An efficient recommendation method based on server Web-logs

摘要

In this paper, we propose a predictive framework based on server Web log. We use profile to discover the community's interest and an effective index structure named session-tree to perform recommendation. We convert server log into a tree structure. After labeling and linking nodes in the session-tree, we can do sequential matching and recommendation. We present an award function in each session based on evolutionary strategy to train the recommendation process and then analyze the common interest within the community. The session-tree structure enables us to retrieve all subsequences both by events and timestamps efficiently. Based on those mechanisms, we predict individuals' coming request and recommend Web pages in server site both on individual and common interest. Through analyzing experimental results, we conclude this method based on adaptive session generation, sequence matching, index structure, evolutionary strategy and common interest profile can recommend well.