Neural Collaborative Filtering Recommendation Model With Multi-factor Information

The recommendation system(RS)can use the user-item history interaction record to learn the potential personal preference of the user and the potential attribute characteristics of the item,which is helpful for the user to locate the content information of the target information accurately and quickly.The collaborative filtering is the earliest and most widely used method in the application of recommended domain.Among them,the matrix factorization(MF)is the most representative algorithm in the collaborative filtering recommendation.The MF technology is mapped the potential factor vector of user and item to the same dimension,and combined them by inner product.However,fitting the relationship between user preference and item characteristics through inner product is limited.The neural network structures can learn arbitrary functions from the data to model the potential features of users and items,which is breaking through the limitations imposed by the inner product,thus improving the nonlinearity model ability of the model.Auxiliary information plays a vital role in the recommendation system to improve user-item interaction.If more auxiliary information introduced into the neural collaborative filtering.The attributes of the item can be improved from more dimensions to build a recommendation model that more accurate,more personalized,effectively mitigating cold start problems.For example,the neighborhood-based approach can help users discover movies with new genres,actors and directors.It is widely used in collaborative filtering recommendation as it is simplicity,effectiveness,and the ability to provide accurate and personalized recommendations.So,we introduce the neighborhood information into the neural collaborative filtering system.The tag information can not only reflect the user's interest characteristics but also describe the characteristics of the resource.Tag-based algorithm will provide a tag description for the recommended item to explain the recommended behavior.Meanwhile,the tag of item is a coarse-grained description of the new item's content attributes,so it can alleviate cold start problem in the model to a certain extent.Therefore,we incorporate tag information into neural collaborative filtering model.The item description text accompanying the new item is a summary description of the item content,which reflects the content characteristics of the item and it is a detailed description of the item.In this paper,we utilized the convolutional neural network to mine the attribute features in the item description text and incorporate it into neural collaborative filtering model,which further alleviates the cold start problem of the model.In this paper,we not only consider the impact of neighboring factors on the results.It also focuses on the mining of item auxiliary information to fully improve the item's attribute characteristics.including text information,tag information,etc.The main innovations and research results of this paper are as follows:(1)Neighborhood-based approach is look for similarities between users and users or items and items to make recommendations.Neighborhood-based method has been widely used in traditional collaborative filtering recommendation.However,there is rarely related work on the neural collaborative filtering model,In order to enhance the ability of personalization and accuracy recommendation,the first part of our work is to introduce the introduce the neighborhood information into the neural collaborative filtering model.(2)Considering that the item's tag information can describe the item content characteristics in a coarse-grained manner,and that can make a tag description for explaining the recommended behavior,which enhancing the semantic interpretability of the model.The introduction of the Tag can alleviate the cold start problem of model.In order to effectively improve the semantic interpretability of the neural collaborative filtering model recommendation results and alleviate the cold start problem of the model,the second part of our work is introducing the tag information into the neural collaborative filtering model.(3)A new item usually be accompanied by relevant item description text when entering the multimedia resource service platform.The text is mainly a general description of the item content,which can reflect the content characteristics of the item.The third part of our work is utilizing neural network to convolute textual description information and introduce it into neural collaborative filtering.As the item description text can provide a high-level overview of the text content information,so it can describe the item content characteristics in a more granular manner,which further alleviate the cold start of the item.(4)Base on three parts work above,we finally propose a new recommendation model TTNNCF(Text,Tag,and Neighborhood-based Neural Collaborative Filtering Model),multi-angle evaluates experiments on real datasets for different cold-start scenarios and general scenarios.The experimental results show that in different experimental scenario,comparing with the existing excellent models,our proposed model can carry out more effective item recommendation and can alleviate the cold start problem faced by the recommendation system.