Research On Collaborative Filtering Recommendation Algorithm Based On Autoencoder And Subgroup

With the advent of the era of big data,major Internet companies are paying more and more attention to data,especially the actual demand of e-commerce websites at home and abroad is the driving force for the promotion of recommendation algorithms.The most common recommendation algorithm on e-commerce websites and social networks at home and abroad is to recommend products or topics that users may purchase or are interested in based on historical behavior data of users.In the actual recommendation system,two main factors affecting the accuracy of recommendation: data sparsity,cold start.Collaborative filtering is currently the most widely used algorithm,and it is also the focus of future research systems and the development direction of recommendation systems.The advantage is that the user's hidden features can be mined through the user-commodity scoring matrix,and its ability to process complex information and low data processing requirements can be well adapted to the actual application environment.However,its existence such as data sparsity and cold start problems are still to be solved.This paper combines the review information of the product to further improve the algorithm.The tag generation is completed by extracting keywords.Further,the improved automatic encoder is used to mine the product feature matrix from the label.After filling the original matrix,the sub-group is divided according to the context semantics.Finally,the sub-matrix with good prediction results is used to generate the approximate matrix to obtain the prediction result.details as follows:According to the characteristics of random,unstructured and colloquial comments,the existing label extraction method has the problems of label redundancy and semantic independence from the short texts extracted from the comments.Therefore,an improved K-means clustering label generation method is proposed.This method ensures semantic independence and selects labels according to the label scores in each cluster.The experimental results show that the improved method ensures that the labels are independent of each other while ensuring label accuracy.The ability to mine hidden and overcome data sparsity for single-layer self-encoders is insufficient.In particular,the sparse self-encoder and edge noise reduction self-encoder are combined into a sparse edge noise reduction self-encoder.At the same time,it has the characteristics of two kinds of encoders.The sparse edge noise reduction self-encoder can have better robustness to noise interference of data input,and can overcome the calculation difficulty and long time consumption of edge noise reduction self-encoder.Difficulties.In view of the sparseness and cold start problem of traditional collaborative filtering methods,the paper uses the review text information and context information to alleviate the problem of data sparsity and cold start.This paper proposes a collaborative filtering recommendation method based on improved subgroups.The paper focuses on the effect of comment texts and context information on recommendations.Therefore,the goods are characterized as label vectors,and the hidden features are extracted by the self-programming machine.The migration of potential information to further predict the scoring matrix.The experimental results show that the proposed method effectively utilizes user tags and context information,which improves the recommendation accuracy.