کنفرانس بین المللی فناوری اطلاعات و دانش

صفحه اصلی / پانزدهمین کنفرانس بین المللی فناوری اطلاعات و دانش

A Mathematical Optimization Approach for Preference Learning in Movie Recommender Systems with Shared Accounts

نویسندگان :

Milad Khademali¹ Fazlollah Aghamohammadi² Marjan Kaedi³ Alireza Nasiri⁴

1- University of Hormozgan 2- University of Hormozgan 3- University of Isfahan 4- University of Hormozgan

کلمات کلیدی :

Movie Recommender System،Convex Optimization،Preference Learning،Shared Account problem

چکیده :

A recommender system assumes that each row of the user-item rating matrix represents a single user preference. However, one account is usually shared among household members, and thus, the ratings data of users in such accounts will be mixed. Consequently, recommendations would severely fail to follow each user's preferences. To solve this problem, we leverage the correspondence of movie and user features from media research and coin a user character concept, a common latent factor in movie and account features. After establishing the movie feature matrix in the character representation, we can identify the presence of different characters in each account by factoring out the account feature binary matrix from the rating matrix. Minimizing the estimation error of a given mixed data matrix leads to a binary quadratic optimization model. Considering scalability, we relax the binary constraint, approximate the solutions to a convex problem, and solve the model via a modified gradient descent algorithm. Finally, based on the identified characters in each account, the preferences will be learned by reconstructing the rating matrix so that each row represents a single user preference. Furthermore, a shared account dataset was generated from MovieLens ratings based on CAMRa2011 to evaluate the method. Experiments on this dataset demonstrate the efficiency of our proposed method.