Search for Sentiment

Post on 26-Jan-2015

108 views 2 download

Tags:

description

Sentiment analysis meets search: presentation slides for Seth Grimes's talk Search for Sentiment at the Search Engine Meeting, April 27, 2010.

transcript

Search for Sentiment

Seth GrimesAlta Plana Corporation

301-270-0795 -- http://altaplana.com

Search Engine MeetingApril 27, 2010

Search for Sentiment

2

Seth Grimes –Principal Consultant with Alta Plana Corporation.

Contributing Editor, TechWeb’s IntelligentEnterprise.com.

Channel Expert (text analytics), B-Eye-Network.com.

Founding Chair, Sentiment Analysis Symposium, sentimentsymposium.com, and Text Analytics Summit, textanalyticsnews.com.

Search for Sentiment

3

Two assertions:

Human communications are inherently subjective.

Opinion often masquerades as Fact.

Search for Sentiment

4

Facts and FeelingsThe unemployment rate is 9.7%.

Unemployment is WAY TOO HIGH!!

The unemployment rate is higher than it was two years ago (5.1%).

Former U.S. Federal Reserve Chairman Alan Greenspan said on Tuesday that the global recession will "surely be the longest and deepest" since the 1930s, adding that the Obama administration's Troubled Asset Relief Program will be insufficient to plug the yawning financial gap. [Reuters, Feb 18, 2009]

Bernanke is doing a better job than Greenspan.

www.google.com/publicdata

Search for Sentiment

6

Questions for business & government:What are people saying? What’s hot/trending?

What are they saying about {topic|person|product} X?

... about X versus {topic|person|product} Y?

How has opinion about X and Y evolved?

How has opinion correlated with {our|competitors’|general} {news|marketing|sales|events}?

What’s behind opinion, the root causes?

Who are opinion leaders?

How does sentiment propagate across multiple channels?

Search for Sentiment

7

Is sentiment a search problem?

Search for Sentiment

8

Information access w/structure, sentiment:

Sentiment+

Sentiment

User intent?

Search for Sentiment

9

“In this example, you can quickly see that the Drooling Dog Bar B Q has gotten lots of positive reviews, and if you want to see what other people have said about the restaurant, clicking this result is a good choice.”

-- http://googleblog.blogspot.com/2009/05/more-search-options-and-other-updates.html

“In the recap of [Searchology] from Google’s Matt Cutts, he tells us that: ‘If you sort by reviews, Google will perform sentiment analysis and highlight interesting comments.’

-- Bill Slawski, “Google's New Review Search Option and Sentiment Analysis,” http://www.seobythesea.com/?p=1488

Search for Sentiment

11

For better information access, understand user intent.

User intent?

Search for Sentiment

12

We have a decision support need. We=

Consumers

Marketers

Competitors

Managers

Decision support requires tools beyond general-purpose search/information access…

Search for Sentiment

13

Counting term hits, in one source, at the doc level, doesn’t take you far...

Good or bad? What’s behind the posts?

Search for Sentiment

14

Counting -- clicks, not even keywords -- leaves you wondering Why? and So What?

Search for Sentiment

15

“Sentiment analysis is the task of identifying positive and negative opinions, emotions, and evaluations.”

-- Wilson, Wiebe & Hoffman, 2005, “Recognizing Contextual Polarity in Phrase-Level Sentiment Analysis”

“Sentiment analysis or opinion mining is the computational study of opinions, sentiments and emotions expressed in text… An opinion on a feature f is a positive or negative view, attitude, emotion or appraisal on f from an opinion holder.”

-- Bing Liu, 2010, “Sentiment Analysis and Subjectivity,” in Handbook of Natural Language Processing

Search for Sentiment

16

Sentiment analysis turns attitudes into data.

Ingredients:Structured and unstructured sources.Subjectivity – WW&H used over 8,000 clues.

Polarity: positive, negative, (both,) or neutral.

Intensity.

Search for Sentiment

17

There are many complications. Simplified:

Sentiment may be of interest at multiple levels.

Corpus / data space, i.e., across multiple sources.

Document.Statement / sentence.Entity / topic / concept.

Human language is noisy and chaotic!Jargon, slang, irony, ambiguity, anaphora,

polysemy, synonymy, etc.Context is key. Discourse analysis comes into

play.Must distinguish the sentiment holder

from the object: Greenspan said the recession will…

Search for Sentiment

18

Sentiment sources (broadly):NewsSocial mediaEnterprise feedback

Consumption models:PushPull (a.k.a. search)

1. General search engine2. Siloed/vertical search interface3. Application embedded4. Widgets/gadgets

Rated negative?

???

Manual focus

Search for Sentiment

22

An accuracy aside: [WWH 2005] describes an inter-annotator agreement test.10 documents w/ 447 subjective expressions. The two annotators agree on 82% of cases.

Excluding of uncertain subjective expressions (18%) boosts agreement to 90%.

(Wilson, Wiebe & Hoffman, 2005, “Recognizing Contextual Polarity in Phrase-Level Sentiment Analysis”)

Search for Sentiment

23

Claim: You fall far short with (only) --Doc-level analysis.Keyword-based analysis.

For text, you need strong natural language processing (NLP) for information extraction:“A direct opinion is a quintuple (oj, fjk, ooijkl, hi, tl),

where oj is an object, fjk is a feature of the object oj, ooijkl is the orientation or polarity of the opinion on feature fjk of object oj, hi is the opinion holder and tl is the time when the opinion is expressed by hi.” [Liu 2010]

… index at will!

Search for Sentiment

24

Boost accuracy via ratings & classification:

Search for Sentiment

25

Next slides have a few more examples.

A Jodange embeddable “gadget.”Newssift.com, a now defunct media

portal from the Financial Times Group.

Search for Sentiment

26

Search for Sentiment

28

Beyond polarity: “We present a system that adds an emotional dimension to an activity that Internet users engage in frequently, search..”

-- Sood & Vasserman & Hoffman, 2009, “ESSE: Exploring Mood on the Web”

Search for Sentiment

29

Happy Sad AngryEnergetic ConfusedAggravatedBouncy Crappy AngryHappy Crushed BitchyHyper Depressed EnragedCheerful Distressed InfuriatedEcstatic Envious IrateExcited Gloomy Pissed offJubilant GuiltyGiddy IntimidatedGiggly JealousLonelyRejectedSadScared

-----------------------The three prominent mood

groups that emerged from K-Means Clustering on the set of LiveJournal mood labels.

Search for Sentiment

30

Questions?

Comments?