Abstract: Understanding user interests is the key to personalize service.User’s interests can be categorized into short-term interests and long-term interests,and may evolve over time.Inspired by artificial immune system (AIS),we skillfully employed immune response process in user interests mining.By combining probability with time,we first introduced the concept temporal dynamics to describe the degree that the user pays attention to a particular interest during a specific time interval.Then,based on AIS,we built classifier to extract user interest tags.Finally,directing at incrementally learning user interest,we built concept temporal dynamics for interest tags,which characterize the attention degree since interests first appears,and on this basis judged whether interests have migrated or being forgotten and assigned weights to each interest tag.The main contribution of this paper is that we creatively applied AIS to user short-term interests and long-term interests mining with incremental feature,which can gracefully reflect the migration of user’s interests.It is a natural and complete model for learning user interests.Experimental result on practical dataset indicates that the proposed learning model can effectively discover topics that user focuses on,with the average precision and recall of 79.5% and 74.4% respectively,which is the most suitable user interest mining model.

Key words: Short-term interests,Long-term interests,Interest-forgotten,Interest migration,Concept temporal dynamics,Incremental mining,Artificial immune system