Research On User Interest And Sentiment Modeling Based On Short-text Comprehension

With the development of the network and the popularity of intelligent devices,more and more people express their views and opinions in the network.So there is a large amount of user generated information(User Generated Content,UGC).How to mine short text information in UGC has important research significance.The existing researches on the user's interest and sentiment analysis are almost separate.In fact,the user interests and emotions are related.This thesis engages in interest modeling and sentiment analysis for users.Meanwhile,our purpose is to find the association between user interest and sentiment,and to construct the user interest emotional model.The main achievement of this thesis included:1)Due to the shortcoming of the traditional methods in short text processing,we use the Twitter-LDA model to analyze the user interest.And considering volatility and time of the user interest,we add time factor to interest model and refine it.Then we construct the user interest knowledge tree by using word vector.In this thesis,a new methodology(V-PSD)for similar interest calculation is proposed based the Point-wise Similarity Distance model and word vector.Our goal is to achieve the mapping between the user's interest and the knowledge tree.2)In this thesis,considering the impact of user interest on emotions,we put forward a new methodology(T-CLSTM)for short text sentiment analysis based on the topic context and long-short term memory(LSTM)model,and reconstruct the user interest emotion model with the interest factor.3)In order to better describe the user emotion,this t introduces the concept of psychological(PAD)in the analysis of text sentiment.Then we put forward the process of generating the user's personality based on PAD model and achieve quantitative analysis of sentiment.Through the experimental analysis of the methods mentioned in the research,this thesis validates the role of interest factors in the promotion of sentiment classification and the validity of PAD model for quantitative analysis of sentiment.