Research On User Interest Mining Based On Behavior Sequence

With the rapid development of information technology,electronic commerce has gradually become an indispensable part of people's daily life.In recent years,Taobao,Jingdong and other e-commerce platforms are more and more popular with users.On the shopping website,users can select the goods they are interested in through simple steps such as browsing,collecting,adding shopping carts and purchasing.E-commerce platform records the interaction between users and commodities in the process of shopping.Although these behavior records contain abundant user interest information,they can not directly express the user's preference for commodities.There are difficulties in extracting user interest information,which results in the inadequate use of data value.In order to solve this problem,this paper focuses on the goal of mining implicit interest information contained in the behavior records.Aiming at the data characteristics in the specific application scenarios,it makes a full analysis from the two aspects of behavior and sequence,digs interest information in depth,and constructs an effective interest model.Based on the time attribute extraction sequence of behavior records,the product is represented by vectors according to the context relevance of interactive commodity sequences.The user interest model is constructed by extracting behavior features and sequence features.The specific research contents are as follows:(1)According to the context-related characteristics of interactive commodity sequences,the implicit vector representation of commodities is acquired by item2 vec algorithm,which enables the vector space to express more abundant commodity-related information.Commodity clusters are formed by clustering.Each commodity cluster is regarded as a coarse-grained interest,and an interest model is constructed on the coarse-grained level.(2)Interest commodity prediction,analysis of the relationship between historical behavior records and user interest commodities,extract behavior characteristics based on experience,use RNN(Recurrent Neural Network)to extract sequence features,establish a prediction model,get each user's interest commodity set and user's interest in commodities,and improve the prediction effect through model optimization.(3)The interest model is constructed to describe the user's interest according to the commodity vector space,coarse-grained interest points and interest commodity set.The calculation method of user's interest is proposed to formally describe the user's preferences and provide the basis for personalized recommendation.(4)Experiments verify and analyze the validity of each part of the algorithm on real data sets.By comparing experiments,adjust the model to the optimal state,and analyze the advantages and disadvantages of the model.