Chapter 11: Opinion MiningIntroduction – facts and opinionsIntroduction – user generated contentIntroduction – ApplicationsAn Interesting Problem!Two types of evaluationOpinion search (Liu, Web Data Mining book, 2007)Typical opinion search queriesFind the opinion of a person on XFind opinions on an objectSearch opinions (contd)Reviews are similar to surveysRoadmapAn Example ReviewOpinion mining – the abstraction (Hu and Liu, KDD-04; Liu, Web Data Mining book 2007)Target Object (Liu, Web Data Mining book, 2006)Model of an objectModel of a reviewWhat is an Opinion? (Liu, Ch. in NLP handbook)Objective – structure the unstructuredFeature-Based Opinion Summary (Hu & Liu, KDD-2004)Visual Comparison (Liu et al. WWW-2005)Feat.-based opinion summary in BingOpinion Mining is Hard!It is not Just ONE ProblemOpinion mining tasksOpinion mining tasks (contd)Slide 28Slide 29Sentiment classificationUnsupervised review classification (Turney, ACL-02)Slide 32Slide 33Sentiment classification using machine learning methods (Pang et al, EMNLP-02)Review classification by scoring features (Dave, Lawrence and Pennock, WWW-03)Slide 36Sentence-level sentiment analysisUsing learnt patterns (Rilloff and Wiebe, EMNLP-03)Subjectivity and polarity (orientation) (Yu and Hazivassiloglou, EMNLP-03)Let us go further?Slide 41But before we go furtherCorpus-based approachesCorpus-based approaches (contd)Slide 45Slide 46Rules from dependency grammarDictionary-based approachesSlide 49Feature-based opinion mining and summarization (Hu and Liu, KDD-04)The tasksAspect extraction(Hu and Liu, KDD-04; Liu, Web Data Mining book 2007)Using part-of relationship and the Web (Popescu and Etzioni, EMNLP-05)Infrequent features extractionUsing dependency relationsSlide 56Identify aspect synonyms (grouping)Aspect sentiment classificationAggregation of opinion words (Hu and Liu, KDD-04; Ding and Liu, 2008)Context dependent opinionsBasic Opinion Rules (Liu, Ch. in NLP handbook)Slide 62Divide and ConquerSlide 64Extraction of Comparatives (Jinal and Liu, SIGIR-06, AAAI-06; Liu’s Web Data Mining book)Two Main Types of OpinionsComparative Opinions (Jindal and Liu, 2006)Types of comparatives: non-gradableMining Comparative OpinionsSlide 70Opinion Spam Detection (Jindal and Liu, 2007)An Example of Practice of Review SpamExperiments with Amazon ReviewsDeal with fake/untruthful reviewsDuplicate ReviewsFour types of duplicatesSupervised model buildingPredictive Power of DuplicatesSpam ReviewsHarmful Spam are Outlier Reviews?Some Tentative ResultsSlide 82SummaryChapter 11: Opinion MiningCS583, UIC2Introduction – facts and opinionsTwo main types of textual information on the Web. Facts and OpinionsCurrent search engines search for facts (assume they are true)Facts can be expressed with topic keywords.Search engines do not search for opinionsOpinions are hard to express with a few keywordsHow do people think of Motorola Cell phones?Current search ranking strategy is not appropriate for opinion retrieval/search.CS583, UIC3Introduction – user generated contentWord-of-mouth on the WebOne can express personal experiences and opinions on almost anything, at review sites, forums, discussion groups, blogs ... (called the user generated content.)They contain valuable informationWeb/global scale: No longer – one’s circle of friendsOur interest: to mine opinions (sentiments) expressed in the user-generated content An intellectually very challenging problem.Practically very useful.CS583, UIC4Introduction – ApplicationsBusinesses and organizations: product and service benchmarking. Market intelligence. Business spends a huge amount of money to find consumer sentiments and opinions.Consultants, surveys and focus groups, etcIndividuals: interested in other’s opinions when Purchasing a product or using a service, Finding opinions on political topics, Ads placements: Placing ads in the user-generated contentPlace an ad when one praises a product. Place an ad from a competitor if one criticizes a product. Opinion retrieval/search: providing general search for opinions.An Interesting Problem!Intellectually challenging & major applications.A very popular research topic in recent years in NLP and Web data mining. 20-60 companies in USA alone It touches everything aspect of NLP and yet is restricted and confined.Little research in NLP/Linguistics in the past.Potentially a major technology from NLP. But it is not easy!CS583, UIC5CS583, UIC6Two types of evaluationRegular Opinions: sentiment expressions on some objects, e.g., products, events, topics, persons.E.g., “the picture quality of this camera is great”SubjectiveComparisons: relations expressing similarities or differences of more than one object. Usually expressing an ordering. E.g., “car x is cheaper than car y.”Objective or subjective.CS583, UIC7Opinion search (Liu, Web Data Mining book, 2007)Can you search for opinions as conveniently as general Web search?Whenever you need to make a decision, you may want some opinions from others, Wouldn’t it be nice? you can find them on a search system instantly, by issuing queries such as Opinions: “Motorola cell phones”Comparisons: “Motorola vs. Nokia”Cannot be done yet! (but could be soon …)CS583, UIC8Typical opinion search queriesFind the opinion of a person or organization (opinion holder) on a particular object or a feature of the object. E.g., what is Bill Clinton’s opinion on abortion?Find positive and/or negative opinions on a particular object (or some features of the object), e.g., customer opinions on a digital camera.public opinions on a political topic. Find how opinions on an object change over time. How object A compares with Object B?Gmail vs. HotmailCS583, UIC9Find the opinion of a person on XIn some cases, the general search engine can handle it, i.e., using suitable keywords. Bill Clinton’s opinion on abortionReason: One person or organization usually has only one opinion on a particular topic. The opinion is likely contained in a single document.Thus, a good keyword query may be sufficient.CS583, UIC10Find opinions on an objectWe use product reviews as an example:Searching for opinions in product reviews is different from general Web search.E.g., search for opinions on “Motorola RAZR V3”General Web search (for a fact):