結合貝氏網路與激勵理論之推薦機制-電影推薦系統設計

Abstract

推薦系統被廣泛應用於網路上，用以幫助使用者快速找到適合或是感興趣的產品，目前已發展出許多的推薦技術，例如內容導向式(Content-based Approach)、協同過濾式 (Collaborative Filtering Approach)、及混合式(Hybrid Approach)等推薦方法。雖然推薦技術已日趨成熟，但其仍存在著一些問題，例如在使用者資訊不足的前提下，推薦系統無法正確的提供推薦資訊。 本論文提出結合貝氏網路與激勵理論(Incentive theory)的推薦機制，在使用者資訊不足的情況下仍然可以產生高度正確率的推薦。透過尋找出使用者輪廓檔(user profile)內資訊之關聯規則，將其視為一個訊息源用以擴大使用者輪廓檔，從而避免使用輪廓檔資訊不足的問題。而在新項目的評比方面，透過激勵理論(Incentive theory)，在符合個人理性(individual rationality)條件與誘因相容(incentive compatible)條件下鼓勵使用者對於新項目分享。實驗證明，透過拓展使用者輪廓檔與鼓勵使用者評比是可以幫助系統在使用者資訊不足的情況下提升系統的推薦準確率。

The recommendation systems are widely used on the network to help users quickly find suitable or interested products. In this area, many recommendation techniques, such as Content-based Approach, Collaborative Filtering Approach, and Hybrid Approach, have been developed. Although the recommendation technology has become mature, there are still some problems. Recommendation systems cannot provide the correct information if we don’t have enough user information. The precision of the recommendation system will be increased dramatically, if a system gets more user potential information. In this study, we attempts to propose a recommendation mechanism design combined with Bayesian networks and incentive theory. In lack of user profile information, our recommendation mechanism still has high precision. This mechanism approach uses Bayesian network to generate association rules from user profiles, a source of information used to expand the user profile, and avoid the problem of user profile shortage. In the new items problem, based on the theory of incentives, we propose a mechanism for encouraging data sharing. This encouraging mechanism satisfies individual rationality and incentive compatibility. Our experiments show that our proposed mechanism can significantly improve the performance of a recommender system under the short user profile situation.