Research On Collaborative Filtering Recommendation Based On Heterogeneous Information Network

With the problem of overloading information becoming more and more serious ine-commerce, collaborative filtering recommendation technology has attracted widespreadattention and research. As information resource in the recommendation system hasdynamic and complex features, the collaborative filtering recommendation technology stillfaces some key issues. In recent years, recommendation technology based on networkstructure is gradually paid attention by scholars and heterogeneous information networksbecome an active area of research. Moreover, the module in the network contains morerelationships among objects and richer semantic information, which provides a newsolution to the research on recommendation technique.Firstly, an in-depth study on collaborative filtering recommendation technology andheterogeneous information networks is conducted in this paper. Then taking heterogeneousinformation networks as the foundation, a collaborative filtering recommendation system isproposed. Finally, the content in this paper is presented as to the critical issues which thecollaborative filtering recommendation technology is facing. Specifically, the whole workis as below:(1) From the point of information networks, heterogeneous recommendation networkmodel is constructed by integrating multi-dimensional information. By linking associatingcharacteristics and updating recommendation network dynamically, user’s interest ofcertain item is predicted which can effectively mitigate the effect of sparseness andscalability and improve the recommended result.(2) Taken into consideration that the data is sparse and user’s behavior is random, it isproposed to generate similarity factor dynamically based on current data environment toimprove the similarity measure. Then the accuracy of recommendation results iseffectively improved by combining comprehensive users with item dimensions in the collaborative filtering recommendation technology.(3) By the analysis of interaction between forecast network and real network, it isproposed collaborative filtering recommendation algorithm based on heterogeneousinformation network. The heterogeneous recommendation network is updated based on thereal-time user behavior data and forecast result in the real network, in which the similarityfactor is generated dynamically to correct neighbor relationship adaptively and therecommendation algorithm based on heterogeneous information network is adopted.Therefore, the method can be applied to large-scale and dynamically updated real networkand has a strong flexibility.In this paper, MovieLens dataset is used to verify the proposed algorithm. Theexperimental result is analyzed to validate accuracy and effectiveness of the proposedalgorithms.