Session-Based Recommendations

Session sequence refers to a set of items used in a user's interactive transaction(such as the user's history of the product clicked during a shopping process).As a special data form in the field of recommendation,session sequence is often difficult to model session sequence data by traditional recommendation methods.Although some scholars have improved the traditional recommendation method to solve the problem of session sequence modeling,the effectiveness of the recommendation system is limited by many problems in the model structure.In recent years,with the breakthrough progress of deep learning in artificial intelligence fields such as image recognition,natural language processing,and speech recognition,incorporating deep learning into the recommendation algorithm can help the traditional recommendation algorithm effectively solve multi-source heterogeneous information and data sparse,Cold start,artificial dependency of data feature design,etc.,also brought a new method to solve the recommendation problem based on session sequence.Considering that in real life,the recommendation system is often recommended based on the user's short-term session sequence data,rather than a large number of user history records.In this case,the recommendation result of the traditional recommendation method is often not accurate enough,and the recommendation result often has lag and repeatability.This article uses session sequence data to model the recommendation algorithm,and according to the characteristics of the main information in the session sequence that is composed of the association between the items,the Apriori algorithm and the recurrent neural network are used to model the user's session sequence data.And through experiments to study the conversation sequence recommendation model based on different algorithms.The main contents of this article are as follows:(1)Apriori algorithm is used to implement the recommendation of the session sequence according to the characteristics of a large number of association relationships between the session sequence data.In the process of modeling using the Apriori algorithm,for the problem of long calculation time and large memory requirements of the Apriori algorithm,an auxiliary matrix is established to improve the Apriori algorithm based on the compression matrix,and the auxiliary matrix is used to reduce the time and scanning range of the calculation support.Improve Apriori algorithm performance.Aiming at the problem that the recommendationalgorithm needs to rank items,a method of calculating item weights is proposed to determine the priority of items.Finally,the effectiveness of the improved Apriori algorithm running speed and recommendation effect are verified through experiments.(2)In order to further improve the recommendation effect of the recommendation system,make full use of the correlation information between the user's session sequence during the interaction process.To solve the problem that the Apriori algorithm cannot mine timing information,the recurrent neural network is used as the recommendation model to achieve the conversation sequence recommend.In the research process,in order to better simulate the change of the session sequence in the real scene,before the modeling,the data is pre-processed according to the characteristics of the large difference in the length of the session sequence.Using a sequence sampling method different from the previous research.The single classification problem is transformed into a sequence modeling problem,which solves the problems of insufficient utilization of sequence information,insufficient priority and lack of timing information in existing studies,and verifies the effectiveness of the improved algorithm through experiments.(3)In response to the problem that user items are full of randomness and chance during the interaction,an improved attention mechanism is added to the network recommendation model based on the conversation-based looping god to learn the changes of the user's conversation sequence during the shopping process To reduce the interference of the recommendation model caused by the user's chance and randomness in the shopping process in the sequence,and further improve the accuracy of the recommendation model.Finally,the validity and superiority of the model were verified through comparison experiments using real data.The results show that the recommended method mentioned in this article is about 40% higher than the existing benchmark in MRR@20 indicator,and the index Recall@20 is about25% higher.