Research On Collaborative Filtering Recommendation Algorithm With Sparse Data

With the rapid development of information technology and the Internet,recommendation systems that can contact users and information have become more and more widely used.The collaborative filtering recommendation algorithm is the most popular algorithm in the recommendation systems.It analyzes the user and item interaction data to model and predict the user's preference for the item.In most recommendation systems,due to the large number of items and the limited spending power of users,the number of items purchased or evaluated by the user is only 1%or less of the total number of items,resulting in sparseness of user and item interaction data.In order to improve the quality of recommendations,many researchers have tried to alleviate the problem of data sparsity in different ways.To solve the problem of data sparsity,this paper first proposes a collaborative filtering algorithm based on item clustering.The algorithm introduces the auxiliary information of the item,and clusters the items according to the auxiliary information of the item,and considers that the user's preference for the item is affected by two parts,from the same kind of items and from the different kinds of items.By using different similarity measures and introducing personalization parameters,the two parts are complementarily combined to form the final model.Experiments show that the algorithm has better effect than the traditional collaborative filtering recommendation algorithms.Then,this paper proposes a collaborative filtering neural network based on quadric polynomial regression.With the advantage of deep learning in unified processing of data and full mining of data features,collaborative filtering can be implemented on neural networks,which can further alleviate data sparsity.Compared with the singular value decomposition and probabilistic matrix factorization methods,the second-order polynomial regression model has stronger representation ability.Therefore,the neural network structure is realized based on the second-order polynomial regression model.At the same time,the relationship between the potential features of the item and the auxiliary information is extracted by the multi-layer perceptron machine,and high-order features are extracted.Experiments show that in the case of sparse data,the algorithm can extract information from the data better,has smaller error in the score prediction,and has better effect in the Top K recommendation system.The two algorithms have their own advantages and disadvantages.The collaborative filtering algorithm based on item clustering can introduce the auxiliary information of items,to some extent alleviate the sparseness of the user and item interaction data,and the requirements for computing resources are not high,and it has a good explanatory.However,the algorithm effect is easily affected by the uneven distribution of data.The collaborative filtering neural network based on second-order polynomial regression can fully exploit the information in the interactive data and automatically extract features,which can avoid the direct influence of uneven data distribution and have better effects.However,the computing resources are relatively high and not interpretable.For each scenario of different needs,the two algorithms can each get the appropriate application.