Research On Recommendation Strategy Based On Multi-objective Optimization

The rapid development of the internet has brought people unlimited choices,but also leads to the data explosion.The information grows so fast that customers are overwhelmed by the choices.How to get the most needed data?Recommended system came into being.Many e-commerce and content service providers offer many products to cater to different tastes and needs of users.How to get what we like?It emphasizes the importance of recommendation systems,which provides personalized products to meet the user's preferences.The recommendation system has been widely used in many websites.However,the traditional recommendation system has three deficiencies:(1)Too much attention to the accuracy makes the similarity of recommended items too high,resulting in the recommended items concentrated in one area;(2)It is not conducive to the user experience,no more innovative options;(3)It makes a large number of service providers' goods were buried,can't produce better benefits.Considering these problems,this paper focuses on the requirements of accuracy,diversity and novelty in recommendation system,and researches the multi-objective optimization of recommendation strategy.Details as follows:(1)Proposed a recommendation model based on singular value decomposition and multi-objective immune optimization.The traditional recommendation algorithm is modeled as a multi-objective problem,which takes into account the accuracy of the recommendations,the diversity of recommended results and the novelty of the recommended items.Besides,immune optimization is very effective in solving multi-objective problems.So,the recommended problem is seen as antigen,the candidate solution of target problem is regarded as antibody,and the immune algorithm for solving the multi-objective problem.Finally,an accurate,diverse and novel recommendation list is obtained.(2)Propose a long-tail group recommendation model based on immune multi-objective.According to the accuracy requirement of recommendation list and the status of the long-tail distribution,a multi-objective group recommendation model is proposed.Through the group discovery,group preference fusion,immune optimization and other operations,the model generates multiple group's recommendation results with different accuracy and popularity weights in one iteration.