A survey of recommendation systems

• Main Authors: Wanzhen ZHOU, Di CAO, Yunfeng XU, Bin LIU
• Format: Article
• Language: zho
• Published: Hebei University of Science and Technology 2020-02-01
• Series: Journal of Hebei University of Science and Technology
• Subjects: computer neural network; recommendation system; data mining; deep learning; information overload
• Online Access: http://xuebao.hebust.edu.cn/hbkjdx/ch/reader/create_pdf.aspx?file_no=b202001009&flag=1&journal_

With the rapid development of the internet, how to mine and analyze massive network information has become a recognized hot and difficult problem. Among them, the recommendation system can provide users with accurate and fast business (commodities, projects, services, etc.) information, which is the common interest and research hotspot of industry and academia in recent years. A recommendation system can help users to solve the problem of information overload when there is no clear demand or a large amount of information. However, at present, the types of data are diverse and the application scenarios are extensive. When faced with this situation, the recommendation system also encounters challenges such as cold start and sparse matrix. Deep learning is an important research field and the most important branch of machine learning. In recent years, deep learning has developed rapidly. Researchers have made great breakthroughs and achievements in speech recognition, image processing, natural language processing and other fields by using deep learning. At present, deep learning has also been favored by a large number of researchers in the field of recommendation and has become a new direction. Incorporating deep learning technology into the recommendation method can effectively solve the problems of cold start and sparse matrix in traditional recommendation systems, and improve the performance and recommendation accuracy of the recommendation system. 河北科技大学学报 2020年 第1期 周万珍，等:推荐系统研究综述 This paper mainly summarizes the application of traditional recommendation methods and the application of neural network in current deep learning technology in recommendation methods, among which the traditional recommendation methods can be divided into the following three categories: 1) Content-based recommendation methods is mainly based on the feature information between the user and the project. The connection between users will not affect the recommendation result, so there is no problem of cold start and sparse matrix, but the content-based recommendation results are low in novelty and face the problem of feature extraction. 2) The collaborative filtering recommendation method is the most widely used method that does not require information about users or items, but only makes accurate recommendations based on the user's interactions with items such as clicks, views, and ratings. Although this method is simple and effective, sparse matrix and cold start problems will occur. 3) The hybrid recommendation method combines the characteristics of the first two traditional recommendation methods and can achieve good recommendation effect. However, this method still faces some challenges and difficulties in processing multi-source heterogeneous auxiliary information such as text and images.Recommendation methods based on deep learning are mainly classified according to neural network categories, which are divided into the following four categories: Recommendation methods based on deep neural network (DNN); recommendation methods based on convolutional neural network (CNN); recommended methods based on cyclic neural network (RNN) and long and short term memory neural network (LSTM); and recommended methods based on graph neural network (GNN). Incorporating deep learning technology into the recommendation field, the constructed model has the following five advantages: it has strong representation ability, and can directly extract the characteristics of users and items from the content; with strong anti-noise ability, it can easily process data with noise; in deep learning, cyclic neural network can model dynamic or sequential data; it can learn user or project characteristics more accurately; and deep learning facilitates the unified processing of data and can process large-scale data. Applying deep learning technology to the recommendation field can effectively overcome the challenges faced by traditional recommendation methods and improve the recommendation effect.