Integrating Attribute-Based Classification for Answer Set Construction

With the exponential growth of information on the WWW, it is becoming increasingly difficult to find and organize relevant documents. Automatic text classification has been considered as a solution to the problem with its focus mostly on the subject or content of text [1]. Recently, researchers have realized that user information needs are not just based on the subject of a document but also on other properties such as patterns of hyperlinks [2] or text genre [3]. Alternatively, document classification can be used as part of an information retrieval system to improve its efficiency rather than as an independent system [4]. This paper rests on the both directions: a new type of text classification and its use for a retrieval system. The main thrust of our paper is based on our experience in developing and applying an attributebased classification technique, which is in and of itself a new attempt in classification research, in the context of an operational “answer set driven” retrieval system. Our answer set driven retrieval system, like AskJeeves [5], attempts to provide high quality answer documents to user queries by maintaining a knowledge base consisting of expected queries and corresponding answer documents. Since it is still considered difficult, if not impossible, to capture semantics and pragmatics of sentences in user queries and documents, such systems require knowledge bases be built manually so that a certain level of quality is guaranteed. Needless to say, this knowledge base construction process is labor-intensive, typically requiring significant and continuous human efforts [6]. In order to reduce the cost of manually constructing and maintaining answer sets, we have devised a new method of automating the answer document selection process by using the notion of attributed-based classification (ABC).