The Research And Implementation Of Joint Reviews Sparse Linear Recommendation Algorithm

The explosive growth of data volume brought by the Internet has hindered the efficiency of users' information acquisition.The goal of the recommendation system is to help users quickly and accurately find the content they are interested in,to some extent,recommendation system alleviates the information overload.Top-N recommendation containing N items list,which is watched because of the common form of recommendation.For the Top-N recommendation,theoretical research and practical applications confirmed that the sparse linear model based on the similarity of learning items has better effects than other models.This paper explores the development of sparse linear models and finds that it has the following shortcomings: 1)The sparse effect of regularized feature selection needs to be improved;2)Just using user-item scoring which is single-type feedback data as the training sample,occurred the problem of data sparsity;3)The characteristics of item similarity learning in the model are single.For the problems above,the paper proposes a regular constraint method,and thoroughly studies the fusion method of comment text information and sparse model,then introduces two improvements,which achieves good results on Beer and Ml data sets.Finally,the above algorithm is integrated into the alcohol rating recommendation system.The details are as follows:1.The paper Proposes a regular method instead of the original model constraint.This method improves the feature selection effect of matrix W,making it easier to obtain sparse solutions,thus ensuring that each prediction score is represented as a linear combination by a small number of other correlation vectors with high correlation,and at the same time,preventing the model from being over-generated.In order to achieve an increase in the performance of the algorithm.2.The paper Proposes a sparse linear model that combines with comments.The comment texts including user preferences and item attributes are integrated into the model: 1)Reconstructing the user-item scoring matrix.This model predicts the user-scoring matrix as a new training sample by extracting the factorization machine model of deep text features with CNN.Compared with the traditional text-based representation method based on the Bag-ofwordsmodel,the deeper features are extracted and the sparsity of the training samples is improved.2)Similar with the linear relationship between the unrated items and the scoring items in SLIM model,the comment text word vector matrix Constrains the model in the same form.3.The paper designed a personalized recommendation system for wine.Using user-item scoring data and user review text data as training samples,Integrating the above recommendation algorithm into Java Web project.User login,alcohol recommendation and user evaluation function modules are designed and implemented in the system,finally the development testing and deployment are completed.