Research On Product Personalization Recommendation System Algorithm Based On Combinatorial Recommendation

The development of Internet technology promotes the vigorous development of online shopping.More and more product information is uploaded on the Internet.When facing hundreds of millions of products,users often fail to know their purchase intentions at once.This is a loss for both merchants and users.Merchants may miss the right users who have implicit demand for their products while users will waste plenty of time in the process of browsing a large number of products.The above-mentioned problems can be solved to a certain extent by studying personalized recommendation.In the calculation process of the recommended algorithm,the user's implicit feedback data is more readily available and the amount of data is larger than the rarer display feedback data.However,with the increasing complexity of data dimensions,a single recommendation algorithm is becoming more and more difficult to meet the increasingly personalized needs of recommendation problems.This paper analyzes the behavior characteristics of the target users by studying the user's implicit behavior data,and makes recommendations to the target users.Different from display behavior data that includes user ratings and can directly express user preferences,implicit behavior data does not include user ratings,so users cannot directly display user preferences for goods.Aiming at the timeliness of user behavior,this paper uses analytic hierarchy process(AHP),sigmoid function and exponential weighted average method to convert user behavior data into the user's preference value for commodities over a period of time,so that the recommendation results are more accurate.After the calculation of user preference is completed,this paper makes serial combination and parallel combination of the recommended algorithm to make up for the defects of the individual recommendation algorithm and improve the accuracy and diversity of the recommended results.After obtaining the user preference value,this paper decomposes the user-item scoring matrix to obtain the user's characteristics,then uses this feature as a parameter to pass in the cluster-based user recommendation algorithm,and obtains the cluster of the target user and the similar user list of the same cluster1.The improved Pearson function is also used to calculate a similar user list 2for the target user.The two user lists use random forests to combine the idea of calculating the results of the decision tree to obtain the end-user list,and the final similar user list to calculate the recommended list of the target users,improving the accuracy of the recommendation results.Finally,the list of recommended items is supplemented by the collaborative filtering recommendation algorithm based on items.Increase the diversity of recommended results.After the combination of the algorithms is completed,the recommended effect of the algorithm is verified by using the Retailrocket dataset and the Alibaba Cloud dataset.Experimental results show that the combination scheme of recommendation algorithm designed in this paper has better recommendation effect than implicit feedback recommendation,traditional recommendation algorithm and other combination recommendation scheme,and can better provide product recommendation service to users and meet their needs.