Chapter 11: Opinion MiningBing LiuDepartment of Computer ScienceUniversity of Illinois at [email protected] Liu, UIC Web Data Mining2Introduction – 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.Bing Liu, UIC Web Data Mining3Introduction – 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 expressed in the user-generated content An intellectually very challenging problem. Practically very useful. Bing Liu, UIC Web Data Mining4Introduction – 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 focused 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.Bing Liu, UIC Web Data Mining5Two types of evaluation Direct 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.Bing Liu, UIC Web Data Mining6Opinion 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 …)Bing Liu, UIC Web Data Mining7Typical 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. HotmailBing Liu, UIC Web Data Mining8Find 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.Bing Liu, UIC Web Data Mining9Find 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): rank pages according to some authority and relevance scores.  The user views the first page (if the search is perfect).  One fact = Multiple facts Opinion search: rank is desirable, however reading only the review ranked at the top is not appropriate because it is only the opinion of one person.  One opinion ≠ Multiple opinionsBing Liu, UIC Web Data Mining10Search opinions (contd) Ranking:  produce two rankings Positive opinions and negative opinions Some kind of summary of both, e.g., # of each Or, one ranking but  The top (say 30) reviews should reflect the natural distributionof all reviews (assume that there is no spam), i.e., with the right balance of positive and negative reviews.  Questions: Should the user reads all the top reviews? OR Should the system prepare a summary of the reviews?Bing Liu, UIC Web Data Mining11Reviews are similar to surveys Reviews can be regarded as traditional surveys. In traditional survey, returned survey forms are treated as raw data.  Analysis is performed to summarize the survey results.  E.g., % against or for a particular issue, etc.  In opinion search,  Can a summary be produced?  What should the summary be?Bing Liu, UIC Web Data Mining12Roadmap Opinion mining – the abstraction Document level sentiment classification Sentence level sentiment analysis Feature-based opinion mining and summarization Comparative sentence and relation extraction SummaryBing Liu, UIC Web Data Mining13Opinion mining – the abstraction(Hu and Liu, KDD-04; Liu, Web Data Mining book 2007) Basic components of an opinion Opinion holder: The person or organization that holds a specific opinion on a particular object. Object: on which an opinion is expressed Opinion: a view, attitude, or appraisal on an object from an opinion holder.  Objectives of opinion mining: many ...  Let us abstract the problem put existing research into a common framework We use consumer reviews of products to