An adaptive personal information filtering system is presented. The adaptive learning algorithm of the system is derived from the statistical consideration - what keywords are useful for retrieving items in which the user want to see. According to the algorithm, the system orgnaizes a personal profile to reflect the user's interests from an empty profile through user-system interaction - the system displays items and the user returns the answer, "interesting' or "uninteresting". The system gives the user items accoridng to his or h4er interest estimated using the user's personal profile. The density of the interesting items at the head of the sequence of items sorted by the system is as high as that in the items retrieved using the keyword matching method whose Boolean expression of keywords was made by a researcher who has much experience in patent retrieval. The number of the interesting items included the head of the sorted sequence is larger than that in the retrieved items. A field test is executed to ensure that hte system responds to the varied interests of many users. The items used in the test are Japanese newspaper headlines which are supplied at a rate of 600 per day from information providers. Not only researchers and engineers in various technical fields, but also staff members participated in the test as users. The results of the test suggest that as users. The results of the test suggest that the system can be put to practical use.
展开▼