Artificial Intelligence Indexing: Creating Knowledge Bases of Index Terms Ordered by Semantic Relations

摘要

This report summarizes an investigation of concepts related to the development of intelligent, automated text indexing. It discusses the construction of knowledge bases of index terms and semantic relations extracted from several different types of full-text documents. Approaches to automating this indexing process are examined, including work related to semantic knowledge representation, knowledge engineering, natural language understanding, and semantic inferencing. This research was conducted using a software package which uses semantic networks both for representing knowledge and for displaying a graphical data base. An approach to language processing is outlined which does not require a large lexicon but instead makes use of high-frequency, low-content words supplemented with knowledge of key verbs and generic and domain-specific seed knowledge. This approach uses word order, prepositions, and word endings to determine parts of speech, to identify index terms, and to infer the semantic relations among objects, modifiers, actions, and attributes. The parameters of the requisite seed knowledge and a framework for representing and making use of verb knowledge are described. This artificial intelligence approach to indexing is contrasted with human indexing performance. Keywords: Information retrieval. (Author)