Abstract
Rapid progress of information services has been increasing the amount of text information which users receive on Internet, teletext broadcasting and so on. A system extracting important part automatically is useful for users to get their interest information easily in such a case. Some methods in the field of natural language processing can extract important keywords or phrases from text information by using word dictionaries. However, these methods can hardly deal with unexpected words such as new proper nouns.On the other hand, the methods based on word occurrence are useful for dealing with unexpected words and they have high applicability for many kinds of text information because they use no dictionary. However, the results are insufficient as a summary in some cases.In this paper, we describe our newly developed system KEIFIS (Keyword Extracting and Information FIltering System). KEIFIS extracts important keywords which represent major topics from a large amount of Japanese text information and collects related data with no dictionary.KEIFIS has the following features : (1) It automatically extracts important keywords and combines some of them to give the users major topics.(2) It retrieves and collects information according to the topic specified by the users, and informs them of its arrival in real time.KEIFIS employs fuzzy information processing in computing the similarity between words. The similarity is utilized in counting co-occurrence of the related words to extract the major topics in KEIFIS. The similarity is also used to define relationship between the specified topic and newly provided information.We applied KEIFIS to news programs on the teletext broadcasting in Japanese, and made sure that KEIFIS was capable of extracting the important keywords which represent current major topics appropriately.
