Research On Interpretable Recommendation Based On Network User's Behavior Logs

With the popularity of the Internet and the increasing number of Internet users,network resources have become more and more abundant.Users can log in to the network anytime,anywhere,leave their footprints on the network,and enjoy the convenience brought by the network.It will also be stored by the service provider through the database,which can help the service provider better understand the user.At the same time,users are often unable to quickly locate their needs when facing massive information resources.Fortunately,in recent years,big data and machine learning technologies have developed rapidly,and the software and hardware facilities in the network market have also become more complete.The recommendation system designed to solve the information explosion is very popular,and a number of the information displayed in web pages is more or less related to the recommendation algorithm.Although there are many types of network resources,they are basically the same.If network service providers want to attract users and occupy market resources,they must seize the user's psychology and start with improving service quality.It is not enough to rely on personalized recommendation systems.In general,the recommendation system is only responsible for providing the user with a recommendation list.The success of the recommendation is still closely related to the user.Even if user accepts the recommendation,the recommendation system cannot know whether the user really likes it or why the user likes it.In order to provide highquality recommendation services,the transparency of the recommendation mechanism and the interpretability of the recommendation system are important.An interpretable recommendation system allows users to basically understand the entire recommendation mechanism,as well as why they recommend this item to themselves.The interpretability will shorten the distance between the service provider and the user,so that the user will trust the system more.To improve user stickiness,service providers can also use this personalized recommendation service to optimize the website structure,adjust network information,and then quickly occupy the online market and carry out commercialization actions.The log data of network users is easy to obtain,as long as the user browses the webpage,the records can be left,but this interaction data is generally sparse.In addition,as an implicit feedback,it cannot reflect the user's preference.Previous researchers usually combined auxiliary information can alleviate sparseness and improve the accuracy of recommendations.Auxiliary information can also provide partial interpretation.This article constructs an item knowledge graph according to specific application scenarios.As a special auxiliary information,it contains rich semantic paths.By combining these paths with log data,higher-quality recommendations can be completed.After in-depth research on the existing literature on knowledge graph path recommendation,this paper proposes an interpretable recommendation model.Unlike the previous recommendation mechanism of most scholars,the model has more userlevel crossover features.Previous studies have shown that crossover Features can improve the accuracy of recommendations and provide an explanation of the results.However,in practical applications,it is difficult and expensive to obtain demographic characteristics.On the contrary,log data contains a large number of user behavior characteristics.It is relatively easier to obtain them using data mining methods,and the two can be complementary.Based on this,this article attempts to use the attention mechanism to complete automatic crossover between features,use it to update user representations in the knowledge graph,and finally use long and short-term memory networks that are good at sequence analysis to complete recommendations.Sex has a positive effect.In addition,on the one hand,the sparseness of the interactive data will cause the number of user-to-item semantic paths in the knowledge graph to become scarce and the distribution will be uneven.This article attempts to improve the mechanism of previous scholars to select paths to alleviate the above.The problem.On the other hand,when studying the explanation mechanism of knowledge graph path recommendation,most scholars will present semantic paths with higher contributions to users,but these semantic paths often include parts that are unknown to users,such as unfamiliar user nodes.It will weaken the interpretability of the model.This article attempts to adjust the direction of the relationship in the knowledge graph to eliminate strange users in the path.At the same time,it extracts the combination of features with a large degree of crossover and the path that contributes a lot in the prediction task to explain to the user.Recommended process.In order to prove the effectiveness of the model,this paper conducted experiments and verifications through log data from a movie website.The results show that the accuracy of recommendations has been significantly improved after adding cross features,and it is also better than some existing recommendation algorithms.The explanation mechanism can also show users the entire recommendation process,making the recommendation system transparent.At the same time,multiple sets of comparative experiments are designed to verify the improvements proposed in the model,and the results show that the improvement mechanism indeed improves the model's effect.Finally,I hope that the interpretable recommendation model proposed in this article can help network service providers to complete a more human-based recommendation service,thereby enhancing user stickiness.Network users can also rely on the service provider to rely on this system to achieve a mutually beneficial situation.