Recommendation Algorithm Based On Constrained Partition Multi-objective Optimization

Recommendation systems are gaining popularity and growing rapidly in applications such as e-commerce,social media and content providers.Although the recommendation technologies in current recommendation systems mainly focuses on improving the accuracy of the recommendation,considering various needs of users,personalized recommendations that meet multiple evaluation criteria are more suitable for modern recommendation systems.Currently,personalized recommendation tasks have been modeled as multi-objective optimization problems already that address the conflicting performance indicators of accuracy and diversity etc.In this paper,the task of personalized recommendation is modeled as a constraint multi-objective optimization problem.A constrained multi-objective recommendation framework for more application scenario is proposed,and a novel constrained multi-objective optimization algorithm is studied to optimize the list of recommended items.The main research work and innovations in this paper are as follows: First,in Chapter 3,the constrained multi-objective optimization algorithm has been studied.We propose a constraint partitioning method based on minimax strategy(CPM/MS)to solve CMOP.The innovations of this research are as follows:(1)The impact of the distribution of infeasible solutions on selecting solutions is analyzed,and the preprocessing method of the infeasible solutions is given.(2)A constraint preprocessing method based on the minimax strategy is designed.(3)A partition-based constraint processing is proposed for individuals in the population.The experiment uses the CEC2009 test problem set.The effectiveness of CPM/MS algorithm is extensively evaluated on a suite of 10 bound-constrained numerical optimization problems,where the results show that CPM/MS algorithm is able to obtain considerably better fronts for some of the problems compared with some the state-of-the-art multi-objective evolutionary algorithms;In Chapter 4,we studied a variety of evaluation indicators that need to be considered in the recommendation system.The innovations of this research are:(1)Considering that these evaluation indicators are not as important as each other,we propose a recommendation framework based on constrained multi-objective optimization,which treats the indicators that required but not need to be maximized as constraints,so that the algorithm can generate a list of recommended items that meets more application scenarios.(2)In the process of generating candidate item sets,the traditional way of taking unions will cause some extreme solutions which have a greater impact in the mixing process.Therefore,in this section,we introduce the concept of dominating solutions in item rating space and propose a method for generating the candidate item set with the most dominant items.The experiment uses the Movie Lens offline movie data set.The test data set is divided by 25%.From the evaluation results of 10 test users,it can be seen that the algorithm proposed in this chapter can generate a list of recommended items for users with higher accuracy,diversity,and novelty,while satisfying the constraints on the coverage of the system,so that the recommendation system can balance multiple evaluation indicators according to different application scenarios.