Web CharacteristicsCE-324: Modern Information Retrieval Sharif University of Technology
M. Soleymani
Fall 2017
Most slides have been adapted from: Profs. Manning, Nayak & Raghavan (CS-276, Stanford)
Some slides have been adapted from: Profs. Leskovec, Rajaraman, and Ullman (Mining of Massive Datasets course, Stanford)
Web document collection
No design/co-ordination
Distributed content creation, linking,democratization of publishing
Content includes truth, lies, obsolete information,contradictions …
Unstructured (text, html,…), semi-structured (XML,annotated photos), structured (Databases)…
Scale much larger than previous text collections …but corporate records are catching up
Growth – slowed down from initial “volumedoubling every few months” but still expanding
Content can be dynamically generatedThe Web
Sec. 19.2
2
Web search basics
The Web
Ad indexes
Web Results 1 - 10 of about 7,310,000 for miele. (0.12 seconds)
Miele, Inc -- Anything else is a compromise At the heart of your home, Appliances by Miele. ... USA. to miele.com. Residential Appliances. Vacuum Cleaners. Dishwashers. Cooking Appliances. Steam Oven. Coffee System ... www.miele.com/ - 20k - Cached - Similar pages
Miele Welcome to Miele, the home of the very best appliances and kitchens in the world. www.miele.co.uk/ - 3k - Cached - Similar pages
Miele - Deutscher Hersteller von Einbaugeräten, Hausgeräten ... - [ Translate this
page ] Das Portal zum Thema Essen & Geniessen online unter www.zu-tisch.de. Miele weltweit ...ein Leben lang. ... Wählen Sie die Miele Vertretung Ihres Landes. www.miele.de/ - 10k - Cached - Similar pages
Herzlich willkommen bei Miele Österreich - [ Translate this page ] Herzlich willkommen bei Miele Österreich Wenn Sie nicht automatisch weitergeleitet werden, klicken Sie bitte hier! HAUSHALTSGERÄTE ... www.miele.at/ - 3k - Cached - Similar pages
Sponsored Links CG Appliance Express Discount Appliances (650) 756-3931 Same Day Certified Installation www.cgappliance.com San Francisco-Oakland-San Jose, CA Miele Vacuum Cleaners Miele Vacuums- Complete Selection Free Shipping! www.vacuums.com Miele Vacuum Cleaners Miele-Free Air shipping! All models. Helpful advice. www.best-vacuum.com
Web spider
Indexer
Indexes
Search
User
Sec. 19.4.1
3
Web graph
HTML pages together with hyperlinks between them
Can be modeled as a directed graph
Anchor text: text surrounding the origin of the hyper-
link on page A
4
SPAM(SEARCH ENGINE OPTIMIZATION)
7
The trouble with paid search ads …
It costs money. What’s the alternative?
Search Engine Optimization:
“Tuning” your web page to rank highly in the algorithmic search
results for selected keywords
Alternative to paying for placement
Thus, intrinsically a marketing function
Performed by companies, webmasters & consultants
(“Search engine optimizers”) for their clients
Some perfectly legitimate, some very shady
Sec. 19.2.2
8
Search Engine Optimizer (SEO)
Motives Commercial, political, religious, lobbies
Promotion funded by advertising budget
Operators Contractors (Search Engine Optimizers) for lobbies, companies
Web masters
Hosting services
Forums E.g.,Web master world ( www.webmasterworld.com )
Search engine specific tricks
Discussions about academic papers
Sec. 19.2.2
9
Simplest forms
First generation engines relied heavily on tf/idf The top-ranked pages for the query maui resort were the ones
containing the most maui’s and resort’s
SEOs responded with dense repetitions of chosen terms e.g.,maui resort maui resort maui resort
Often, the repetitions would be in the same color as thebackground of the web page
Repeated terms got indexed by crawlers
But not visible to humans on browsers
Pure word density cannot
be trusted as an IR signal
Sec. 19.2.2
10
Variants of keyword stuffing
Misleading meta-tags, excessive repetition
Hidden text with colors, style sheet tricks, etc.
Meta-Tags = “… London hotels, hotel, holiday inn, hilton, discount, booking, reservation, …”
Sec. 19.2.2
11
Cloaking
Serve fake content to search engine spider
DNS cloaking: Switch IP address. Impersonate
Is this a Search
Engine spider?
N
Y
SPAM
Real
DocCloaking
Sec. 19.2.2
12
More spam techniques
Doorway pages
Pages optimized for a single keyword that re-direct to thereal target page
Link spamming
Mutual admiration societies, hidden links, awards
Domain flooding: numerous domains that point or re-direct toa target page
Robots
Fake query stream – rank checking programs
“Curve-fit” ranking programs of search engines
Millions of submissions via Add-Url
Sec. 19.2.2
13
The war against spam
Quality signals: Preferauthoritative pages based on: Votes from authors (linkage signals)
Votes from users (usage signals)
Policing of URL submissions Anti robot test
Limits on meta-keywords
Robust link analysis Ignore statistically implausible
linkage (or text)
Use link analysis to detectspammers (guilt by association)
Spam recognition by machinelearning Training set based on known spam
Family friendly filters Linguistic analysis, general
classification techniques, etc.
For images: flesh tone detectors,source text analysis, etc.
Editorial intervention Blacklists
Top queries audited
Complaints addressed
Suspect pattern detection
14
More on spam
Web search engines have policies on SEO practices theytolerate/block http://help.yahoo.com/help/us/ysearch/index.html
http://www.google.com/intl/en/webmasters/
Adversarial IR: the unending (technical) battle betweenSEO’s and web search engines
Research http://airweb.cse.lehigh.edu/
15
Understanding the users
16
User Needs Need [Brod02, RL04]
Informational – want to learn about something (~40% / 65%)
Navigational – want to go to that page (~25% / 15%)
Transactional – want to do something (web-mediated) (~35% / 20%)
Access a service
Downloads
Shop
Gray areas
Find a good hub
Exploratory search “see what’s there”
Low hemoglobin
United Airlines
Seattle weather
Mars surface images
Canon S410
Car rental Brasil
Sec. 19.4.1
17
Users’ empirical evaluation of results
Quality of pages varies widely Relevance is not enough
Other desirable qualities (non IR!!)
Content:Trustworthy, diverse, non-duplicated, well maintained
Web readability: display correctly & fast
No annoyances: pop-ups, etc.
Precision vs. recall:
On the web, recall seldom matters
What matters Precision at 1? Precision above the fold?
Comprehensiveness – must be able to deal with obscure queries
Recall matters when the number of matches is very small
User perceptions may be unscientific, but are significant overa large aggregate
19
Users’ empirical evaluation of engines
Relevance and validity of results
Coverage of topics for polysemic queries
Trust – Results are objective
UI – Simple, no clutter, error tolerant
Pre/Post process tools provided Mitigate user errors (auto spell check, search assist,…)
Explicit: Search within results, more like this, refine ...
Anticipative: related searches
Deal with idiosyncrasies Web specific vocabulary (Impact on stemming, spell-check, etc.)
Web addresses typed in the search box
20
DUPLICATE DETECTION
21
Duplicate documents
The web is full of duplicated content
Strict duplicate detection = exact match
Not as common
But many, many cases of near duplicates
E.g., last-modified date the only difference between two copies
of a page
Sec. 19.6
22
Duplicate/near-duplicate detection
Duplication: Exact match can be detected with fingerprints
Near-Duplication:Approximate match
Overview
Compute syntactic similarity with an edit-distance
measure
Use similarity threshold to detect near-duplicates E.g., Similarity > 80% => Docs are “near duplicates”
Not transitive though sometimes used transitively
Sec. 19.6
23
Computing Similarity
Features:
Segments of a doc (natural or artificial breakpoints)
Shingles (Word N-Grams)
Similarity Measure between two docs (= sets of shingles)
Jaccard coefficient:𝐴∩𝐵
𝐴∪𝐵
Sec. 19.6
24
Example
25
Doc A:“a rose is red a rose is white”
Doc B:“a rose is white a rose is red”
Doc A: 4 shingles
“a rose is red”
“rose is red a”
“is red a rose”
“red a rose is”
“a rose is white”
Doc B: 4 shingles
“a rose is white”
“rose is white a”
“is white a rose”
“white a rose is”
“a rose is red”
𝐽𝑎𝑐𝑐𝑎𝑟𝑑 = 0.25
Shingles + Set Intersection
Doc AShingle set A Sketch A
Doc BShingle set B Sketch B
Jaccard
Sec. 19.6
26
Computing exact set intersection of shingles between all
pairs of docs is expensive/intractable
Approximate using a cleverly chosen subset of shingles from
each (called sketch)
Estimate𝑆(𝐴)∩𝑆(𝐵)
𝑆(𝐴)∪𝑆(𝐵)based on short sketches of Doc A and B
From sets to Boolean matrices
27
Rows =elements of the universal set.
Example: the set of all k shingles.
Columns =sets.
View sets as columns of a matrix 𝐶; one row for each element
in the universe of shingles.
𝐶𝑖𝑗 = 1 indicates presence of shingle 𝑖 in set 𝑗
Typical matrix is sparse.
Example: Column similarity
28
For types of rows (for a pair of columns)
29
For columns 𝐶𝑖,𝐶𝑗, four types of rows
𝑪𝒊 𝑪𝒋1 1
1 0
0 1
0 0
𝑛11: # of rows where both columns are one (# of the items that
exist in both sets 𝐶𝑖 and 𝐶𝑗)
𝑛10: # of rows where 𝐶𝑖 contains 1 but 𝐶𝑗 contains 0 (# of the items
that exist in both sets 𝐶𝑖 and 𝐶𝑗)
and so on
𝐽𝑎𝑐𝑐𝑎𝑟𝑑 𝐶𝑖 , 𝐶𝑗 =𝑛11
𝑛10 + 𝑛01 + 𝑛11
Minhashing
30
Imagine the rows permuted randomly, define minhash function:
ℎ𝜋(𝐶𝑖) = index of first row with 1 in column 𝐶𝑖 (after the permutation 𝜋on rows)
Use several (e.g., 100) independent permutations to create a
signature for each column.
The signatures can be displayed in another matrix:
The signature matrix – whose columns represent the sets and the rows
represent the minhash values, in order for that column.
Minhashing: example
31
Randomly permute rows
through permutation 𝜋
For each permutation 𝜋 , ℎ𝜋(𝐶𝑖)denotes the index of first row with 1
in column 𝐶𝑖 (after the permutation 𝜋)
A property of minhashing
32
The probability (over all permutations of the rows) that
ℎ𝜋(𝐶𝑖) = ℎ𝜋(𝐶𝑗) is the same as 𝐽𝑎𝑐𝑐𝑎𝑟𝑑(𝐶𝑖 , 𝐶𝑗).
Both are 𝑛11 /(𝑛01 + 𝑛10 + 𝑛11 )!
Proof sketch:
• Look down the permuted columns 𝐶𝑖 and 𝐶𝑗 until we see a 1.
• If both columns have one in this row then ℎ𝜋(𝐶𝑖) = ℎ𝜋(𝐶𝑗) .
However, if only one of them contains 1 ℎ𝜋(𝐶𝑖) ≠ ℎ𝜋(𝐶𝑗)
Proof
We intend to show 𝑃 ℎ𝜋 𝐶𝑖 = ℎ𝜋 𝐶𝑗 = 𝐽𝑎𝑐𝑐𝑎𝑟𝑑(𝐶𝑖 , 𝐶𝑗)
Look down columns 𝐶𝑖, 𝐶𝑗 until first row that at least one of 𝐶𝑖 or 𝐶𝑗are non-zero ⇒ the corresponding shingle to this row is in the 𝐶𝑖 ∪ 𝐶𝑗.
If both 𝐶𝑖 and 𝐶𝑗 are non-zero in this row (ℎ𝜋 𝐶𝑖 = ℎ𝜋 𝐶𝑗 ), the
corresponding shingle is in the 𝐶𝑖 ∩ 𝐶𝑗.
Thus, in each permutation we indeed select a random sample from 𝐶𝑖∪ 𝐶𝑗 and check that if it exist also in 𝐶𝑖 ∩ 𝐶𝑗.
Therefore, the expectation of ℎ𝜋 𝐶𝑖 = ℎ𝜋 𝐶𝑗 on different
permutations 𝜋 shows 𝐽𝑎𝑐𝑐𝑎𝑟𝑑(𝐶𝑖 , 𝐶𝑗)
Sec. 19.6
33
Similarity for signatures
34
The similarity of signatures is the fraction of the
minhash functions in which they agree.
Thinking of signatures as columns of integers, the similarity of
signatures is the fraction of rows in which they agree.
Thus, the expected similarity of two signatures equals the
Jaccard similarity of the columns or sets that the
signatures represent.
And the longer the signatures, the smaller will be the expected
error.
Min Hashing: Example
35
Sketch of a document
Create a “sketch vector” (of size ~200) for each doc 𝐷:
Let 𝑓 map all shingles to 0. . . 2𝑚 − 1 (e.g., 𝑓= fingerprinting).
Indeed, it maps each set of shingles to an m-bit integer.
For i=1 to size of sketch vector
Let 𝜋𝑖 be a random permutation
Pick 𝑆𝑘𝑒𝑡𝑐ℎ𝐷 𝑖 be the minimum of shingle value for doc 𝐷 after
performing permutation 𝜋𝑖 on these numbers
Docs that share ≥ threshold (e.g. 90%) corresponding
sketch vector elements are near duplicates
If the size of sketch vector is 𝑀 , we have 𝐽𝑎𝑐𝑐𝑎𝑟𝑑(𝐴, 𝐵)
≈ 𝑖=1𝑀 𝐼(𝑆𝑘𝑒𝑡𝑐ℎ𝐴 𝑖 =𝑆𝑘𝑒𝑡𝑐ℎ𝐵 𝑖 )
𝑀
Sec. 19.6
36
Computing Sketch[i] for Doc1
Document 1
264
264
264
264
Start with 64-bit f(shingles)
Permute on the number line
Pick the min value
Sec. 19.6
37
Test if Doc1.Sketch[i] = Doc2.Sketch[i]
Document 1 Document 2
264
264
264
264
264
264
264
264
Are these equal?
Test for 200 random permutations: p1, p2,… p200
A B
Sec. 19.6
38
However…
Document 1 Document 2
264
264
264
264
264
264
264
264
BA
Sec. 19.6
39
the shingle with the MIN value in both of Doc1 and Doc2 is
common to both (i.e., lies in the intersection)
Min-Hash sketches: summary
Use 𝑓 to map shingles to 𝑚 bites
Pick P random row permutations of the numbers
MinHash sketch
𝑆𝑘𝑒𝑡𝑐ℎ𝐶[𝑖] shows the first row with 1 in the column Cin the 𝑖-th permutation
Similarity of signatures
Let 𝑠𝑖𝑚[𝑠𝑘𝑒𝑡𝑐ℎ𝐶 , 𝑠𝑘𝑒𝑡𝑐ℎ𝐶′] = fraction of identical elements inthe vectors 𝑠𝑘𝑒𝑡𝑐ℎ𝐶 and 𝑠𝑘𝑒𝑡𝑐ℎ𝐶′ fraction of permutations where MinHash values agree
Observe 𝑠𝑖𝑚[𝑠𝑘𝑒𝑡𝑐ℎ𝐶 , 𝑠𝑘𝑒𝑡𝑐ℎ𝐶′] ≈ 𝐽𝑎𝑐𝑐𝑎𝑟𝑑(𝐴, 𝐵)
Sec. 19.6
40
Implementation of Min-Hashing
41
Suppose one billion rows.
Hard to pick a random permutation of 1…billion.
Representing a random permutation requires 1 billion
entries.
Accessing rows in permuted order leads to thrashing.
Implementation of Min-Hashing
42
A good approximation to permuting rows:
Pick, say, 100 hash functions.
ℎ𝑖(.) gives order of rows for i-th permutation.
For each column and each hash function ℎ𝑖, 𝑆𝑘𝑒𝑡𝑐ℎ𝐶(𝑖)will become the smallest value of ℎ𝑖(𝑟) for which column
𝐶 has 1 in row r.
All signature pairs
We have an extremely efficient method for estimating a
Jaccard coefficient for a single pair of docs.
But we still have to estimate N2 coefficients where N is
the number of web pages (Still slow)
One solution: locality sensitive hashing (LSH)
Another solution: sorting (Henzinger 2006)
Sec. 19.6
43
More resources
IIR Chapter 19
44
