Design And Implementation Of Users' Intention Recognition System Based On Weibo's Hashtags Searching

Sina Weibo is a wildly used platform in Chinese social society,many people use Sina Weibo to search their interested blog,but sometimes people would find that the intention of the search result is not what they want.In modern Internet,the search engine takes a lot of time which people spending on Internet,so it's very important to return a better searching result to users.This paper is aiming to design and implement a system that can boost the accuracy of recognizing the intention of weibo's searching query.Recognizing intention of weibo's searching query is a typical short text classification problem,the insufficient information of short text and ambiguous problem make it very hard to solve.The system which this paper proposed uses the Python language and uses the idea of modules designing,and the system is focusing the problem of insufficient information and ambiguous problem of short text,and designs a method of query expansion,and uses topic modeling technique.This system uses the state-of-the-art classification model and mainly include five modules which are text preprocessing module,query expansion module,topic model generation module,personized representation generation module and users' intention recognition module.The author designs and implements these five modules independently.Text preprocessing module which contains data loading,text cleaning,text segmenting,word's vector training and saving.Topic model generation module which contains LDA model training,LDA model comparison and saving and topics allocation of users' query log generating.Query expansion module which contains tables of locally sensitive hashing generating,expanding candidates generating and filtering.Personalized representation of query generation module which contains LDA's topics vectorization,compressing method or expanded method to generate personalized representation.Users' intention recognition module which contains neural network classification model's training and user query intention recognition.The system's verification results show that the system designed in this paper raises the accuracy rate of user intention recognition from the initial 74% to 85.4%,which achieves the expected effect of the system design.