Research On Personalized Recommendation Algorithm Based On User Online Shopping Behavior

With the increasing problem of information overload,the recommended technology emerged.However,there are still some problems in the current recommendation research,among which the recommendation homogenization and the repeated recommendation after shopping are particularly prominent,which has seriously affected the quality of recommendations.For this reason,this thesis studies the two issues and proposes solutions to improve the recommendation quality of the recommendation algorithm.The recommendation should originate from the user's needs and it has an important influence on the quality of recommendations.This thesis first studied the needs of users.user demand is an abstract concept of marketing that is highly changeable and difficult to measure.The study found that the behavior of online shopping users affected by user needs often has a certain purpose.Further research shows that this purpose of online shopping is mostly expressed as user interest.Therefore,based on the study of user interest,this thesis proposes a user-interest-based demand discriminant method.This thesis studies the two issues mentioned above.For the problem of homogeneity of recommendation,this thesis proposes a collaborative filtering algorithm based on user interest and time.First of all,analyze the behavior of online shopping users to extract the characteristics of user clustering,and use the demand discriminant method to determine the user needs,reduce the incorrect recommendation to improve the accuracy when the there is no user needs;then use the heat optimization method and similarity optimization method to optimize the candidate list which can reduces the recommendation rate of bestsellers(a higher percentage of bestsellers will limit the user's horizon,and is not conducive to the sale of long-tailed goods)and the similarity between the recommended list items,so that the purpose of recommending diversity enhancement is achieved.For the repeated recommendation problem,this thesis proposes a collaborative filtering algorithm based on user interest and time.As user interest gradually disappears over time,the time context is introduced into the algorithm to consider the effect of “time effect” on user interest,and an interest decay function is proposed to dynamically track user interest and improve the traditional predictive scoring formula.When the recommendation list is formed,the item which user alreadypurchased has filtered,so that the repeated recommendation rate is decreased.In order to verify the feasibility and effectiveness of the proposed algorithm,a comparison experiment was conducted with Taobao users accessing datasets using existing similar algorithms.The experimental results show that the proposed algorithm has better diversity performance than other similar algorithms under the condition of same accuracy and the repeated recommendation rate of the proposed algorithm is lower compares with other algorithm.