Application And Research On Personalized Information Recommendation System

The latest development of the Internet and network technology has led to the rapid growth of information accumulation,which has also led to the problem of information overload.In this context,the concept of recommender system and various recommender system technologies have been put forward one after another,and recommender system brings convenience to both users and platforms.The field of recommender system has faced with many problems and challenges,including data sparsity and cold start problems in most existing recommendation algorithms,which result in inaccurate recommendation.There are many methods to improve the recommender system.This paper studies in collaborative filtering and deep learning,and the main innovations are as follows:Firstly,in view of the limitations of traditional collaborative filtering algorithms and data sparsity,a hybrid recommendation model CF-K based on CF and K-means clustering was proposed.This algorithm firstly extracts a certain number of feature types using PCA dimension reduction technology,then uses K-means method to cluster,and divides the whole data set into a specific number of clusters.Finally,SVD collaborative filtering method is applied to alleviate the sparsity of the scoring matrix,effectively shorten the recommendation time and improve the recommendation accuracy.Secondly,in order to learn the potential features of users and items more effectively,Neural FC,a neural model fusing context for recommendation,is proposed based on the deep learning method on the basis of explicit feedback and implicit feedback.The model combines the deep matrix decomposition technology with the convolutional neural network to extract the comment text information while acquiring the matrix information,and then to better integrate the feature vectors.Neural FC consists of four parallel neural networks,two of which use multi-layer perceptron(MLP)to extract the vector information of the rating matrix to obtain the potential feature vectors of users and items.In the other two works,convolutional neural network(CNN)is used to extract relevant information of comment text to obtain additional feature vectors.Finally,the deep features of users and items are fused by constructing the fusion layer to achieve the purpose of rating prediction.Finally,through the experiment and analysis of models,the results figure out that the models can effectively alleviate the drawbacks of data sparsity to some extent,and improve the performance of the recommender system by utilizing deep features.