Research On User Interest Mining Algorithm Based On Network Comment And Behavior Serialization Information

User interest modeling is a research hotspot in recommendation systems,which requires accurately identify users interests from their purchase records and exactly extract relevant interest features.The results of user interest modeling directly determine the subsequent decisions of the recommendation system.Therefore,how to extract user interests from user interaction data becomes a very valuable task.Among them,user behavior sequence and review information are two kinds of important data information in user interaction records.User interest modeling can be divided into two types based on different predicted goals.One is rating prediction,which predicts the user's rating of unpurchased items.Rating is an exact score,so this is an explicit feedback prediction problem.Currently,most methods use review information to enhance user interest expression.The other is click-through rate prediction,which predicts the probability of a user clicking on an ad.This is an implicit feedback prediction problem.Now,most methods extract user preference characteristics by modeling user behavior sequence information.Based on the study of basic technology,this article focuses on the research and proposes a method for extracting review information and a method for modeling serialized information.The main research contents are as follows:The current method only uses two separate attention mechanisms to judge the usefulness of each review.Ignoring the fact that the usefulness of each review is dynamic and depends on the target user and product pair,which may make it impossible to effectively model user preferences and product characteristics.In this article,we use the review-level dynamic topic collaborative attention mechanism to collaboratively assign a corresponding weight score to each review in the user and the product.At the same time,due to the sparseness of the review data and the rating data,we use a heterogeneous information network composed of the product category characteristics and the user's common purchase information to alleviate the cold start problem.Experimental results show that the accuracy of this method is 2% to 5% higher than the baseline method,and it also works well on sparse data sets.The existing user behavior sequence modeling methods ignore the problems of user interest shift and timeliness of short-term interests of user,this paper proposes a method,which combines multi-channel memory network and self-attention mechanism to extract long-term and short-term interests of users.On the one hand,the self-attention mechanism is used to mine the short-term interests of users.It divides the entire click sequence into different sessions according to the time interval,and then uses the user's recent session click sequence to represent the user's short-term interests.On the other hand,a multi-channel memory network is used to mine the long-term interests of users and their shift.The multiple channels are used to obtain the user's interest shift of the same category of items,while memory network is used to mine the user's long-term interests.When compared with the traditional RNN-based sequence model,the memory network has not only better memory storage capacity,but also higher interpretability.Finally,experiments show that our model has higher accuracy than the baseline method.