Interst Modeling Based On Short Text Analysising

With the development of mobile internet,the internet has been integrated into every aspects of people's work and life,becomes an important way for people to obtain and disseminate information.However,the explosion of information attributed to the rapid development of information technology makes it more difficult for people to obtain and identify useful information in the vast sea of information.Therefore,personalized recommender service is particularly significant for providing better internet service,and user profile generation has become one of the hottest research topics at the right moment.The work of this thesis mainly focuses on interest modeling based on short text analysis.The main work is as follows:Firstly,we researched on short text analysis and solved basic problems on short text processing by improving segmentation algorithm,new words discovery algorithm and keyword extraction algorithm for further analysis.Secondly,we proposed a new method to train sentence vector based on word representations in vector space for short text categorization after analyzing the pros and cons of traditional classification algorithms.Finally,we proposed an interest modeling method based on keywords extraction and expansion which also combines the word embedding.The topic model and user communities of interest were used to improve the scalability and stability and achieved good results.