Research Of Recommendation Algorithm Based On Users' Shopping Behaviors

The rapid development of information technology brings convenience to people while also leads to information overload problem.Recommendation system can be used to explore the interests of users,and accurately predict the possible link between users and items,which has a great significance on alleviating the pressure of information overloading.And it has been widely used in electronic commerce,but still faces the problem of sparse user data,algorithm with cold-start and so on.With the development of internet technology,electronic commerce has accumulated a large number of user data,and how to rationally utilize these data has become a breakthrough to improve the performance of the recommendation system.In this thesis,taking the Alibaba Tmall e-commerce platform as study background,analyzing the operation of purchases online to detect the interest and preference for goods of users,consequently predict the purchase behavior.The main research work is as follows:When it comes to the old users who has have purchase behavior before,due to the low frequency of user's shopping and seldom leaving the evaluation information,the recommendation system faces the problem of sparse user data.To solve this problem,by considering the behavioral characteristics of the users'online shopping,the UIAction algorithm based on characteristics of users'online shopping behavior is proposed.In the algorithm,the history operation data of users is collected,and the potential interest of users is mined by applying the behavior model to complete the lost information of users.Data sets of ali mobile recommendation algorithm are used to simulate with the precision,recall and F₁-measure as the evaluation standard.The simulation results show that the prediction effect of UIAction is better than the existing recommendation algorithm.When it comes to the new users who has not yet have purchase behavior but have some other behavior before,the history information of the users is almost zero,and thus the recommendation system faces the cold start problem.To solve this problem,starting with the users'current operational characteristics and mining the interest and preference of users,the cold-start recommendation algorithm based on user behavior similarity is proposed.Firstly,users with similar behavior are clustered in different categories according to the theory of group behavior similarity and subspace clustering algorithm.And then the conversion rate of users'purchase behavior in each category is calculated.Finally,the naive Bayesian forecasting model is used to calculate the purchase probability of new users on some items and give the recommendation results.Experimental results show that the algorithm can accurately describe the user's current interest and improve the prediction accuracy of the proposed algorithm.