Research And Application Of Recommendation Algorithm Based On Latent Co-occurrence

The recommendation system has been developed for more than 20 years.It was mostly used for e-commerce websites at the beginning,and now it has been widely used in a wide range of fields,including e-commerce,film and television,music,social platforms,news information,and advertising,etc.The development of network technology has made us get more and more user behavior data,and the type of data has also increased.Therefore,we should make full use of these data to explore the potential needs of users.On the one hand,it can show the information that the user needs and the hot items or long tail items to users.On the other hand,it is also beneficial for the information producer to marketing his own products.Almost all recommendation systems consist of three parts: the front display page,the background log system,and the recommended algorithm system.The recommendation algorithm is the core of research on the recommendation system.At present,the single recommendation algorithm usually has problems such as cold start,etc.Therefore,this paper combines the neighborhood-based recommendation algorithm with the content-based recommendation algorithm to perform mixed recommendation.The hybrid recommendation algorithm makes up for the shortcomings of the single recommendation algorithm,and it makes the recommendation list for the users more precise and the recommendation more comprehensive.The recommendation algorithm is inseparable from the similarity measure between users or projects.Usually,due to the large number of projects and users,the user-item scoring matrix is very sparse,and the proposed recommendation algorithm has the problem of inaccurate recommendation.The paper proposes to apply the co-occurrence latent semantic space model(CLSVSM) to the design of recommendation algorithm,and to apply the concept of “co-occurrence” to the recommendation algorithm,And it also proposes to replace the similarity measure of traditional recommendation algorithm with “co-occurrence intensity”,and to complete the user-score matrix.The traditional recommendation algorithm also has the problem of modeling recommendation only by using a single score data or collecting network user browsing data.These recommendation models have shown great deficiencies in the increasing content of web pages and the improvement of user personalizationrequirements.In the era of big data,the personalized needs of online users are increasing,and the structure of data is also enriched,which including not only numbers,but also texts and pictures.Therefore,this paper adds text commentary data to model which based on the traditional modeling of scoring data.The comment data can not only describe the characteristics of the product,but also express the user's emotional tendency towards the product.The article uses the Meituan comment data,including the score data and the review text data,and it fully extracts the comment data from different dimensions and carry out word segmentation and feature word extraction,and then builds an emotional dictionary for the emotional analysis of the comment text.It not only makes full use of the review text data,but also improves the recommendation accuracy of the recommendation system.The experimental results are good,which can confirm the feasibility of the algorithm.