The Research Of Recommendation Algorithm Based On Multi-objective Optimization Algorithm

The rapid development of the information technology brings huge information meanwhile results in the problem of information overload. As an effective solution,recommender system has been proposed and introduced e-commerce, search engine, video service, social media and other industries. Along with recommender system providing highly personalized recommendation service for users, the users' interests have gradually led into the small world network. Users in interest group browse as well as click interesting goods, which would constantly make these goods recommended firstly by system in the future, and the highly profitable goods always been relegated to the long tail therefore resulted long-tail phenomenon, this has seriously affected the profits of businesses. Hence, in view of both businesses and users, establish the recommender system for multi-objective requirements is an important issue need to be addressed urgently. Aiming at this issue, this paper made researches as follows:(1) For the efficiency of recommender system, a computing strategy which separates online and offline is adopted and a new architecture of recommender system is constructed.Firstly, the architecture separates the section which need more time and space to do the calculation on offline. Secondly, updates online without interruption for the section which needs timely response. Finally, gives the data interface of mutual feedback between online and offline, thereby form an architecture of recommender system that fast-response, dynamic and feedback each other.(2) For the multi-objective requirements of the recommender system and the numerous limitations of the current recommendation algorithm, by adopting the hybrid strategy, a new multi-objective recommendation algorithm is proposed. Firstly, the algorithm adopts weighted combination with multiple recommendation algorithms,and construct the multiple-objectiveoptimization model whose objective function is built with weight sequence as variables and assessed with F-score, diversity and novelty. Secondly, employs multiple-objective optimization algorithm to solve the model. Lastly, recommends goods to the target users using users' shopping preferences and Pareto solutions.(3) Compare and analyze the multi-objective recommendation architecture and algorithm with separation and mixing experiments, not only verified the feasibility of the architecture and algorithms, but studied the multi-objective recommendation architecture and the issue that selecting child recommendation algorithm in the present algorithm. Introduce two performance evaluations of multi-objective optimization algorithm, homogeneity and extension. The experimental results present that the multi-objective recommendation architecture and algorithm is effective and feasible; more and better single performance evaluations of child recommendation algorithm applied in the proposed method would obtain the distribution of non-dominated solutions having better homogeneity, and the better single performance evaluation of the child recommendation algorithm, the better extension of the non-dominated solutions.