Eureka! Institutional Repository

With the growing use of the World Wide Web, user-generated content (UGC) isinductively increased, which has been proven to have a significant impact on people'severyday lives, that is why making its effective analysis vital, with a keen interest in thescientific, industrial and even political community. In cases where the content to beanalyzed in terms of its opinion is in the form of text, the concept of Sentiment Analysis ofthe Natural Language Processing (NLP) field is introduced, which aims at the automaticrecognition of subjective information from written sources. In this diploma thesis, SentimentAnalysis is described as a challenging problem and its solution is investigated throughMachine Learning methods, applied to product reviews. Therefore, the problem is treated asa problem of classifying product reviews texts in an emotional class, focusing on theproduct attributes for which the reviewer is expressed. Approach to such problems is thefield that we examine and it is usually implemented at the sentence or aspect level, and theanalysis of this level is referred to as an Aspect Based Sentiment Analysis (ABSA) aimed atboth identifying the aspects (product attributes) in the sentence, as well as the sentimentthey carry. We identify the difficulties of solving such problems and possible MachineLearning algorithms as solutions, as the comprehensive problem solving process isdescribed step by step. Finally, we present a large number of related work that attempted toresolve this problem by quoting a subjective comparison with important observations. Oneof the remarks and the final conclusion of this diploma is that none implemented ABSAsystem has succeeded in solving the problem altogether, a fact that discovers the adversityof managing the natural language