The Research Of Collaborative Filtering Recommendation Algorithm Based On Semi-supervised

The core of the recommendation system is the recommendation algorithm.The index to measure the quality of a recommendation algorithm is accuracy.The higher the accuracy,the better the model and the wider the application.However,in reality,the modeling of recommendation algorithm often encounters problems such as data sparsity,poor scalability and cold start.Among them,data sparsity is the most important problem,because the model is based on real training set.The more real labeled samples in training set,the more accurate the training model is.In reality,the real marked data cannot be produced out of thin air,but there are a lot of unmarked data.The problem of data sparsity is manifested in the two-dimensional scoring matrix of user items.Semi-supervised learning can not only use labeled data to train the model,but also use unlabeled data to enhance the model.The work direction of this paper is to introduce semi-supervised learning technology into the research of recommendation algorithm.Based on the new idea of the combination of semi supervised learning and recommendation algorithm,the self-training recommendation algorithm is constructed by combining the semi supervised self-training method with the recommendation algorithm model.On the basis of realizing the algorithm,the semi supervised co-training method is combined with the recommendation algorithm again to construct a more complex but effective co-training recommendation algorithm,and improvement to the original approach to ensemble of co-training method for weighted ensemble,eventually forming the weighted integrated co-training recommendation algorithm.The semi-supervised recommendation algorithm studied in this paper is modularized by the original recommendation algorithm model and can be applied to various recommendation algorithm models for semi-supervised learning and training.The main research content of this article includes the following aspects:1.Aiming at the problem of data sparseness in the recommendation system,this article first introduces the self-training method and proposes a self-training recommendation algorithm.Self-training is a tokenizer that retrains in its own token buffer at each iteration.The algorithm model in the self-training recommendation algorithm proposed in this paper can execute the self-training process to achieve the purpose of enhancing the self-model.2.In the case of data sparseness,in order to further improve the scalability and applicability of the algorithm framework,this paper proposes a co-training recommendation algorithm.The co-training method is a semi-supervised learning technique in which two models learn and enhance each other.After one model in the framework has labeled unlabeled data,it is put into the training set of another model,and the model is retrained.The two modes guide each other until the best effect of each mode is achieved.3.Improve the ensemble result of traditional semi supervised co-training algorithm.Traditionally,the ensemble result of co-training method is only the average value of the sum of RMSE of two learning machines.Considering that the accuracy of the two recommendation algorithm models in co-training recommendation algorithm is not the same,this paper improves it by weighting first and then evaluating.According to the initial RMSE of the model,the ensemble weight of the algorithm model with high accuracy is set to high.