1.UNIT-8UNIT-8 Mining Complex Types of DataMining Complex Types of Data LectureLecture TopicTopic **************************************************************************************************** Lecture-50Lecture-50 Multidimensional analysis and descriptiveMultidimensional analysis and descriptive mining of complex data objectsmining of complex data objects Lecture-51Lecture-51 Mining spatial databasesMining spatial databases Lecture-52Lecture-52 Mining multimedia databasesMining multimedia databases Lecture-53Lecture-53 Mining time-series and sequenceMining time-series and sequence datadata Lecture-54Lecture-54 Mining text databasesMining text databases Lecture-55Lecture-55 Mining the World-Wide WebMining the World-Wide Web

2. Lecture-50Lecture-50 Multidimensional analysis andMultidimensional analysis and descriptive mining of complexdescriptive mining of complex data objectsdata objects 3. Mining Complex Data Objects:Mining Complex Data Objects: Generalization of Structured DataGeneralization of Structured Data Set-valued attributeSet-valued attribute Generalization of each value in the set into itsGeneralization of each value in the set into its corresponding higher-level conceptscorresponding higher-level concepts Derivation of the general behavior of the set, suchDerivation of the general behavior of the set, such as the number of elements in the set, the types oras the number of elements in the set, the types or value ranges in the set, or the weighted average forvalue ranges in the set, or the weighted average for numerical datanumerical data hobbyhobby = {= {tennis, hockey, chess, violin,tennis, hockey, chess, violin, nintendo_gamesnintendo_games} generalizes to {} generalizes to {sports, music,sports, music, video_gamesvideo_games}} List-valued or a sequence-valued attributeList-valued or a sequence-valued attribute Same as set-valued attributes except that the orderSame as set-valued attributes except that the order of the elements in the sequence should beof the elements in the sequence should be observed in the generalizationobserved in the generalization Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 4. Generalizing Spatial and Multimedia DataGeneralizing Spatial and Multimedia Data Spatial data:Spatial data: Generalize detailed geographic points into clustered regions,Generalize detailed geographic points into clustered regions, such as business, residential, industrial, or agricultural areas,such as business, residential, industrial, or agricultural areas, according to land usageaccording to land usage Require the merge of a set of geographic areas by spatialRequire the merge of a set of geographic areas by spatial operationsoperations Image data:Image data: Extracted by aggregation and/or approximationExtracted by aggregation and/or approximation Size, color, shape, texture, orientation, and relative positionsSize, color, shape, texture, orientation, and relative positions and structures of the contained objects or regions in the imageand structures of the contained objects or regions in the image Music data:Music data: Summarize its melody: based on the approximate patterns thatSummarize its melody: based on the approximate patterns that repeatedly occur in the segmentrepeatedly occur in the segment Summarized its style: based on its tone, tempo, or the majorSummarized its style: based on its tone, tempo, or the major musical instruments playedmusical instruments played Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 5. Generalizing Object DataGeneralizing Object Data Object identifier: generalize to the lowest level of class in theObject identifier: generalize to the lowest level of class in the class/subclass hierarchiesclass/subclass hierarchies Class composition hierarchiesClass composition hierarchies generalize nested structured datageneralize nested structured data generalize only objects closely related in semantics to the currentgeneralize only objects closely related in semantics to the current oneone Construction and mining of object cubesConstruction and mining of object cubes Extend the attribute-oriented induction methodExtend the attribute-oriented induction method Apply a sequence of class-based generalization operators onApply a sequence of class-based generalization operators on different attributesdifferent attributes Continue until getting a small number of generalized objects thatContinue until getting a small number of generalized objects that can be summarized as a concise in high-level termscan be summarized as a concise in high-level terms For efficient implementationFor efficient implementation Examine each attribute, generalize it to simple-valued dataExamine each attribute, generalize it to simple-valued data Construct a multidimensional data cube (object cube)Construct a multidimensional data cube (object cube) Problem: it is not always desirable to generalize a set of values toProblem: it is not always desirable to generalize a set of values to single-valued datasingle-valued data Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 6. An Example: Plan Mining by Divide andAn Example: Plan Mining by Divide and ConquerConquer Plan: a variable sequence of actionsPlan: a variable sequence of actions E.g., Travel (flight): airline, price, seat> Plan mining: extraction of important or significant generalizedPlan mining: extraction of important or significant generalized (sequential) patterns from a planbase (a large collection of plans)(sequential) patterns from a planbase (a large collection of plans) E.g., Discover travel patterns in an air flight database, orE.g., Discover travel patterns in an air flight database, or find significant patterns from the sequences of actions in thefind significant patterns from the sequences of actions in the repair of automobilesrepair of automobiles MethodMethod Attribute-oriented induction on sequence dataAttribute-oriented induction on sequence data A generalized travel plan: A generalized travel plan: Divide & conquer:Mine characteristics for each subsequenceDivide & conquer:Mine characteristics for each subsequence E.g., big*: same airline, small-big: nearby regionE.g., big*: same airline, small-big: nearby region Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 7. A Travel Database for PlanA Travel Database for Plan MiningMining Example: Mining a travel planbaseExample: Mining a travel planbase plan# action# departure depart_time arrival arrival_time airline 1 1 ALB 800 JFK 900 TWA 1 2 JFK 1000 ORD 1230 UA 1 3 ORD 1300 LAX 1600 UA 1 4 LAX 1710 SAN 1800 DAL 2 1 SPI 900 ORD 950 AA . . . . . . . . . . . . . . . . . . . . . . . . airport_code city state region airport_size 1 1 ALB 800 1 2 JFK 1000 1 3 ORD 1300 1 4 LAX 1710 2 1 SPI 900 . . . . . . . . . . . . . . . Travel plans table Airport info table Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 8. Multidimensional AnalysisMultidimensional Analysis StrategyStrategy Generalize theGeneralize the planbase inplanbase in differentdifferent directionsdirections Look forLook for sequentialsequential patterns in thepatterns in the generalizedgeneralized plansplans Derive high-levelDerive high-level plansplans A multi-D model for the planbase Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 9. MultidimensionalMultidimensional GeneralizationGeneralization Plan# Loc_Seq Size_Seq State_Seq 1 ALB - JFK - ORD - LAX - SAN S - L - L - L - S N - N - I - C - C 2 SPI - ORD - JFK - SYR S - L - L - S I - I - N - N . . . . . . . . . Multi-D generalization of the planbase Plan# Size_Seq State_Seq Region_Seq 1 S - L+ - S N+ - I - C+ E+ - M - P+ 2 S - L+ - S I+ - N+ M+ - E+ . . . . . . . . . Merging consecutive, identical actions in plans %]75[)()( ),(_),(_),,( yregionxregion LysizeairportSxsizeairportyxflight = Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 10. Generalization-Based Sequence MiningGeneralization-Based Sequence Mining Generalize planbase in multidimensional wayGeneralize planbase in multidimensional way using dimension tablesusing dimension tables Use no of distinct values (cardinality) at eachUse no of distinct values (cardinality) at each level to determine the right level oflevel to determine the right level of generalization (level-planning)generalization (level-planning) Use operatorsUse operators mergemerge +,+, optionoption [] to further[] to further generalize patternsgeneralize patterns Retain patterns with significant supportRetain patterns with significant support Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 11. Generalized Sequence PatternsGeneralized Sequence Patterns AirportSize-sequence survives the min thresholdAirportSize-sequence survives the min threshold (after applying(after applying mergemerge operator):operator): SS--LL++ --SS [35%],[35%], LL++ --SS [30%],[30%], SS--LL++ [24.5%],[24.5%], LL++ [9%][9%] After applyingAfter applying optionoption operator:operator: [S][S]--LL++ -[-[S]S] [98.5%][98.5%] Most of the time, people fly via large airports to get toMost of the time, people fly via large airports to get to final destinationfinal destination Other plans: 1.5% of chances, there are otherOther plans: 1.5% of chances, there are other patterns: S-S, L-S-Lpatterns: S-S, L-S-L Lecture-50 - Multidimensional analysis and descriptive mining of complex data objectsLecture-50 - Multidimensional analysis and descriptive mining of complex data objects 12. Lecture-51Lecture-51 Mining spatial databasesMining spatial databases 13. Spatial Data WarehousingSpatial Data Warehousing Spatial data warehouse: Integrated, subject-oriented,Spatial data warehouse: Integrated, subject-oriented, time-variant, and nonvolatile spatial data repository fortime-variant, and nonvolatile spatial data repository for data analysis and decision makingdata analysis and decision making Spatial data integration: a big issueSpatial data integration: a big issue Structure-specific formats (raster- vs. vector-based,Structure-specific formats (raster- vs. vector-based, OO vs. relational models, different storage andOO vs. relational models, different storage and indexing)indexing) Vendor-specific formats (ESRI, MapInfo, Integraph)Vendor-specific formats (ESRI, MapInfo, Integraph) Spatial data cube: multidimensional spatial databaseSpatial data cube: multidimensional spatial database Both dimensions and measures may contain spatialBoth dimensions and measures may contain spatial componentscomponents Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 14. Dimensions and Measures in SpatialDimensions and Measures in Spatial Data WarehouseData Warehouse Dimension modelingDimension modeling nonspatialnonspatial e.g. temperature: 25-30e.g. temperature: 25-30 degrees generalizes todegrees generalizes to hothot spatial-to-nonspatialspatial-to-nonspatial e.g. region B.C.e.g. region B.C. generalizes togeneralizes to description description westernwestern provincesprovinces spatial-to-spatialspatial-to-spatial e.g. region Burnabye.g. region Burnaby generalizes to regiongeneralizes to region Lower MainlandLower Mainland MeasuresMeasures numericalnumerical distributive ( count, sum)distributive ( count, sum) algebraic (e.g. average)algebraic (e.g. average) holistic (e.g. median, rank)holistic (e.g. median, rank) spatialspatial collection of spatialcollection of spatial pointers (e.g. pointers topointers (e.g. pointers to all regions with 25-30all regions with 25-30 degrees in July)degrees in July) Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 15. Example: BC weather patternExample: BC weather pattern analysisanalysis InputInput A map with about 3,000 weather probes scattered in B.C.A map with about 3,000 weather probes scattered in B.C. Daily data for temperature, precipitation, wind velocity, etc.Daily data for temperature, precipitation, wind velocity, etc. Concept hierarchies for all attributesConcept hierarchies for all attributes OutputOutput A map that reveals patterns: merged (similar) regionsA map that reveals patterns: merged (similar) regions GoalsGoals Interactive analysis (drill-down, slice, dice, pivot, roll-up)Interactive analysis (drill-down, slice, dice, pivot, roll-up) Fast response timeFast response time Minimizing storage space usedMinimizing storage space used ChallengeChallenge A merged region may contain hundreds of primitive regionsA merged region may contain hundreds of primitive regions (polygons)(polygons) Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 16. Star Schema of the BCStar Schema of the BC Weather WarehouseWeather Warehouse Spatial dataSpatial data warehousewarehouse DimensionsDimensions region_nameregion_name timetime temperaturetemperature precipitationprecipitation MeasurementsMeasurements region_mapregion_map areaarea countcount Fact tableDimension table Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 17. Spatial MergeSpatial Merge Precomputing all: too much storage space On-line merge: very expensive Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 18. Methods for Computation of SpatialMethods for Computation of Spatial Data CubeData Cube On-line aggregation: collect and store pointers to spatialOn-line aggregation: collect and store pointers to spatial objects in a spatial data cubeobjects in a spatial data cube expensive and slow, need efficient aggregationexpensive and slow, need efficient aggregation techniquestechniques Precompute and store all the possible combinationsPrecompute and store all the possible combinations huge space overheadhuge space overhead Precompute and store rough approximations in a spatialPrecompute and store rough approximations in a spatial data cubedata cube accuracy trade-offaccuracy trade-off Selective computation: only materialize those which will beSelective computation: only materialize those which will be accessed frequentlyaccessed frequently a reasonable choicea reasonable choice Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 19. Spatial Association AnalysisSpatial Association Analysis Spatial association rule:Spatial association rule: AA BB [[ss%,%, cc%]%] A and B are sets of spatial or nonspatial predicatesA and B are sets of spatial or nonspatial predicates Topological relations:Topological relations: intersects, overlaps, disjoint,intersects, overlaps, disjoint, etc.etc. Spatial orientations:Spatial orientations: left_of, west_of, under,left_of, west_of, under, etc.etc. Distance information:Distance information: close_to, within_distance,close_to, within_distance, etc.etc. ss% is the support and% is the support and cc% is the confidence of the rule% is the confidence of the rule ExamplesExamples is_a(x, large_town) ^ intersect(x, highway)is_a(x, large_town) ^ intersect(x, highway) adjacent_to(x,adjacent_to(x, water) [7%, 85%]water) [7%, 85%] is_a(x, large_town) ^adjacent_to(x, georgia_strait)is_a(x, large_town) ^adjacent_to(x, georgia_strait) close_to(x, u.s.a.)close_to(x, u.s.a.) [1%, 78%][1%, 78%] Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 20. Progressive Refinement Mining of SpatialProgressive Refinement Mining of Spatial Association RulesAssociation Rules Hierarchy of spatial relationship:Hierarchy of spatial relationship: g_close_tog_close_to:: near_bynear_by,, touchtouch,, intersectintersect,, containcontain, etc., etc. First search for rough relationship and then refine itFirst search for rough relationship and then refine it Two-step mining of spatial association:Two-step mining of spatial association: Step 1: Rough spatial computation (as a filter)Step 1: Rough spatial computation (as a filter) Using MBR or R-tree for rough estimationUsing MBR or R-tree for rough estimation Step2: Detailed spatial algorithm (as refinement)Step2: Detailed spatial algorithm (as refinement) Apply only to those objects which have passed the roughApply only to those objects which have passed the rough spatial association test (no less thanspatial association test (no less than min_supportmin_support)) Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 21. Spatial classificationSpatial classification Analyze spatial objects to derive classificationAnalyze spatial objects to derive classification schemes, such as decision trees in relevance toschemes, such as decision trees in relevance to certain spatial properties (district, highway, river, etc.)certain spatial properties (district, highway, river, etc.) Example: Classify regions in a province intoExample: Classify regions in a province into richrich vs.vs. poorpoor according to the average family incomeaccording to the average family income Spatial trend analysisSpatial trend analysis Detect changes and trends along a spatial dimensionDetect changes and trends along a spatial dimension Study the trend of nonspatial or spatial data changingStudy the trend of nonspatial or spatial data changing with spacewith space Example: Observe the trend of changes of the climateExample: Observe the trend of changes of the climate or vegetation with the increasing distance from anor vegetation with the increasing distance from an oceanocean Spatial Classification and Spatial TrendSpatial Classification and Spatial Trend AnalysisAnalysis Lecture-51 - Mining spatial databasesLecture-51 - Mining spatial databases 22. Lecture-52Lecture-52 Mining multimedia databasesMining multimedia databases 23. Similarity Search in Multimedia DataSimilarity Search in Multimedia Data Description-based retrieval systemsDescription-based retrieval systems Build indices and perform object retrieval based onBuild indices and perform object retrieval based on image descriptions, such as keywords, captions,image descriptions, such as keywords, captions, size, and time of creationsize, and time of creation Labor-intensive if performed manuallyLabor-intensive if performed manually Results are typically of poor quality if automatedResults are typically of poor quality if automated Content-based retrieval systemsContent-based retrieval systems Support retrieval based on the image content, suchSupport retrieval based on the image content, such as color histogram, texture, shape, objects, andas color histogram, texture, shape, objects, and wavelet transformswavelet transforms Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 24. Queries in Content-Based RetrievalQueries in Content-Based Retrieval SystemsSystems Image sample-based queries:Image sample-based queries: Find all of the images that are similar to the givenFind all of the images that are similar to the given image sampleimage sample Compare the feature vector (signature) extracted fromCompare the feature vector (signature) extracted from the sample with the feature vectors of images thatthe sample with the feature vectors of images that have already been extracted and indexed in the imagehave already been extracted and indexed in the image databasedatabase Image feature specification queries:Image feature specification queries: Specify or sketch image features like color, texture, orSpecify or sketch image features like color, texture, or shape, which are translated into a feature vectorshape, which are translated into a feature vector Match the feature vector with the feature vectors of theMatch the feature vector with the feature vectors of the images in the databaseimages in the database Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 25. Approaches Based on ImageApproaches Based on Image SignatureSignatureColor histogram-based signatureColor histogram-based signature The signature includes color histograms based on colorThe signature includes color histograms based on color composition of an image regardless of its scale orcomposition of an image regardless of its scale or orientationorientation No information about shape, location, or textureNo information about shape, location, or texture Two images with similar color composition may containTwo images with similar color composition may contain very different shapes or textures, and thus could bevery different shapes or textures, and thus could be completely unrelated in semanticscompletely unrelated in semantics Multifeature composed signatureMultifeature composed signature The signature includes a composition of multipleThe signature includes a composition of multiple features: color histogram, shape, location, and texturefeatures: color histogram, shape, location, and texture Can be used to search for similar imagesCan be used to search for similar images Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 26. Wavelet AnalysisWavelet Analysis Wavelet-based signatureWavelet-based signature Use the dominant wavelet coefficients of an image as itsUse the dominant wavelet coefficients of an image as its signaturesignature Wavelets capture shape, texture, and locationWavelets capture shape, texture, and location information in a single unified frameworkinformation in a single unified framework Improved efficiency and reduced the need for providingImproved efficiency and reduced the need for providing multiple search primitivesmultiple search primitives May fail to identify images containing similar in locationMay fail to identify images containing similar in location or size objectsor size objects Wavelet-based signature with region-based granularityWavelet-based signature with region-based granularity Similar images may contain similar regions, but a regionSimilar images may contain similar regions, but a region in one image could be a translation or scaling of ain one image could be a translation or scaling of a matching region in the othermatching region in the other Compute and compare signatures at the granularity ofCompute and compare signatures at the granularity of regions, not the entire imageregions, not the entire image Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 27. C-BIRD:C-BIRD: CContent-ontent-BBasedased IImagemage RRetrieval frometrieval from DDigital librariesigital libraries Search by image colors by color percentage by color layout by texture density by texture Layout by object model by illumination invariance by keywords Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 28. Multi-Dimensional Search inMulti-Dimensional Search in Multimedia DatabasesMultimedia DatabasesColor layout Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 29. Color histogram Texture layout Multi-Dimensional AnalysisMulti-Dimensional Analysis in Multimedia Databasesin Multimedia Databases Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 30. Refining or combining searches Search for blue sky (top layout grid is blue) Search for blue sky and green meadows (top layout grid is blue and bottom is green) Search for airplane in blue sky (top layout grid is blue and keyword = airplane) Mining Multimedia DatabasesMining Multimedia Databases Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 31. Multidimensional Analysis of MultimediaMultidimensional Analysis of Multimedia DataData Multimedia data cubeMultimedia data cube Design and construction similar to that of traditionalDesign and construction similar to that of traditional data cubes from relational datadata cubes from relational data Contain additional dimensions and measures forContain additional dimensions and measures for multimedia information, such as color, texture, andmultimedia information, such as color, texture, and shapeshape The database does not store images but theirThe database does not store images but their descriptorsdescriptors Feature descriptor: a set of vectors for each visualFeature descriptor: a set of vectors for each visual characteristiccharacteristic Color vector: contains the color histogramColor vector: contains the color histogram MFC (Most Frequent Color) vector: five color centroidsMFC (Most Frequent Color) vector: five color centroids MFO (Most Frequent Orientation) vector: five edge orientationMFO (Most Frequent Orientation) vector: five edge orientation centroidscentroids Layout descriptor: contains a color layout vector and anLayout descriptor: contains a color layout vector and an edge layout vectoredge layout vectorLecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 32. Mining Multimedia Databases inMining Multimedia Databases in Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 33. RED WHITE BLUE GIFJPEG By Format By Colour Sum Cross Tab RED WHITE BLUE Colour Sum Group By Measurement JPEG GIF Small Very Large RED WHITE BLUE By Colour By Format & Colour By Format & Size By Colour & Size By Format By Size Sum The Data Cube and the Sub-Space Measurements Medium Large Format of image Duration Colors Textures Keywords Size Width Height Internet domain of image Internet domain of parent pages Image popularity Mining Multimedia DatabasesMining Multimedia Databases Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 34. Classification in MultiMediaMinerClassification in MultiMediaMiner Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 35. Special features:Special features: Need # of occurrences besides Boolean existence, e.g.,Need # of occurrences besides Boolean existence, e.g., Two red square and one blue circle implies themeTwo red square and one blue circle implies theme air-showair-show Need spatial relationshipsNeed spatial relationships Blue on top of white squared object is associatedBlue on top of white squared object is associated with brown bottomwith brown bottom Need multi-resolution and progressive refinementNeed multi-resolution and progressive refinement miningmining It is expensive to explore detailed associationsIt is expensive to explore detailed associations among objects at high resolutionamong objects at high resolution It is crucial to ensure the completeness of search atIt is crucial to ensure the completeness of search at multi-resolution spacemulti-resolution space Mining Associations in Multimedia Data Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 36. Spatial Relationships from Layout property P1 next-to property P2property P1 on-top-of property P2 Different Resolution Hierarchy Mining Multimedia DatabasesMining Multimedia Databases Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 37. From Coarse to Fine Resolution Mining Mining Multimedia DatabasesMining Multimedia Databases Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 38. Challenge: Curse of DimensionalityChallenge: Curse of Dimensionality Difficult to implement a data cube efficiently given aDifficult to implement a data cube efficiently given a large number of dimensions, especially serious in thelarge number of dimensions, especially serious in the case of multimedia data cubescase of multimedia data cubes Many of these attributes are set-oriented instead ofMany of these attributes are set-oriented instead of single-valuedsingle-valued Restricting number of dimensions may lead to theRestricting number of dimensions may lead to the modeling of an image at a rather rough, limited, andmodeling of an image at a rather rough, limited, and imprecise scaleimprecise scale More research is needed to strike a balance betweenMore research is needed to strike a balance between efficiency and power of representationefficiency and power of representation Lecture-52 - Mining multimedia databasesLecture-52 - Mining multimedia databases 39. Lecture-53Lecture-53 Mining time-series and sequenceMining time-series and sequence datadata 40. Mining Time-Series and SequenceMining Time-Series and Sequence DataData Time-series databaseTime-series database Consists of sequences of values or eventsConsists of sequences of values or events changing with timechanging with time Data is recorded at regular intervalsData is recorded at regular intervals Characteristic time-series componentsCharacteristic time-series components Trend, cycle, seasonal, irregularTrend, cycle, seasonal, irregular ApplicationsApplications Financial: stock price, inflationFinancial: stock price, inflation Biomedical: blood pressureBiomedical: blood pressure Meteorological: precipitationMeteorological: precipitation Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 41. Mining Time-Series andMining Time-Series and Sequence DataSequence Data Time-series plot Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 42. Mining Time-Series and Sequence Data:Mining Time-Series and Sequence Data: Trend analysisTrend analysis A time series can be illustrated as a time-series graphA time series can be illustrated as a time-series graph which describes a point moving with the passage of timewhich describes a point moving with the passage of time Categories of Time-Series MovementsCategories of Time-Series Movements Long-term or trend movements (trend curve)Long-term or trend movements (trend curve) Cyclic movements or cycle variations, e.g., businessCyclic movements or cycle variations, e.g., business cyclescycles Seasonal movements or seasonal variationsSeasonal movements or seasonal variations i.e, almost identical patterns that a time seriesi.e, almost identical patterns that a time series appears to follow during corresponding months ofappears to follow during corresponding months of successive years.successive years. Irregular or random movementsIrregular or random movements Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 43. Estimation of Trend CurveEstimation of Trend Curve The freehand methodThe freehand method Fit the curve by looking at the graphFit the curve by looking at the graph Costly and barely reliable for large-scaled data miningCostly and barely reliable for large-scaled data mining The least-square methodThe least-square method Find the curve minimizing the sum of the squares ofFind the curve minimizing the sum of the squares of the deviation of points on the curve from thethe deviation of points on the curve from the corresponding data pointscorresponding data points The moving-average methodThe moving-average method Eliminate cyclic, seasonal and irregular patternsEliminate cyclic, seasonal and irregular patterns Loss of end dataLoss of end data Sensitive to outliersSensitive to outliers Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 44. Discovery of Trend in Time-SeriesDiscovery of Trend in Time-Series Estimation of seasonal variationsEstimation of seasonal variations Seasonal indexSeasonal index Set of numbers showing the relative values of a variable duringSet of numbers showing the relative values of a variable during the months of the yearthe months of the year E.g., if the sales during October, November, and December areE.g., if the sales during October, November, and December are 80%, 120%, and 140% of the average monthly sales for the80%, 120%, and 140% of the average monthly sales for the whole year, respectively, then 80, 120, and 140 are seasonalwhole year, respectively, then 80, 120, and 140 are seasonal index numbers for these monthsindex numbers for these months Deseasonalized dataDeseasonalized data Data adjusted for seasonal variationsData adjusted for seasonal variations E.g., divide the original monthly data by the seasonal indexE.g., divide the original monthly data by the seasonal index numbers for the corresponding monthsnumbers for the corresponding months Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 45. Discovery of Trend in Time-SeriesDiscovery of Trend in Time-Series Estimation of cyclic variationsEstimation of cyclic variations If (approximate) periodicity of cycles occurs, cyclicIf (approximate) periodicity of cycles occurs, cyclic index can be constructed in much the same mannerindex can be constructed in much the same manner as seasonal indexesas seasonal indexes Estimation of irregular variationsEstimation of irregular variations By adjusting the data for trend, seasonal and cyclicBy adjusting the data for trend, seasonal and cyclic variationsvariations With the systematic analysis of the trend, cyclic,With the systematic analysis of the trend, cyclic, seasonal, and irregular components, it isseasonal, and irregular components, it is possible to make long- or short-term predictionspossible to make long- or short-term predictions with reasonable qualitywith reasonable quality Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 46. Similarity Search in Time-Series AnalysisSimilarity Search in Time-Series Analysis Normal database query finds exact matchNormal database query finds exact match Similarity search finds data sequences that differSimilarity search finds data sequences that differ only slightly from the given query sequenceonly slightly from the given query sequence Two categories of similarity queriesTwo categories of similarity queries Whole matching: find a sequence that is similar to theWhole matching: find a sequence that is similar to the query sequencequery sequence Subsequence matching: find all pairs of similarSubsequence matching: find all pairs of similar sequencessequences Typical ApplicationsTypical Applications Financial marketFinancial market Market basket data analysisMarket basket data analysis Scientific databasesScientific databases Medical diagnosisMedical diagnosisLecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 47. Multidimensional IndexingMultidimensional Indexing Multidimensional indexMultidimensional index Constructed for efficient accessing using theConstructed for efficient accessing using the first few Fourier coefficientsfirst few Fourier coefficients Use the index can to retrieve theUse the index can to retrieve the sequences that are at most a certainsequences that are at most a certain small distance away from the querysmall distance away from the query sequencesequence Perform postprocessing by computing thePerform postprocessing by computing the actual distance between sequences inactual distance between sequences in the time domain and discard any falsethe time domain and discard any false matchesmatches Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 48. Subsequence MatchingSubsequence Matching Break each sequence into a set of pieces ofBreak each sequence into a set of pieces of window with lengthwindow with length ww Extract the features of the subsequenceExtract the features of the subsequence inside the windowinside the window Map each sequence to a trail in theMap each sequence to a trail in the feature spacefeature space Divide the trail of each sequence intoDivide the trail of each sequence into subtrails and represent each of them withsubtrails and represent each of them with minimum bounding rectangleminimum bounding rectangle Use a multipiece assembly algorithm toUse a multipiece assembly algorithm to search for longer sequence matchessearch for longer sequence matches Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 49. Enhanced similarity search methodsEnhanced similarity search methods Allow for gaps within a sequence or differences in offsetsAllow for gaps within a sequence or differences in offsets or amplitudesor amplitudes Normalize sequences with amplitude scaling and offsetNormalize sequences with amplitude scaling and offset translationtranslation Two subsequences are considered similar if one liesTwo subsequences are considered similar if one lies within an envelope ofwithin an envelope of width around the other, ignoringwidth around the other, ignoring outliersoutliers Two sequences are said to be similar if they have enoughTwo sequences are said to be similar if they have enough non-overlapping time-ordered pairs of similarnon-overlapping time-ordered pairs of similar subsequencessubsequences Parameters specified by a user or expert: sliding windowParameters specified by a user or expert: sliding window size, width of an envelope for similarity, maximum gap,size, width of an envelope for similarity, maximum gap, and matching fractionand matching fraction Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 50. Steps for performing a similarity searchSteps for performing a similarity search Atomic matchingAtomic matching Find all pairs of gap-free windows of a small lengthFind all pairs of gap-free windows of a small length that are similarthat are similar Window stitchingWindow stitching Stitch similar windows to form pairs of large similarStitch similar windows to form pairs of large similar subsequences allowing gaps between atomicsubsequences allowing gaps between atomic matchesmatches Subsequence OrderingSubsequence Ordering Linearly order the subsequence matches toLinearly order the subsequence matches to determine whether enough similar pieces existdetermine whether enough similar pieces exist Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 51. Query Languages for Time SequencesQuery Languages for Time Sequences Time-sequence query languageTime-sequence query language Should be able to specify sophisticated queries likeShould be able to specify sophisticated queries like Find all of the sequences that are similar to some sequence in classFind all of the sequences that are similar to some sequence in class AA, but not similar to any sequence in class, but not similar to any sequence in class BB Should be able to support various kinds of queries: rangeShould be able to support various kinds of queries: range queries, all-pair queries, and nearest neighbor queriesqueries, all-pair queries, and nearest neighbor queries Shape definition languageShape definition language Allows users to define and query the overall shape of timeAllows users to define and query the overall shape of time sequencessequences Uses human readable series of sequence transitions or macrosUses human readable series of sequence transitions or macros Ignores the specific detailsIgnores the specific details E.g., the pattern up, Up, UP can be used to describeE.g., the pattern up, Up, UP can be used to describe increasing degrees of rising slopesincreasing degrees of rising slopes Macros: spike, valley, etc.Macros: spike, valley, etc. Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 52. Sequential Pattern MiningSequential Pattern Mining Mining of frequently occurring patterns related toMining of frequently occurring patterns related to time or other sequencestime or other sequences Sequential pattern mining usually concentrate onSequential pattern mining usually concentrate on symbolic patternssymbolic patterns ExamplesExamples Renting Star Wars, then Empire Strikes Back, thenRenting Star Wars, then Empire Strikes Back, then Return of the Jedi in that orderReturn of the Jedi in that order Collection of ordered events within an intervalCollection of ordered events within an interval ApplicationsApplications Targeted marketingTargeted marketing Customer retentionCustomer retention Weather predictionWeather prediction Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 53. Mining SequencesMining Sequences (cont.)(cont.) CustId Video sequence 1 {(C), (H)} 2 {(AB), (C), (DFG)} 3 {(CEG)} 4 {(C), (DG), (H)} 5 {(H)} Customer-sequenceCustomer-sequence Sequential patterns with supportSequential patterns with support > 0.25> 0.25 {(C), (H)}{(C), (H)} {(C), (DG)}{(C), (DG)} Map Large ItemsetsMap Large Itemsets Large Itemsets MappedID (C) 1 (D) 2 (G) 3 (DG) 4 (H) 5 Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 54. Sequential pattern mining: Cases andSequential pattern mining: Cases and ParametersParameters Duration of a time sequenceDuration of a time sequence TT Sequential pattern mining can then be confined to theSequential pattern mining can then be confined to the data within a specified durationdata within a specified duration Ex. Subsequence corresponding to the year of 1999Ex. Subsequence corresponding to the year of 1999 Ex. Partitioned sequences, such as every year, orEx. Partitioned sequences, such as every year, or every week after stock crashes, or every two weeksevery week after stock crashes, or every two weeks before and after a volcano eruptionbefore and after a volcano eruption Event folding windowEvent folding window ww IfIf w = Tw = T, time-insensitive frequent patterns are found, time-insensitive frequent patterns are found IfIf w = 0w = 0 (no event sequence folding), sequential(no event sequence folding), sequential patterns are found where each event occurs at apatterns are found where each event occurs at a distinct time instantdistinct time instant IfIf 0 < w < T0 < w < T, sequences occurring within the same, sequences occurring within the same periodperiod ww are folded in the analysisare folded in the analysisLecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 55. Sequential pattern mining: Cases andSequential pattern mining: Cases and ParametersParameters Time interval,Time interval, intint, between events in the, between events in the discovered patterndiscovered pattern intint = 0: no interval gap is allowed, i.e., only strictly= 0: no interval gap is allowed, i.e., only strictly consecutive sequences are foundconsecutive sequences are found Ex. Find frequent patterns occurring in consecutive weeksEx. Find frequent patterns occurring in consecutive weeks min_intmin_int intint max_intmax_int: find patterns that are: find patterns that are separated by at leastseparated by at least min_intmin_int but at mostbut at most max_intmax_int Ex. If a person rents movie A, it is likely she will rent movie BEx. If a person rents movie A, it is likely she will rent movie B within 30 days (within 30 days (intint 30)30) intint = c= c 0: find patterns carrying an exact interval0: find patterns carrying an exact interval Ex. Every time when Dow Jones drops more than 5%, whatEx. Every time when Dow Jones drops more than 5%, what will happen exactly two days later? (will happen exactly two days later? (intint = 2)= 2) Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 56. Episodes and Sequential Pattern MiningEpisodes and Sequential Pattern Mining MethodsMethods Other methods for specifying the kinds ofOther methods for specifying the kinds of patternspatterns Serial episodes: ASerial episodes: A BB Parallel episodes: A & BParallel episodes: A & B Regular expressions: (A | B)C*(DRegular expressions: (A | B)C*(D E)E) Methods for sequential pattern miningMethods for sequential pattern mining Variations of Apriori-like algorithms, e.g., GSPVariations of Apriori-like algorithms, e.g., GSP Database projection-based pattern growthDatabase projection-based pattern growth Similar to the frequent pattern growth withoutSimilar to the frequent pattern growth without candidate generationcandidate generation Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 57. Periodicity AnalysisPeriodicity Analysis Periodicity is everywhere: tides, seasons, daily powerPeriodicity is everywhere: tides, seasons, daily power consumption, etc.consumption, etc. Full periodicityFull periodicity Every point in time contributes (precisely orEvery point in time contributes (precisely or approximately) to the periodicityapproximately) to the periodicity Partial periodicit: A more general notionPartial periodicit: A more general notion Only some segments contribute to the periodicityOnly some segments contribute to the periodicity Jim reads NY Times 7:00-7:30 am every week dayJim reads NY Times 7:00-7:30 am every week day Cyclic association rulesCyclic association rules Associations which form cyclesAssociations which form cycles MethodsMethods Full periodicity: FFT, other statistical analysis methodsFull periodicity: FFT, other statistical analysis methods Partial and cyclic periodicity: Variations of Apriori-likePartial and cyclic periodicity: Variations of Apriori-like mining methodsmining methods Lecture-53 - Mining time-series and sequence dataLecture-53 - Mining time-series and sequence data 58. Lecture-54Lecture-54 Mining text databasesMining text databases 59. Text Databases and IRText Databases and IR Text databases (document databases)Text databases (document databases) Large collections of documents from various sources:Large collections of documents from various sources: news articles, research papers, books, digital libraries,news articles, research papers, books, digital libraries, e-mail messages, and Web pages, library database,e-mail messages, and Web pages, library database, etc.etc. Data stored is usuallyData stored is usually semi-structuredsemi-structured Traditional information retrieval techniques becomeTraditional information retrieval techniques become inadequate for the increasingly vast amounts of textinadequate for the increasingly vast amounts of text datadata Information retrievalInformation retrieval A field developed in parallel with database systemsA field developed in parallel with database systems Information is organized into (a large number of)Information is organized into (a large number of) documentsdocuments Information retrieval problem: locating relevantInformation retrieval problem: locating relevant documents based on user input, such as keywords ordocuments based on user input, such as keywords or example documentsexample documents Lecture-54 - Mining text databasesLecture-54 - Mining text databases 60. Information RetrievalInformation Retrieval Typical IR systemsTypical IR systems Online library catalogsOnline library catalogs Online document management systemsOnline document management systems Information retrieval vs. database systemsInformation retrieval vs. database systems Some DB problems are not present in IR, e.g., update,Some DB problems are not present in IR, e.g., update, transaction management, complex objectstransaction management, complex objects Some IR problems are not addressed well in DBMS,Some IR problems are not addressed well in DBMS, e.g., unstructured documents, approximate searche.g., unstructured documents, approximate search using keywords and relevanceusing keywords and relevance Lecture-54 - Mining text databasesLecture-54 - Mining text databases 61. Basic Measures for TextBasic Measures for Text RetrievalRetrieval Precision:Precision: the percentage of retrieved documentsthe percentage of retrieved documents that are in fact relevant to the query (i.e., correctthat are in fact relevant to the query (i.e., correct responses)responses) Recall:Recall: the percentage of documents that arethe percentage of documents that are relevant to the query and were, in fact, retrievedrelevant to the query and were, in fact, retrieved|}{| |}{}{| Relevant RetrievedRelevant precision = |}{| |}{}{| Retrieved RetrievedRelevant precision = Lecture-54 - Mining text databasesLecture-54 - Mining text databases 62. Keyword-Based RetrievalKeyword-Based Retrieval A document is represented by a string, whichA document is represented by a string, which can be identified by a set of keywordscan be identified by a set of keywords Queries may use expressions of keywordsQueries may use expressions of keywords E.g., carE.g., car andand repair shop, tearepair shop, tea oror coffee, DBMScoffee, DBMS butbut notnot OracleOracle Queries and retrieval should consider synonyms,Queries and retrieval should consider synonyms, e.g., repair and maintenancee.g., repair and maintenance Major difficulties of the modelMajor difficulties of the model Synonymy: A keywordSynonymy: A keyword TT does not appear anywheredoes not appear anywhere in the document, even though the document isin the document, even though the document is closely related toclosely related to TT, e.g., data mining, e.g., data mining Polysemy: The same keyword may mean differentPolysemy: The same keyword may mean different things in different contexts, e.g., miningthings in different contexts, e.g., mining Lecture-54 - Mining text databasesLecture-54 - Mining text databases 63. Similarity-Based Retrieval in TextSimilarity-Based Retrieval in Text DatabasesDatabases Finds similar documents based on a set of commonFinds similar documents based on a set of common keywordskeywords Answer should be based on the degree of relevanceAnswer should be based on the degree of relevance based on the nearness of the keywords, relativebased on the nearness of the keywords, relative frequency of the keywords, etc.frequency of the keywords, etc. Basic techniquesBasic techniques Stop listStop list Set of words that are deemed irrelevant, evenSet of words that are deemed irrelevant, even though they may appear frequentlythough they may appear frequently E.g.,E.g., a, the, of, for, witha, the, of, for, with, etc., etc. Stop lists may vary when document set variesStop lists may vary when document set varies Lecture-54 - Mining text databasesLecture-54 - Mining text databases 64. Similarity-Based Retrieval in TextSimilarity-Based Retrieval in Text DatabasesDatabases Word stemWord stem Several words are small syntactic variants of eachSeveral words are small syntactic variants of each other since they share a common word stemother since they share a common word stem E.g.,E.g., drugdrug,, drugs, druggeddrugs, drugged A term frequency tableA term frequency table Each entryEach entry frequent_table(i, j)frequent_table(i, j) = # of occurrences of= # of occurrences of the wordthe word ttii in documentin document ddii Usually, theUsually, the ratioratio instead of the absolute number ofinstead of the absolute number of occurrences is usedoccurrences is used Similarity metrics: measure the closeness of aSimilarity metrics: measure the closeness of a document to a query (a set of keywords)document to a query (a set of keywords) Relative term occurrencesRelative term occurrences Cosine distance:Cosine distance: |||| ),( 21 21 21 vv vv vvsim = Lecture-54 - Mining text databasesLecture-54 - Mining text databases 65. Latent Semantic IndexingLatent Semantic Indexing Basic ideaBasic idea Similar documents have similar word frequenciesSimilar documents have similar word frequencies Difficulty: the size of the term frequency matrix is very largeDifficulty: the size of the term frequency matrix is very large Use a singular value decomposition (SVD) techniques to reduceUse a singular value decomposition (SVD) techniques to reduce the size of frequency tablethe size of frequency table Retain theRetain the KK most significant rows of the frequency tablemost significant rows of the frequency table MethodMethod Create a term frequency matrix,Create a term frequency matrix, freq_matrixfreq_matrix SVD construction: Compute the singular valued decomposition ofSVD construction: Compute the singular valued decomposition of freq_matrixfreq_matrix by splitting it into 3 matrices,by splitting it into 3 matrices, UU,, SS,, VV Vector identification: For each documentVector identification: For each document dd, replace its original, replace its original document vector by a new excluding the eliminated termsdocument vector by a new excluding the eliminated terms Index creation: Store the set of all vectors, indexed by one of aIndex creation: Store the set of all vectors, indexed by one of a number of techniques (such as TV-tree)number of techniques (such as TV-tree) Lecture-54 - Mining text databasesLecture-54 - Mining text databases 66. Other Text Retrieval Indexing TechniquesOther Text Retrieval Indexing Techniques Inverted indexInverted index Maintains two hash- or B+-tree indexed tables:Maintains two hash- or B+-tree indexed tables: document_table: a set of document records postings_list> term_table: a set of term records, term_table: a set of term records, Answer query: Find all docs associated with one or a set of termsAnswer query: Find all docs associated with one or a set of terms Advantage: easy to implementAdvantage: easy to implement Disadvantage: do not handle well synonymy and polysemy, andDisadvantage: do not handle well synonymy and polysemy, and posting lists could be too long (storage could be very large)posting lists could be too long (storage could be very large) Signature fileSignature file Associate a signature with each documentAssociate a signature with each document A signature is a representation of an ordered list of terms thatA signature is a representation of an ordered list of terms that describe the documentdescribe the document Order is obtained by frequency analysis, stemming and stop listsOrder is obtained by frequency analysis, stemming and stop lists Lecture-54 - Mining text databasesLecture-54 - Mining text databases 67. Types of Text Data MiningTypes of Text Data Mining Keyword-based association analysisKeyword-based association analysis Automatic document classificationAutomatic document classification Similarity detectionSimilarity detection Cluster documents by a common authorCluster documents by a common author Cluster documents containing information from aCluster documents containing information from a common sourcecommon source Link analysis: unusual correlation between entitiesLink analysis: unusual correlation between entities Sequence analysis: predicting a recurring eventSequence analysis: predicting a recurring event Anomaly detection: find information that violates usualAnomaly detection: find information that violates usual patternspatterns Hypertext analysisHypertext analysis Patterns in anchors/linksPatterns in anchors/links Anchor text correlations with linked objectsAnchor text correlations with linked objects Lecture-54 - Mining text databasesLecture-54 - Mining text databases 68. Keyword-based association analysisKeyword-based association analysis Collect sets of keywords or terms that occurCollect sets of keywords or terms that occur frequently together and then find the associationfrequently together and then find the association or correlation relationships among themor correlation relationships among them First preprocess the text data by parsing,First preprocess the text data by parsing, stemming, removing stop words, etc.stemming, removing stop words, etc. Then evoke association mining algorithmsThen evoke association mining algorithms Consider each document as a transactionConsider each document as a transaction View a set of keywords in the document as a set ofView a set of keywords in the document as a set of items in the transactionitems in the transaction Term level association miningTerm level association mining No need for human effort in tagging documentsNo need for human effort in tagging documents The number of meaningless results and the executionThe number of meaningless results and the execution time is greatly reducedtime is greatly reduced Lecture-54 - Mining text databasesLecture-54 - Mining text databases 69. Automatic document classificationAutomatic document classification MotivationMotivation Automatic classification for the tremendous number ofAutomatic classification for the tremendous number of on-line text documents (Web pages, e-mails, etc.)on-line text documents (Web pages, e-mails, etc.) A classification problemA classification problem Training set: Human experts generate a training data setTraining set: Human experts generate a training data set Classification: The computer system discovers theClassification: The computer system discovers the classification rulesclassification rules Application: The discovered rules can be applied toApplication: The discovered rules can be applied to classify new/unknown documentsclassify new/unknown documents Text document classification differs from theText document classification differs from the classification of relational dataclassification of relational data Document databases are not structured according toDocument databases are not structured according to attribute-value pairsattribute-value pairs Lecture-54 - Mining text databasesLecture-54 - Mining text databases 70. Association-Based DocumentAssociation-Based Document ClassificationClassification Extract keywords and terms by information retrieval and simpleExtract keywords and terms by information retrieval and simple association analysis techniquesassociation analysis techniques Obtain concept hierarchies of keywords and terms usingObtain concept hierarchies of keywords and terms using Available term classes, such as WordNetAvailable term classes, such as WordNet Expert knowledgeExpert knowledge Some keyword classification systemsSome keyword classification systems Classify documents in the training set into class hierarchiesClassify documents in the training set into class hierarchies Apply term association mining method to discover sets of associatedApply term association mining method to discover sets of associated termsterms Use the terms to maximally distinguish one class of documents fromUse the terms to maximally distinguish one class of documents from othersothers Derive a set of association rules associated with each documentDerive a set of association rules associated with each document classclass Order the classification rules based on their occurrence frequencyOrder the classification rules based on their occurrence frequency and discriminative powerand discriminative power Used the rules to classify new documentsUsed the rules to classify new documents Lecture-54 - Mining text databasesLecture-54 - Mining text databases 71. Document ClusteringDocument Clustering Automatically group related documents based on theirAutomatically group related documents based on their contentscontents Require no training sets or predetermined taxonomies,Require no training sets or predetermined taxonomies, generate a taxonomy at runtimegenerate a taxonomy at runtime Major stepsMajor steps PreprocessingPreprocessing Remove stop words, stem, feature extraction, lexicalRemove stop words, stem, feature extraction, lexical analysis, analysis, Hierarchical clusteringHierarchical clustering Compute similarities applying clustering algorithms,Compute similarities applying clustering algorithms, SlicingSlicing Fan out controls, flatten the tree to configurableFan out controls, flatten the tree to configurable number of levels, number of levels, Lecture-54 - Mining text databasesLecture-54 - Mining text databases 72. Lecture-55Lecture-55 Mining the World-Wide WebMining the World-Wide Web 73. Mining the World-Wide WebMining the World-Wide Web The WWW is huge, widely distributed, globalThe WWW is huge, widely distributed, global information service center forinformation service center for Information services: news, advertisements,Information services: news, advertisements, consumer information, financial management,consumer information, financial management, education, government, e-commerce, etc.education, government, e-commerce, etc. Hyper-link informationHyper-link information Access and usage informationAccess and usage information WWW provides rich sources for data miningWWW provides rich sources for data mining ChallengesChallenges Too huge for effective data warehousing and dataToo huge for effective data warehousing and data miningmining Too complex and heterogeneous: no standards andToo complex and heterogeneous: no standards and structurestructure Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 74. Mining the World-Wide WebMining the World-Wide Web Growing and changing very rapidlyGrowing and changing very rapidly Broad diversity of user communitiesBroad diversity of user communities Only a small portion of the information on the Web is trulyOnly a small portion of the information on the Web is truly relevant or usefulrelevant or useful 99% of the Web information is useless to 99% of Web users99% of the Web information is useless to 99% of Web users How can we find high-quality Web pages on a specified topic?How can we find high-quality Web pages on a specified topic? Internet growth 0 5000000 10000000 15000000 20000000 25000000 30000000 35000000 40000000 Sep-69 Sep-72 Sep-75 Sep-78 Sep-81 Sep-84 Sep-87 Sep-90 Sep-93 Sep-96 Sep-99 Hosts Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 75. Web search enginesWeb search engines Index-based: search the Web, index Web pages,Index-based: search the Web, index Web pages, and build and store huge keyword-based indicesand build and store huge keyword-based indices Help locate sets of Web pages containing certainHelp locate sets of Web pages containing certain keywordskeywords DeficienciesDeficiencies A topic of any breadth may easily containA topic of any breadth may easily contain hundreds of thousands of documentshundreds of thousands of documents Many documents that are highly relevant to aMany documents that are highly relevant to a topic may not contain keywords defining themtopic may not contain keywords defining them (polysemy)(polysemy) Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 76. Web Mining: A more challenging taskWeb Mining: A more challenging task Searches forSearches for Web access patternsWeb access patterns Web structuresWeb structures Regularity and dynamics of Web contentsRegularity and dynamics of Web contents ProblemsProblems The abundance problemThe abundance problem Limited coverage of the Web: hidden Web sources,Limited coverage of the Web: hidden Web sources, majority of data in DBMSmajority of data in DBMS Limited query interface based on keyword-orientedLimited query interface based on keyword-oriented searchsearch Limited customization to individual usersLimited customization to individual users Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 77. Web Mining Web Structure Mining Web Content Mining Web Page Content Mining Search Result Mining Web Usage Mining General Access Pattern Tracking Customized Usage Tracking Web Mining TaxonomyWeb Mining Taxonomy Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 78. Web Mining Web Structure Mining Web Content Mining Web Page Content Mining Web Page Summarization WebLog (Lakshmanan et.al. 1996), WebOQL(Mendelzon et.al. 1998) : Web Structuring query languages; Can identify information within given web pages Ahoy! (Etzioni et.al. 1997):Uses heuristics to distinguish personal home pages from other web pages ShopBot (Etzioni et.al. 1997): Looks for product prices within web pages Search Result Mining Web Usage Mining General Access Pattern Tracking Customized Usage Tracking Mining the World-Wide WebMining the World-Wide Web Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 79. Web Mining Mining the World-Wide WebMining the World-Wide Web Web Usage Mining General Access Pattern Tracking Customized Usage Tracking Web Structure Mining Web Content Mining Web Page Content Mining Search Result Mining Search Engine Result Summarization Clustering Search Result (Leouski and Croft, 1996, Zamir and Etzioni, 1997): Categorizes documents using phrases in titles and snippets Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 80. Web Mining Web Content Mining Web Page Content Mining Search Result Mining Web Usage Mining General Access Pattern Tracking Customized Usage Tracking Mining the World-Wide WebMining the World-Wide Web Web Structure Mining Using Links PageRank (Brin et al., 1998) CLEVER (Chakrabarti et al., 1998) Use interconnections between web pages to give weight to pages. Using Generalization MLDB (1994), VWV (1998) Uses a multi-level database representation of the Web. Counters (popularity) and link lists are used for capturing structure. Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 81. Web Mining Web Structure Mining Web Content Mining Web Page Content Mining Search Result Mining Web Usage Mining General Access Pattern Tracking Web Log Mining (Zaane, Xin and Han, 1998) Uses KDD techniques to understand general access patterns and trends. Can shed light on better structure and grouping of resource providers. Customized Usage Tracking Mining the World-Wide WebMining the World-Wide Web Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 82. Web Mining Web Usage Mining General Access Pattern Tracking Customized Usage Tracking Adaptive Sites (Perkowitz and Etzioni, 1997) Analyzes access patterns of each user at a time. Web site restructures itself automatically by learning from user access patterns. Mining the World-Wide WebMining the World-Wide Web Web Structure Mining Web Content Mining Web Page Content Mining Search Result Mining Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 83. Mining the Web's Link StructuresMining the Web's Link Structures Finding authoritative Web pagesFinding authoritative Web pages Retrieving pages that are not only relevant, but also ofRetrieving pages that are not only relevant, but also of high quality, or authoritative on the topichigh quality, or authoritative on the topic Hyperlinks can infer the notion of authorityHyperlinks can infer the notion of authority The Web consists not only of pages, but also ofThe Web consists not only of pages, but also of hyperlinks pointing from one page to anotherhyperlinks pointing from one page to another These hyperlinks contain an enormous amount ofThese hyperlinks contain an enormous amount of latent human annotationlatent human annotation A hyperlink pointing to another Web page, this can beA hyperlink pointing to another Web page, this can be considered as the author's endorsement of the otherconsidered as the author's endorsement of the other pagepage Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 84. Mining the Web's Link StructuresMining the Web's Link Structures Problems with the Web linkage structureProblems with the Web linkage structure Not every hyperlink represents an endorsementNot every hyperlink represents an endorsement Other purposes are for navigation or for paidOther purposes are for navigation or for paid advertisementsadvertisements If the majority of hyperlinks are for endorsement,If the majority of hyperlinks are for endorsement, the collective opinion will still dominatethe collective opinion will still dominate One authority will seldom have its Web page point toOne authority will seldom have its Web page point to its rival authorities in the same fieldits rival authorities in the same field Authoritative pages are seldom particularlyAuthoritative pages are seldom particularly descriptivedescriptive HubHub Set of Web pages that provides collections of links toSet of Web pages that provides collections of links to authoritiesauthorities Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 85. HITS (HITS (HHyperlink-yperlink-IInducednduced TTopicopic SSearch)earch) Explore interactions between hubs and authoritativeExplore interactions between hubs and authoritative pagespages Use an index-based search engine to form the root setUse an index-based search engine to form the root set Many of these pages are presumably relevant to theMany of these pages are presumably relevant to the search topicsearch topic Some of them should contain links to most of theSome of them should contain links to most of the prominent authoritiesprominent authorities Expand the root set into a base setExpand the root set into a base set Include all of the pages that the root-set pages link to,Include all of the pages that the root-set pages link to, and all of the pages that link to a page in the root set,and all of the pages that link to a page in the root set, up to a designated size cutoffup to a designated size cutoff Apply weight-propagationApply weight-propagation An iterative process that determines numericalAn iterative process that determines numerical estimates of hub and authority weightsestimates of hub and authority weights Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 86. Systems Based on HITSSystems Based on HITS Output a short list of the pages with large hub weights,Output a short list of the pages with large hub weights, and the pages with large authority weights for the givenand the pages with large authority weights for the given search topicsearch topic Systems based on the HITS algorithmSystems based on the HITS algorithm Clever, Google: achieve better quality search resultsClever, Google: achieve better quality search results than those generated by term-index engines such asthan those generated by term-index engines such as AltaVista and those created by human ontologists suchAltaVista and those created by human ontologists such as Yahoo!as Yahoo! Difficulties from ignoring textual contextsDifficulties from ignoring textual contexts Drifting: when hubs contain multiple topicsDrifting: when hubs contain multiple topics Topic hijacking: when many pages from a single WebTopic hijacking: when many pages from a single Web site point to the same single popular sitesite point to the same single popular site Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 87. Automatic Classification of WebAutomatic Classification of Web DocumentsDocuments Assign a class label to each document from a set ofAssign a class label to each document from a set of predefined topic categoriespredefined topic categories Based on a set of examples of preclassified documentsBased on a set of examples of preclassified documents ExampleExample Use Yahoo!'s taxonomy and its associatedUse Yahoo!'s taxonomy and its associated documents as training and test setsdocuments as training and test sets Derive a Web document classification schemeDerive a Web document classification scheme Use the scheme classify new Web documents byUse the scheme classify new Web documents by assigning categories from the same taxonomyassigning categories from the same taxonomy Keyword-based document classification methodsKeyword-based document classification methods Statistical modelsStatistical models Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 88. Multilayered Web Information BaseMultilayered Web Information Base LayerLayer00:: the Web itselfthe Web itself LayerLayer11:: the Web page descriptor layerthe Web page descriptor layer Contains descriptive information for pages on the WebContains descriptive information for pages on the Web An abstraction of LayerAn abstraction of Layer00: substantially smaller but still: substantially smaller but still rich enough to preserve most of the interesting, generalrich enough to preserve most of the interesting, general informationinformation Organized into dozens of semistructured classesOrganized into dozens of semistructured classes document, person, organization, ads, directory,document, person, organization, ads, directory, sales, software, game, stocks, library_catalog,sales, software, game, stocks, library_catalog, geographic_data, scientific_datageographic_data, scientific_data, etc., etc. LayerLayer22 and up:and up: various Web directory services constructedvarious Web directory services constructed on top ofon top of LayerLayer11 provide multidimensional, application-specific servicesprovide multidimensional, application-specific services Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 89. Multiple Layered Web ArchitectureMultiple Layered Web Architecture Generalized Descriptions More Generalized Descriptions Layer0 Layer1 Layern ... Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 90. Mining the World-Wide WebMining the World-Wide Web Layer-0: Primitive data Layer-1: dozen database relations representing types of objects (metadata) document, organization, person, software, game, map, image, document(file_addr, authors, title, publication, publication_date, abstract, language, table_of_contents, category_description, keywords, index, multimedia_attached, num_pages, format, first_paragraphs, size_doc, timestamp, access_frequency, links_out,...) person(last_name, first_name, home_page_addr, position, picture_attached, phone, e-mail, office_address, education, research_interests, publications, size_of_home_page, timestamp, access_frequency, ...) image(image_addr, author, title, publication_date, category_description, keywords, size, width, height, duration, format, parent_pages, colour_histogram, Colour_layout, Texture_layout, Movement_vector, localisation_vector, timestamp, access_frequency, ...) Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 91. Mining the World-Wide WebMining the World-Wide Web doc_brief(file_addr, authors, title, publication, publication_date, abstract, language, category_description, key_words, major_index, num_pages, format, size_doc, access_frequency, links_out) person_brief (last_name, first_name, publications,affiliation, e-mail, research_interests, size_home_page, access_frequency) Layer-2: simplification of layer-1 Layer-3: generalization of layer-2 cs_doc(file_addr, authors, title, publication, publication_date, abstract, language, category_description, keywords, num_pages, form, size_doc, links_out) doc_summary(affiliation, field, publication_year, count, first_author_list, file_addr_list) doc_author_brief(file_addr, authors, affiliation, title, publication, pub_date, category_description, keywords, num_pages, format, size_doc, links_out) person_summary(affiliation, research_interest, year, num_publications, count) Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 92. XML and Web MiningXML and Web Mining XML can help to extract the correct descriptorsXML can help to extract the correct descriptors Standardization would greatly facilitate informationStandardization would greatly facilitate information extractionextraction Potential problemPotential problem XML can help solve heterogeneity for vertical applications, butXML can help solve heterogeneity for vertical applications, but the freedom to define tags can make horizontal applicationsthe freedom to define tags can make horizontal applications on the Web more heterogeneouson the Web more heterogeneous eXtensible Markup LanguageWorld-Wide Web Consortium19981.0Meta language that facilitates more meaningful and precise declarations of document contentDefinition of new tags and DTDs Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 93. Benefits of Multi-Layer Meta-WebBenefits of Multi-Layer Meta-Web Benefits:Benefits: Multi-dimensional Web info summary analysisMulti-dimensional Web info summary analysis Approximate and intelligent query answeringApproximate and intelligent query answering Web high-level query answering (WebSQL, WebML)Web high-level query answering (WebSQL, WebML) Web content and structure miningWeb content and structure mining Observing the dynamics/evolution of the WebObserving the dynamics/evolution of the Web Is it realistic to construct such a meta-Web?Is it realistic to construct such a meta-Web? Benefits even if it is partially constructedBenefits even if it is partially constructed Benefits may justify the cost of tool development,Benefits may justify the cost of tool development, standardization and partial restructuringstandardization and partial restructuring Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 94. Web Usage MiningWeb Usage Mining Mining Web log records to discover user access patternsMining Web log records to discover user access patterns of Web pagesof Web pages ApplicationsApplications Target potential customers for electronic commerceTarget potential customers for electronic commerce Enhance the quality and delivery of Internet informationEnhance the quality and delivery of Internet information services to the end userservices to the end user Improve Web server system performanceImprove Web server system performance Identify potential prime advertisement locationsIdentify potential prime advertisement locations Web logs provide rich information about Web dynamicsWeb logs provide rich information about Web dynamics Typical Web log entry includes the URL requested, theTypical Web log entry includes the URL requested, the IP address from which the request originated, and aIP address from which the request originated, and a timestamptimestamp Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 95. Techniques for Web usage miningTechniques for Web usage mining Construct multidimensional view on the Weblog databaseConstruct multidimensional view on the Weblog database Perform multidimensional OLAP analysis to find the topPerform multidimensional OLAP analysis to find the top NN users, topusers, top NN accessed Web pages, most frequentlyaccessed Web pages, most frequently accessed time periods, etc.accessed time periods, etc. Perform data mining on Weblog recordsPerform data mining on Weblog records Find association patterns, sequential patterns, andFind association patterns, sequential patterns, and trends of Web accessingtrends of Web accessing May need additional information,e.g., user browsingMay need additional information,e.g., user browsing sequences of the Web pages in the Web server buffersequences of the Web pages in the Web server buffer Conduct studies toConduct studies to Analyze system performance, improve system designAnalyze system performance, improve system design by Web caching, Web page prefetching, and Web pageby Web caching, Web page prefetching, and Web page swappingswapping Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web 96. Mining the World-Wide WebMining the World-Wide Web Design of a Web Log MinerDesign of a Web Log Miner Web log is filtered to generate a relational databaseWeb log is filtered to generate a relational database A data cube is generated form databaseA data cube is generated form database OLAP is used to drill-down and roll-up in the cubeOLAP is used to drill-down and roll-up in the cube OLAM is used for mining interesting knowledgeOLAM is used for mining interesting knowledge 1 Data Cleaning 2 Data Cube Creation 3 OLAP 4 Data Mining Web log Database Data Cube Sliced and diced cube Knowledge Lecture-55 - Mining the World-Wide WebLecture-55 - Mining the World-Wide Web