Research On Board Game Hybrid Recommendation Algorithm Based On Multi-Source Heterogeneous Data

At the beginning of the 21 st century,people's interest in board games has gradually revived,the share of board games in the entertainment market has increased day by day,and this trend is global.With the increasing types and numbers of board games,it is difficult for people to find their favorite board game products among a large number of board games.Therefore,the application of board game recommendation algorithms is born.With the enrichment of data,the recommendation algorithm is gradually developing in the direction of hybrid recommendation integrating multi-source heterogeneous data.Using multi-source heterogeneous data can improve the performance of board game recommendation algorithm.The data used in this article includes user ratings,board game description text,and board game attribute information.These data sources are diverse and have different data structures.Probabilistic Matrix Factorization(PMF)has high prediction accuracy and good scalability,but it only uses scoring data and still has some shortcomings.Based on the PMF algorithm as the basic framework,this paper proposes two board game hybrid recommendation algorithms from different perspectives based on multi-source heterogeneous data.For the PMF algorithm only uses the scoring matrix,there is a problem of data sparsity.This paper uses multi-source heterogeneous data and proposes a board game hybrid recommendation algorithm(PMF-SIM)that integrates PMF and semantic similarity.Use distributed memory model of paragraph vectors(PV-DM)to learn the feature vector of the description text of the board game,and calculate the similarity between the text feature vectors through the cosine similarity,so as to obtain the similarity of the board game description text;the attribute information is preprocessed and spliced into a vector form,using the cosine similarity calculates the similarity between the board game attribute feature vectors to obtain the board game attribute similarity;after fusing the two similarities of text and attributes,the board game similarity regularization item is used to integrate the similarity relationship between the board games into the objective function of the PMF algorithm,so that the feature vectors of similar board games are also similar.Solved the problem of data sparsity and improve the accuracy of scoring prediction.The PMF algorithm can only learn linear features and cannot obtain deep nonlinear features,which leads to insufficient feature extraction.Based on multi-source heterogeneous data,this paper proposes a board game hybrid recommendation algorithm(UACD-PMF)that integrates PMF and deep features.Use the convolutional neural network to learn the deep features of the board game description text,and use the deep neural network to learn the deep features of the board game attributes.After the two deep features are merged,they are added to the board game feature vectors of the PMF algorithm;the content information of the user's rated board games reflects the user's interest,but the content information of each rated board game of the user has a different contribution to the user's interest.This article uses the self-attention mechanism combined with the content characteristics of the user's rated board games to learn the user's deep interest features,and it is integrated into the user feature vectors of the PMF algorithm.The problem of insufficient feature extraction is solved,and the accuracy of scoring prediction is improved.This paper conducts experiments on the two methods proposed.The experiment proves that the accuracy of the score prediction of the two methods is improved to different degrees compared with different baseline algorithms,which proves the effectiveness and superiority of the methods proposed in this paper.