Hibridni pristop za zaznavo elementov subjektivnosti v besedilnih tokovih : doktorska disertacija

slovenian

doktorska dela

2009

Univerza v Mariboru, Fakulteta za elektrotehniko računalništvo in informatiko

Sentiment analysis or opinion mining is relatively new branch of text mining and is very popular lately especially because of the potential for mining public opinion in different areas of interest. In comparison to Information retrieval and Computational Linguistics, opinion mining is not focused on topics or objective data (location, time, people) but on subjective opinion of the document writer. Sentiment analysis is especially interesting for analyzing text streams, which are texts with special characteristics available on the web. There are many possibilities of applying sentiment analysis, for example, it can be used to track opinions on some special product or service, tracking public opinion on political candidates during election time or on some political situations. Opinion mining is not only about determining opinions, but also to extract features that can be used to determine, whether opinion was positive (good) or negative (bad). This thesis describes we researched semi-automated and automated detection of subjective elements - terms bearing opinion or sentiment and which served as abasis for lexicons to be later used in classification. New hybrid method of sentiment classification has been proposed and implemented in our tool SentimentHitchhicker. Thesis combines ideas and methods from different fields, including machine learning, information retrieval, computer linguistics and psychology.

07.08.2023