Analysis Of Browsing Behavior Based On Representation Learning

The browsing behavior analysis of large-scale websites excavates similarity relationships between web pages,user browsing preferences and clustering of similar behaviors.The results is applied to e-commerce activities such like website marketing.This paper extract semantics,preferences and behaviors information from the semantics of the URLs of websites,the probability distribution of browsing preferences and the browsing behavior characteristics to.Finally,this paper construct multi-order Markov model and deep memory neural network model considering the respective advantages of graph probabilistic model and neural network model about long-dependency expression ability and trainability,are constructed.The main contributions are summarized below:(1)URL semantic representation learning: Using the distributed semantics of URL to build a neural probability symbolic model.It maps the URL to the high-dimensional semantic feature space to express the similar semantics of the URL by the distance of the feature vector.The URL vector representation can be used as the input of the LSTM model and for the analysis of the website Structural.(2)Browsing preference pattern: Using the multi-order Markov graph probability model and LSTM neural network model to learn browsing behavior habits and infer the probability distribution of the page access under conditions of current environment.This can express the browsing behavior preferences in different environments to apply in the field of personalized recommendation and web page link structural optimization.(3)Behavioral semantics representation learning: The browsing trajectory of users is mapped to the high-dimensional semantic feature space.It can approximately express the similar semantic information between different users by the distance in the vector space which is used for website traffic analysis,real-time monitoring and hot spot analysis.(4)Visual comparison of experiments: Experiments were conducted on actual ecommerce website data to calculate the page access probability distribution by the graph model and the neural network model in the current environment and analyze the match between the page structure and user behavior.Mapping URLs and user behaviors into high-dimensional semantic feature space to mine the semantic relevance between different types of pages.This is to analyze the reasonableness of the page structure,analyze traffic distribution and user groups of the site from two dimensions of time and space;Finally,experiment is conducted to analyze hot spot on specific stores.