Improvement Of Neural Collaborative Filtering Model And Its Application In Recommendation System

With the development of information technology,the ways in which people can obtain information resources are becoming more and more convenient,and the information resources they are facing are becoming more and more abundant.The problem of "information overload" has become increasingly prominent,making it difficult for people to choose what they want in the face of massive information resources.content.In order to solve the above problems,a recommendation system came into being.The recommendation system is one of the most effective ways to solve the problem of information overload.It mainly learns the user's preferences based on the user's explicit and implicit feedback information on the project,and quickly locates the content of interest for the user.The recommendation algorithm is the core of the recommendation system,which determines the performance of the recommendation.Collaborative filtering algorithm is the most widely used algorithm in the recommendation field,but this algorithm faces problems such as data sparseness and cold start.Although collaborative filtering based on matrix factorization solves the data sparse problem of collaborative filtering algorithm well,there are deficiencies in the inner product process of the vector between the user and the item,which limits the expressive ability of the model.Due to the limitations of traditional recommendation algorithms,researchers are gradually exploring collaborative filtering based on neural networks.Neural collaborative filtering algorithms are one of the research results.Although the neural collaborative filtering model is better than the traditional recommendation model in performance,the model has shortcomings in the use of auxiliary information and the extraction of information features.The neural collaborative filtering model uses historical behavior information of users and items to generate embedded features,and simulates the interaction between users and items through a multilayer perceptron,which solves the shortcomings of some traditional recommendation algorithms.Although the performance has been improved,the characteristic information used by the algorithm is relatively single,and only the interaction information between the user and the project is used,and no other auxiliary information is used.Auxiliary information such as user and item attribute tag information plays an important role in improving the personalized recommendation performance of the model and alleviating the cold start of the system.If this information can be introduced into the neural collaborative filtering model,the recommendation performance of the neural collaborative filtering model can be improved.Good,the recommendation is more interpretable.Aiming at the above shortcomings,this paper introduces the attribute tag information of users and items into the neural collaborative filtering model,and proposes a neural collaborative filtering model that combines multiple information features.Through research,it is found that the neural collaborative filtering model will have the following shortcomings after fusing multiple information features: the parameters of the model and the convergence time increase,the model cannot dig a variety of information features well,the model regards the feature information as equally important,and the digestion part has Value information.In response to the above problems,this paper makes further improvements to the neural collaborative filtering model,and proposes the neural collaborative filtering model AHINNCF,which combines the embedded features of heterogeneous information networks and the attention mechanism.In this paper,the effective attribute information of users and items is uniformly converted into label information.By constructing a heterogeneous information network of users and items and their respective attribute label information,the network representation learning algorithm metapath2 vec based on metapath random walks is used to further explore more The characteristics of this attribute information are integrated with the neural collaborative filtering model to improve the performance of the model.The attention mechanism network is introduced into the model to solve the problem that the model regards multiple features as equally important,leading to the elimination of valuable feature information.This article uses two public data sets Movie Lens-1M and Pinterest to experiment with the model in general scenarios,simulated cold start scenarios,and convergence speed.The experimental results show that the model AHINNCF proposed in this paper is more superior.In general scenarios,the performance of the model AHINNCF is improved by an average of 2.5%.In the simulated cold start scenario,the performance of the model AHINNCF is improved by 3%-10%.In terms of model convergence speed,the model AHINNCF converges faster.In summary,the model AHINNCF proposed in this paper is better than the neural collaborative filtering model.Finally,this paper combines the improved neural collaborative filtering model with the actual needs of users to design and develop a recommendation system that integrates relevant online learning platform course information to alleviate the information overload caused by massive course information resources to learners and other related problems.