Research On Neural Collaborative Filtering Recommendation Model Based On Multi-Objective Optimization

The recommendation model is of great significance in solving information overload.However,most of the existing recommendation models are only optimized for a single objective,without considering the diversity of objectives,especially multiple conflicting objectives,so they cannot meet the needs of real users.It has become an urgent and challenging task to select the best information that can optimize multiple objectives at the same time from the massive data and recommend it to specific users.Its importance is increasing.This thesis proposes a multi-objective optimizationbased neural collaborative filtering recommendation model that combines deep neural networks' efficient automatic learning and feature modeling capabilities,as well as the achievements of multi-objective optimization strategies in solving combinatorial optimization problems.The model defines the recommendation problem as a multiobjective optimization problem that simultaneously optimizes the accuracy and diversity of the recommendation results,and obtains good recommendation performance.The main research contents and work are as follows:(1)This thesis proposes a recommendation model that combines a neural network with enhanced attention mechanism.Through implicit feedback learning,a variety of data fusion methods,and the enhanced attention mechanism proposed in this thesis,the model simulates the deep-level characteristics of user-item correlation,and increases the accuracy of recommended performance.(2)In this thesis,we present a new multi-objective optimization strategy.By implementing greedy strategy to check and correct the individuals,as well as the worst individual inversion operation,etc.,an improved whale algorithm is proposed.Then it is combined with optimization strategy,effectively overcoming the issue that conventional optimization strategies are prone to local optimality.(3)On the basis of point(1),we propose a multi-objective optimization based on neural collaborative filtering recommendation model.The recommendation problem is regarded as a multi-objective problem,with the two objectives of the recommendation results are optimized simultaneously.The model proposed in the previous chapter is regarded as a candidate set generation model,then the candidate set is put into the multiobjective optimization strategy for training,and the results can take into account the diversity without affecting the accuracy,so as to improve the user experience.(4)Comparative experiments were carried out for the proposed model with the existing excellent models using two large public datasets,including Movielens-1m and Pinterest-20.It is proved that the proposed model is superior to the existing models,which can not only improve the accuracy of recommendation results,but also improve the problem of incomplete measurement of recommendation results.