Technology Researches Of Collaborative Filtration Recommendation Based On Trust

The rapid development of the Internet and e-commerce make the rapid growth ofnetwork information, and multiple network applications bring us a convenient service,meanwhile, it sends people to a deep "information overload" swamp. For these userswho clearly know what they need most, with the help of search engines they can findthe very information quickly. But unfortunately, in most case, the users can’t expresstheir needs clearly and accurately, only by consulting online or referencing others’suggestion. Besides that, because lack of transparent and reliable consulting channelsand experience share platforms, result in consulting on-line and sharing shopping-experience information can’t be complete reliably. Therefore, to ensure the user’snormal, reliable online activities, introducing the trust mechanisms has become anurgent issue for the development of e-commerce.The main content of this papercontains the following sections:First, simulate the trust relationship handling process of real life in thee-commerce domain, and model the trust relationships management. Afterinitialization，local trust relationship expansion and trust value calculation, we built amixture network composed by the user’s Auth-trust and the user’s local-trust.Compared to other similar models, the most important feature of this model isincreased the interest factor as the guiding role in the expansion process of trust,providing the trust path probing with a certain selectivity.Second, changed the similarity calculation method by traditional collaborativefiltration algorithms, and raised a new similarity method which based on userspreferences positions, then, combined the trust factor and hobby factor into itemscollaborative filtration recommendation process, providing a new itemrecommendation algorithm, helping users to discover more two-dimensionalsimilarity neighbors which taking both the interest factor and the trust factor intoaccount.Thirdly, based on the analysis of users behavior, the paper made use of user’sin-degree and out-degree trust information, designed a calculation method to measureusers’ trust-behavior similarity (UTBS). Meanwhile, for different rate behaviorcontains different amount of interest to characterize the users, designed a userrating-behavior similarity calculation method (URBS), then, with these to analysis the potential friends’ similar association in behavioral aspect.Fourthly, take full advantage of the potential friends’ similarity in behavior, andcombined the mixed-trust evaluation of the trust model, providing a new friendsrecommendation algorithm for platform users with reliable friend recommendation.Compared to other system of precision10%-15%and recall10%-20%, we can getthe best performance of precision22.47%and recall21.15%.In summary, introducing trust into collaborative filteration recommendationprogress, describing users’ association from interest, trust, behavior and other relatedangles, fully exploited the potential association between users, eased the pressure ofdata sparsity and improved the prediction accuracy.