Research And Implementation On Collaborative Filtering Algorithm Base On Item And Mood

In recent years, context awareness recommendation technology becomes one of the hottest topics in the research field of recommender system, its purpose is to use context information further improve the prediction accuracy of recommendation system. Therefore, it is very important significance to study how to use the item-based collaborative filtering system with mood contextual.In this paper, firstly we introduce the related technology of recommendation system; emphatically introduce the collaborative filtering recommendation technology. To solve the recommendation system widespread data sparseness and overfitting problems, this paper puts forward to collaborative filtering (CF) which combines factor decomposition model, neighbor model and mood contextual to alleviate the above problems. The paper gives indepth study to three new proposed algorithms; finally experiments in pubic datasets verify that these algorithms are effective. The main work of this thesis includes the following several aspects:(1) Research recommendation algorithm based on factor decomposition model and emotional context, its purpose is to effectively alleviate the sparse data. The algorithm puts forward to two kinds of factor decomposition models:the first model firstly constructs a two-dimensional "user-item" matrix, uses of SVD matrix decomposition method to reverse filling sparse matrix, then uses the mood contextual to fill the matrix for preliminary filter processing, finally using the Top-K strategy to prediction recommendation; the second model firstly constructs "users-item-mood" multivariate matrix, then trains the model prediction formula using the known ratings information, finally using the model formula to make prediction. Experimental comparisons show that these two models can alleviate the problem of data sparseness.(2) Research recommendation algorithm based on neighbor model model and emotional context, its purpose is to effectively alleviate overfitting problem caused by data sparsity. This glgorithm firstly constructs "user-item-mood" three-dimensional matrix, using the new proposed mood contextual similarity calculation method to reduce weft of three-dimensional model, and get the items’similar neighbor set, finally using the least squares fitting calculation of similar items’interpolation weights to make prediction. Experimental comparisons show that this algorithm can alleviate overfitting problem.(3) Research hybird recommendation algorithm based on factor decomposition model, neighbor model and mood contextual. Summing up the above research, put forward to a new mood contextual similarity calculation method, improvement the construction method of factor decomposition model and neighbor model, and using the method of combination two kinds of model to make prediction. Experimental comparisons show that this algorithm can obtain better recommendation accuracy than the one only uses a kind of model algorithm.The paper aims at the existing data sparseness and overfitting problems of recommender system, respectively proposes the recommendation algorithm which combines the factor decomposition model, neighbor weighted model with mood contextual information, and experimental comparisons show that these three algorithms proposed above are effective, and the hybrid recommendation algorithm can get better recommendation accuracy than the previous two algorithm.