Distance Metric Based Semi-supervised Learning For Text Sentiment Analysis

Sentiment analysis is a popular research field of artificial intelligence(AI).With the development of the Internet era,information in the text form is experiencing explosive growth and has become a resource that people can easily obtain and communicate.Since people want to understand specific internal relationships and meanings of the collected text data,text sentiment analysis is widely applied.It is one of the hottest issues in the sentiment analysis field on how to use limited labeled samples to improve the classification precision of text sentiments.This thesis mainly studies the graph-based semi-supervised algorithm for sentiment analysis.Taking the impact of text vector similarity on sentiment judgment into consideration and existing research results as references,this thesis integrates distance metric learning into the similarity calculation process and reinforces semantic relations between key features based on the graph-based semi-supervised algorithm.Therefore,a semi-supervised method is proposed for sentiment analysis based on metric learning.This thesis aims at the application of distance metric learning in semi-supervised text for sentiment analysis and summarizes all related achievements in this field on the basis of literatures.It introduces some commonly used semi-supervised machine learning algorithms,describes detailed process of text sentiment analysis,analyzes similarity calculation,and finally proposes a series of improvement schemes according to the features of text sentiment analysis and metric learning algorithm.This thesis mainly includes the following:(1)Consider the impact of different structural maps on experiment results on the basis of semi-supervised algorithms,propose improvement schemes,and construct proximity maps through the algorithm that uses thresholds to eliminate relevance of cosine distances.(2)Use the semantic similarity to find the Mahalanobis distance between text vectors,which is more suitable for text sentiment calculation.Finally,based on the theory mentioned before,we realized the entire text sentiment analysis system,including classification module,pre-processing module,feature selection module and the distance metric learning modules.