Sentiment Analysis Based On Ensemble Methods

Online social media provides a fast and convenient communication environment through which people can express their views and opinions on social media hot spots.Emotional analysis is a technique to analyze people’s emotions from their views and opinions,has become a hot issue in the current Natural Language Processing,has important application prospect in emotional forecasting field.Aiming at the weak generalization ability of sentiment analysis in single classification model,the affective analysis of ensemble method is presented,which has good generalization ability and good adaptability in sentiment analysis.The main work of this thesis focuses on the following two aspects.1.This thesis analyzes the popular affective lexicon,constructs the affective lexicon used in affective analysis,and divides the lexical library according to the affective tendency.According to the characteristics of micro-blog short text data,the corpus of the data pre-processing was screened out,and the corpus was represented by vector based on the selected features.The existing models of sentiment analysis and the advantages and disadvantages of the commonly used single classification algorithm model are investigated,and the single classification algorithm model is experimentally verified.2.A heterogeneous ensemble method is proposed,which solves the problems such as poor generalization performance caused by mis-selection in the training set of hypothesis space in the single classification algorithm model.The basic idea is,first of all,using the built Sentiment dictionary for micro-Bo short text features to express and select.Then,according to the characteristics of the single classification model of different models according to the strategy of integration,including two ways,one is two or more than two kinds of single classification model and experimental results using the method of confidence degree weighted average combination of sentiment analysis model,choose a high degree of confidence sub classifiers classification decision;another is the classification results of classifier combination method and confidence as the feature of a higher level integrated classifier into sentiment analysis model.Finally,the design of public sentiment analysis platform.Experimental results show that the proposed ensemble method is effective and feasible for sentiment analysis.