Research On Recommendation Algorithm Based On User Behavior Sequence

The recommendation system uses the user's existing selection process or similar relationship between projects to discover the hidden preference information or items of the target user,thus having the ability to alleviate the increasingly serious "information overload" problem,resulting in a wide range of academia and industry.Focus on and apply in e-commerce,online learning and digital libraries.The core of the recommendation system is the recommendation algorithm.In the current popular recommendation calculation,the prediction of user interest requires a large number of user ratings,comments,trust values,etc.,which are subjective data that clearly reflect the interests of users.However,subjective data is often sparse and will constrain the recommendation quality of recommendation algorithms.This requires researchers to mine more hidden information in user behavior to optimize the recommendation algorithm.This paper studies the hidden upper and lower order relations in the user's historical behavior.In the historical behavior record of the user,there is a certain correlation between behavior and behavior.This paper maps the similarity between behaviors by extracting the semantic relationship between behaviors.The semantic relationship is extracted using the best Word2 v EC technology in natural language processing.Word2 vec calculates the behavior similarity with content semantics,in which the behavioral content as a word is mapped into the vector space,and the Euclidean distance between the vectors.It is described as the similarity between behaviors.Based on the semantic feature extraction of Word2 vec,this paper proposes a recommendation model that uses content semantics and user ratings.In this model,first collect the user's historical behavior,add all user behaviors into the blank dictionary to establish a user behavior dictionary,and then user behavior.The dictionary uses Word2 vec technology to obtain the similarity between user behaviors.The results of the training are used to predict the user's behavior.The first N recommended algorithms are used to obtain the recommended item candidate list 1.Next,the traditional collaborative filtering recommendation algorithm is selected,and the score information is used.In the prediction,the first N recommendation algorithms are still used to obtain the recommended item candidate list 2.Finally,the two users are combined according to the same user to obtain the final item recommendation list.Through Word2 vec,the semantic association between keywords can also be extracted,and the similarity between keywords can be obtained,and the current demand of the user can be obtained by the user's search keyword.The above algorithm is compared with the traditional recommendation algorithm in the MovieLens,FilmTrust and Online_Retail data sets.Experiments show that the recommendation accuracy of the model is greatly improved,and the data sparse problem is alleviated to some extent.