Based On Neural Network And Singular Value Decomposition Research On Recommended Algorithm

After entering the era of data information explosion,it becomes very difficult for us to obtain the necessary information from the massive data.In this situation,the recommendation system is playing an important role as an effective tool for filtering information,and the way that we get information is upgraded from a simple and limited keyword searches to an efficient accurate information mining.In this paper,we focus on a recommendation algorithm based on singular value decomposition(SVD)and a recommendation algorithm based on word vector driven by data.Based on the research of these two algorithms,this paper proposes a SVD recommendation algorithm based on word vector driven by data.First of all,we had a study on a recommendation algorithm based on Word Vector Data Driven Neural Network(WVNN).Among them,we get the word vector through a neural network model training and then we use it to complete the task of scoring prediction and product recommendation.In this paper,we propose a method of word series concatenation instead of the traditional word vector summing method,which solves the problem that they do not belong to the same word space.Then,We analyzed a recommendation algorithm for SVD and its improved recommendation algorithm: BSVD.To solve the missing data filling problem of SVD,we proposed a method which adding the user's bias in the average score.The experimental analysis of BSVD was carried out to study the effect of different experimental parameters on the performance of the algorithm.To further improve the recommended accuracy,we combine the WVNN algorithm with BSVD algorithm linearly.Compared with the single model,the experimental results show that the linear fusion model's performance is significantly improved.At last,we propose a Base line SVD recommendation algorithm based on Word vector data Driven(W-BSVD).The training of word vector from WVNN is used as the pre-training process of BSVD's factor matrix which can complete the automatic initialization for the factor matrix.The experimental results show that the W-BSVD model has a stable and improved performance.