Research On Recommendation Algorithm Based On User Behavior Feedback

With the continuous development of computer technology,massive amounts of information are continuously updated and increased every day.In the face of the complicated data on the Internet,it is often difficult to find the information resources that they really need,and the recommendation algorithm is proposed.The recommendation algorithm can predict the user's preference according to the user's past behavior and score,generate a unique recommendation list for the current user,and actively guide the user to discover the information.Most of the current recommendation algorithms use explicit feedback from users(such as scoring),but these explicit feedback collections are difficult and the authenticity cannot be guaranteed.In fact,exploring user interest in an item can also rely on a large amount of implicit feedback in the presence of a web log.As far as e-commerce platforms are concerned,users must generate a lot of intermediate data from search to order--click,browse,follow,collect,purchase,etc.These behavior data are collected without being perceived by the user,and can be truly and reliably Reflecting user interest,and these behavior data can be updated in real time,so the recommendation algorithm based on user behavior has important research value.This paper has carried out sufficient research on the development status,common data sets and evaluation indicators of the recommended algorithm at home and abroad,and analyzed the shortcomings of the existing research results,and carried out the following work:(1)To use behavior data for recommendation,we must first establish the relationship between user behavior and user interest,and quantify user behavior.This requires in-depth understanding and analysis of user behavior,and accurately explore the implicit relationship between behavior and interest.The general algorithm is to artificially assign scores to user behavior.This method is very rough and poorly interpretable,resulting in the behavior model not accurately indicating user interest.Therefore,this paper proposes a user behavior model that can accurately measure user interest.The model fully analyzes the motivation of behavior generation,and uses the subjective G1 method and the objective entropy method to empower behaviors,and combines the influence of time factors on the model,which improves the accuracy of the model and provides information on how to use behavioral data.A new solution.The experimental results show that the algorithm can effectively improve the update speed of the model and further improve the accuracy of the recommendation results.(2)The recommendation algorithm based on user behavior is generally updated offline by the platform or dataset.However,in real life,the user interest changes every moment,so the behavior model embodying user interest should also be constantly Change,which puts higher demands on the time complexity of the algorithm.Changes in intrinsic interest affect the expression of external behavior,so the feedback collected by the system is also updated in real time,so we will use these continuous feedback correction models to use matrix decomposition to achieve dynamic update of the model,and on this basis A recommendation algorithm based on user behavior feedback is proposed.The experimental results show that the algorithm effectively improves the update speed of the model and the accuracy of the recommendation results is also improved.