In this paper we propose a hybrid method of literature recommendation in the academic community. First, we refer the objective recommendation based on HITS algorithm by constructing a directed graph according to the literature citation relation and then select the articles considering the authority and hub score of each article synthetically and add them to the recommendation list. This can narrow the recommendation scope and give a more authoritive recommendation. Second, the subjective recommendation is based on collaborative filtering by comparing the ratings of other similar users for the objects in recommendation list. The difference is we discover the similar user by clustering them. And the experiment shows the method can provide better recommendation results and is timesaving.

The paper presents a novel method to cluster users who share the common interest and discover their common interest domain by mining different users’ search behaviors in the user session, mainly the consecutive search behavior and the click sequence considering the click order and the syntactic similarity. The community is generated and this information will be used in the recommendation system in the future. Also the method is ‘content-ignorant’ to avoid the storage and manipulation of a large amount of data when clustering the web pages by content. The experiment proved it an available and effective way.