Automatic Discovery of Attribute Words from Web Documents

Automatic Discovery of Attribute Words from Web Documents Kosuke Tokunaga, Jun’ichi Kazama, and Kentaro Torisawa Japan Advanced Institute of Science and Technology (JAIST), Asahidai 1-1, Nomi, Ishikawa, 923-1292 Japan {kosuke-t, kazama, torisawa}@jaist.ac.jp Abstract. We propose a method of acquiring attribute words for a wide range of objects from Japanese Web documents. The method is a simple unsupervised method that utilizes the statistics of words, lexico-syntactic patterns, and HTML tags. To evaluate the attribute words, we also es- tablish criteria and a procedure based on question-answerability about the candidate word. 1 Introduction Knowledge about how we recognize objects is of great practical importance for many NLP tasks. Knowledge about attributes, which tells us from what view- points objects are usually understood or described, is one of such type of knowl- edge. For example, the attributes of car objects will be weight, engine, steering wheel, driving feel, and manufacturer. In other words, attributes are items whose values we want to know when we want to know about the object. More analyti- cally, we tend to regard A as an attribute for objects of class C when A works as if function v = A(o), o ∈ C where v is necessary to us to identify o (especially to distinguish o from o′(̸= o) ∈ C). Therefore, obvious applications of attributes are ones such as summarization [1,2] and question-answering [3]. Moreover, they can be useful as features in word clustering [4] or machine learning. Although the knowledge base for attributes can be prepared manually (e.g., WordNet [5]), problems are cost and coverage. To overcome these, we propose a method that automatically acquires attribute knowledge from the Web. To acquire the attributes for a given class, C (e.g., car), the proposed method first downloads documents that contain class label C (e.g., “car”) from the Web.1 We extract the candidates of attribute words from these documents and score them according to the statistics of words, lexico-syntactic patterns, and HTML tags. Highly scored words are output as attributes for the class. Lexico-syntactic patterns and other statistics have been used in other lexical knowledge acquisi- tion systems [3,4,6,7,8]. We specifically used lexico-syntactic patterns involving the Japanese postposition “no” as used in [8] such as “C no A” where A is an attribute word, which is almost equivalent to pattern “A of C” used in [7] to 1 We use C to denote both the class and its class label (the word representing the class). We also use A to denote both the attribute and the word representing it. R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 106–118, 2005. c ⃝ Springer-Verlag Berlin Heidelberg 2005 Automatic Discovery of Attribute Words from Web Documents 107 find part-whole relations. Novel features of our method are its use of Web search engines to focus on documents highly relevant to the class and its use of statistics concerning attribute words and surrounding HTML tags. One of the difficulties in studying attribute knowledge is that there are no standard definitions of …

Content truncated to 500 words