Research On Diversified Recommendation Algorithm Based On Supervised Learning

With the advent of both the large data era and mobile Internet era,people communicate more frequently in closer relationship.The time of lack of information never returns but the era of overload information follows to come.Enormous quantities of e-commerce websites came into being like a blowout,which greatly influenced people’s lifestyle as well as the way of shopping.Under such an internet environment,traditional search algorithm cannot provide individual search list for different customers,so it is unable to satisfy the needs of both customers and market.Therefore,the recommender system came into being as a kind of individual search tool,in order to provide appropriate choices for customers while shopping.According to user’s previous query history,transaction records,user information or product content and other information,the recommender system can help customers select and filter the shopping items,and return the products that consumers are most likely to be interested in.In this way,it not only improves systems’ working efficiency,but also saves users’ time in filtering information as well.With accuracy being the primary concern in recommendation tasks,the recommender systems with poor diversity might narrow down the users’ horizons and make them frustrated.The diversity in recommender systems represents the abundance of a recommendation list.A system with broad horizons may provide a win-win situation:users can find more interesting items and e-commerce enterprises can increase their sales and improve users’ satisfaction.Recently,the approaches about the diversity issue in recommender systems use the heuristic strategy to re-rank the items for recommendation based on certain diversity metric,which mainly involves two steps:generating a candidate set of favorable items based on the accuracy metric,and then selecting k items from the candidates by maximizing the recommendation diversity metric.However,these algorithms either use a limited feature space or require extensive tuning for different parameter settings.In this study,we investigate diversified recommendation problem by supervised learning,seeking significant improvement in diversity while maintaining accuracy.Our main contributions are summarized as follows.(1)We investigate learning-based diversified recommendation problem,targeting significant improvement in recommendation diversity meanwhile maintaining accuracy.Then we formulate two coupled optimization problems:parameterized matrix factorization problem and structural support vector machine learning problem.(2)We propose DCF,a diversified collaborative filtering algorithm to solve the coupled problems.In DCF,structural support vector machine learns a recommendation model that can predict a set of items for each user,and a parameterized matrix factorization process is integrated in each iteration of the structural SVM for generating representations of users and items.(3)We introduce a new pairwise accuracy metric and a normalized topic coverage diversity metric to measure the performance of accuracy and diversity respectively,and show the consistency with traditional measures.Extensive experiments on benchmark datasets demonstrate the significant performance gains of DCF in comparison with the state-of-the-art algorithms.