Design And Implementation Of Content Recommendation System Based On Text Mining And Learning To Rank

With the rapid development of information technology,especially mobile Internet technology,the amount of infonnation on the network platform has grown exponentially.How to find interesting content from massive data is a big challenge for users.As an effective means of solving this challenge,recommendation systems are increasingly being used to provide personalized services to users.However,the diversification of the topic of the article and the timing of changes in user interest make it difficult for traditional recommendation strategies to meet the individual needs of users.How to deeply mine text information and combine machine learning algorithms to recall and sort the recommendation list has become the focus of this system research.Aiming at the above problems,this paper designs a recommendation system based on text mining and learning to rank,and studies the new recommendation schemes,including text modeling algorithm,candidate set recall algorithm and candidate set sorting algorithm,and then designs and implements based on the recommendation strategy.Content recommendation system.The main tasks are as follows:(1)Combining text topic probability model and deep neural network,a recommendation candidate set triggering strategy is proposed.The strategy uses deep neural networks to mine the deep attributes of users and commodities,uses gradient descent and backpropagation to train the model,calculates the feature sequences of users and commodities in the same space,and finds the most relevant product collections through user neighbors..Trying to use the implicit Dirichlet distribution to calculate the text topic distribution and embedding it into the user's reading historical interest sequence as the model's pre-training data not only improves the accuracy of the deep neural network model,but also makes the model loss function quickly converge and improve.Computational efficiency.(2)Combining the deep neural network with the factorization machine,a fusion model is constructed.By constructing the combination of low-order features and high-order features of users and commodities,the model explores the hidden relationship between users and commodities,and realizes the recommended products.Click-through rate forecasting to accurately filter and sort the recall recommendation list in the case of a limited number of client impressions.Online A/B testing showed that the algorithm increased the user's click-through rate by 14%compared to the original tag-based recommendations.(3)Combining the requirements of users and systems,the overall architecture of the content recommendation system is designed,and the online service of the recommendation system is designed based on the Spring Boot framework.The user can request the recommendation and the recommended software and the complex business logic.The list is returned to the client in a timely manner and verifies the functionality and performance of the system.At present,the system has been successfully applied to get APP large-scale article recommendation,which greatly improves the user experience.Figure 40,Table 16,Reference 42.