The Kernel Research Of Beijing Housing Price Prediction Based On Weibo Information

Nowadays,technology of Machine Learning and Natural Language Processing developed rapidly.homes' prices attract people's attention.Due to the difficulties of acquiring information,most current researches were often based on statistics and financial data,but did not consider human's attitude.In this thesis,I try to dump mass personal comments and replies on Sina Wei Bo,and build a more precise model which mixes subjective attitude to form a forecast system for prices of real estates.My thesis includes:(1)By artificial collection with the housing information of Internet,I determine the topic collections of housing field which oriented to the subjective attitude of housing field,and make the topic filter by the influence degree between topic and housing price.Finally I realized the extract of subjective attitude.(2)Aim to the problem of incorrect recognition with topic word in word segmentation algorithm,I use the similarity which between the vocabulary and the keyword of the topics in housing field to be the relevance between the vocabulary and the topics,Then I use this relevance to improve the Boundary Entropy algorithm,and make the Model integration between this improved Boundary Entropy algorithm and Conditional Random Field algorithm.Finally I realized a Wei Bo's word segmentation basic on housing filed.By comparing with the other word segmentation algorithm,the test results show that the method is quickly and accurate.(3)Aim to the problem of incorrect split with topic in the topic model algorithm,I add a maximum entropy priori layer in front of Probabilistic Latent Semantic Analysis(PLSA)Model,and I use the Jaccard coefficient to calculate the word similarity matrix,raising an Topic Modle algorithm for housing filed.By comparing with the other word segmentation algorithm,the test results show that the method is accurate and efficient.(4)According to the Weibo content of uncertainty and complex,I design a more suitable method for model selection and model parameter design,which is much better for Weibo's prediction model,and using object-oriented design idea,designing the general framework of the system,each function module and the related class design,the realize prediction system based on B/S architecture and JSP technology.Finally,by comparing the situation of public sentiment in Beijing in recent years and the change of the real estate price trend,this paper compares the forecasting system with other house price trends prediction system.The test results show that the method is efficient and comprehensive,and can be used to predict the price of the house based on house price trends by subjective factors.And it has high prediction accuracy.