Research On Dynamic Self-adaptive Feature Weighting For Personalized Micro-blog Recommendation System

As a kind of broadcast social network platform which shares short real-time information through following mechanism,micro-blog has become an important channel for people to exchange and get information.The micro-blog which be published by friends of user`s attention-list is the main source for user to acquire information,however,with the growing number of micro-blog users and active users,user's attention list has become longer,user maybe face the problem of information overload by these factors.How to dig out the value of micro-blog from a wide range of information flow is the key to improve the quality of service for micro-blog users.This paper studied the existing method of micro-blog recommendation and made full use of them,then combed with the practical application scenarios,this paper transformed the problem of personalized micro-blog recommendation into the reordering of the micro-blog received by user.First of all,the recommended range is determined by session segmentation,then this paper constructed a lot of micro-blog features based on user preferences,micro-blog content and publisher authority,and then put forward a multi-feature fusion method based on dynamic self-adaptive feature weighting(dynamic self-adaptive feature weighting,DAFW),finally this paper designed a real-time personalized recommendation system of micro-blog after validating the validity of this scheme.The main results of this paper are as follows:(1)This paper established a series of scheme which processing micro-blog corpus,then trained the general theme model based on the processed micro-blog,and then used method of Gibbs Sampling to estimate the theme distribution of the target text from general theme model,this method can solve the difficult of topic modeling for short text.(2)This paper excavated the publisher who forwarded by but not followed by user,the micro-blog of publisher will be as a supplement to the recommended content in case of insufficient,so users can also receive interest micro-blog through other resources.(3)This paper constructed ten features of micro-blog based on user preferences,micro-blog content and publisher authority,including many analysis features,such as micro-blog heat,interactive TF-IDF,etc.,adding these features could significantly improve the quality of recommendation.(4)In order to solve the multi-index fusion problem,this paper introduced the information entropy into multi-factor weighting,to calculate the objective weights according to the variance of each features value,and combined features mean and parameters to get the sorting function,experimental results show the effectiveness of this method.(5)This paper studied and designed a real-time recommendation system,to apply the theoretical research of this paper to real life which provides value.The work of this paper can effectively deal with the information overload problem in the social network era,users can spend less time to capture more characteristic information in the circle of friends and get the recommended results in real time,and our work has better theoretical and practical significance for improving the user experience.