Chapter Preview
TopBackground
Large amount of information is uploaded to the Web and is regarded as a knowledge database. We believe that the usage of the knowledge database can help our decision making in various situations and improve our daily living. However, the knowledge database is too huge and it is not always well-defined. Most of the knowledge database is described by natural language. We cannot sufficiently use it without appropriate text mining techniques. Many studies based on this theme have been still proposed.
For example, Hu and Liuite (2004) proposed a method that analyzes customer reviews on the Web. The method extracts product features and identifies which sentences include opinions for the products. In addition, it summarizes the identified opinions. Morinaga, Yamanishi, Tateishi, and Fukushima (2002) proposed a method that analyzes reputations for products. The method extracts opinions by using syntactic and linguistic rules described by experts. It identifies whether the reviews include positive opinions or negative opinions. Either positive labels or negative labels are assigned to the reviews. The method also attaches the product name and the degree of the confidence to the opinions. Kobayashi, Iida, Inui, and Matsumoto (2005) proposed a method that extracts attribute-value pairs and judges whether the pairs are an opinion of the author. The method uses machine learning techniques. Here, the pairs are composed of target objects and adjective expressions related to them. They are extracted from textual data on the Web sites. On the other hand, Esuli and Sebastiani (2006) proposed SENTIWORDNET for text mining methods which especially focus on opinions in the textual data. It is a lexical resource associated with WordNet (Miller, Fellbaum, Tengi, Langone, Ernst, & Jose, 2012) where WordNet is a huge lexical database of English. It assigns three scores to the terms in WordNet synset. The scores represent how objective, positive, and negative the terms are. Even if these methods do not always focus on the analysis of bulletin board sites, we can apply them to the analysis to some extent. This is because the sites include various textual data. However, they focus on individual sentences or texts and do not sufficiently consider text sets sequentially described by many authors. In the following, this article focuses on the textual data collected from bulletin board sites.
Key Terms in this Chapter
Bulletin Board Site: It is such a Web site offering the place that some participants discuss specific topics. Some discussions are limited to specific participants and the other discussions are open to all participants.
Thread: It is a set of articles contributed by some participants and is a discussion performed on the Web. The articles are related to a specific topic and are arranged in order of their contribution time.
Relevant Expression: It is a word and a phrase directly expressing a specific event. Synonyms are its examples. It is in advance described by users.
SVM: It is an inductive learning method. Originally, it deals with the two-class classification task. It acquires a hyperplane identifying the classes from training examples.
Reputation Analysis: It is a technique that analyzes the evaluation information for services and products. The information includes their rating information and users' comments for them.
Relevant Incident: It is a word and a phrase characterizing an article. It is related to a specific event. Its causes or actions for it are examples. It is appropriately extracted from the article.
Event Extraction: It is a technique extracting events from an article. Here, the events are representative objects, actions, their impressions, and so on. They are designated by users according to their interests.