Multi-agent Web recommendation method based on indirect association rules

Recommendation systems often use association rules as main technique to discover useful links among the set of transactions, especially web usage data-historical user sessions. Presented in the paper new approach extends typical, direct association rules with indirect ones, which reflect associations existing "between" rather than "within" web user sessions. Both rule types are combined into complex rules which are used to obtain ranking lists needed for recommendation of pages in the web site. All recommendation tasks are distributed between many agents that communicate and transfer their knowledge one another.