Research On Top-N Recommendation Algorithm Based On Integrated Network

In the context of big data,the main task of the recommendation system is to filter valuable information from the complex and massive data and recommend it to users.The recommend accuracy of traditional recommend algorithms are affected severely by the sparsity of the rating data.Although there are some traditional method solving the sparsity problem,they can only mine the linear relations of users' rating data.It cannot consider the fluctuant preference of users as well.Aiming at the sparseness of rating date and the fluctuation of user interest,this paper proposes a Top-N recommendation model based on integrated neural networks(CDAETN),which is recommended by mining the hidden and non-linear links between data.First,we should consider the data sparsity of the user-item rating matrix,and use the combination of explicit feedback and implicit feedback data.Using the user comment information as an implicit feedback behavior,the convolution neural network is used to extract the comment text and obtain the implicit feature vector of the user interest degree.Then we should use the item vector as the initial weight parameter of the denoising auto-encoders from input layer to the hidden layer.The rating matrix of the explicit feedback data is used as the input data of the denoising auto-encoders,and the unrated items in the matrix are rated and predicted.Second,we should consider that user interest preferences are not stable.The trend of user preference over time is similar to the human forgetting curve.Introducing nonlinear forgetting function and item attributes,we can obtain similarity between items according to item attribute similarity.Finding a set of unrated items that are similar to the recent rating items,we can use similar items with the same time weighting strategy to time-weight the predictions of unrated items and calculate the rating again.Finally,we select Top-N items to generate a recommendation list.Through simulation experiments,the proposed method is compared with the traditional methods(CDAETN),such as collaborative filtering algorithm,SVD-based recommendation algorithm and denoising auto-encoders by F1 value.It can be seen from the simulation experiment results that the Top-N recommendation algorithm(CDAETN)of the integrated neural network proposed in this paper has largest F1 value.It is indicated that the(CDAETN)algorithm proposed in this paper improves the recommendation accuracy rate.