To enable semantics based recommender systems, products and user profiles need to be represented in knowledge uniformly where ontology can be exploited. Product ontology describes the attributes of the product such as appearance, structure, behavior and function, and has a property "service" which describes the services related to the product supplied by the products provider. So service ontology need to be constructed due to its great influences on users when they browse and purchase products. User profile is modeled as a set of triple where Goal is the product a user searches for, Constraint indicates the conditions a user prescribes that must be satisfied by the attributes of the goals and Preference indicates users' preferences in specific dimensions of the attributes of the goals. The constraint and preference in product attributes are obtained through mining user's past browsing behaviors and transaction records. The mining algorithm is given in this paper. The method of implicit rating and weight evaluation of product attributes are also explored in this paper. A hybrid approach combining semantic similarity with collaborative filtering is proposed to generate the recommendation lists for users where the semantic similarity algorithm is adopted to get the nearest neighbors of the active user. The experiment results are presented which demonstrate that our approach is feasible.
展开▼
