The development of semantic sentiment analyser utilising sentiment composition for financial news

Tan, Li Im (2016) The development of semantic sentiment analyser utilising sentiment composition for financial news. Masters thesis, Universiti Malaysia Sabah.

[img]
Preview
Text
The development of semantic.pdf

Download (11MB) | Preview

Abstract

Sentiment analysis is a technique to determine and extract subjective information from source materials. This thesis studies the effectiveness of a lexicon-based sentiment analysis that used sentiment composition rules and semantic similarity techniques to perform polarity classification for financial news articles. This method utilized a prior polarity lexicon to determine the polarity of the analysed text. The semantic sentiment analyser is developed to assist investors in their stock investment by providing them the news sentiment as a source of references in their investment decision. This work compares and combines a few existing sentiment analysis methods to determine the positive and negative classification of the news articles. There is set of 893 financial news articles were collected for experiment purposes from early of year 2013 until June 2013. The research project started off with the development of the Baseline Sentiment Analyser based on existing sentiment composition rules and a mathematical formula namely Positivity/Negativity ratio to determine the sentiment value of the analysed text. This sentiment value is used to determine the polarity of the financial news article. In this model, a phrase extraction tool is needed for phrase extraction according to the Part-of-Speech of the text. Various data mining methods such as stemming and lemmatization algorithms were used to produce different representations of data. These sets of data are combined with the different phrase extraction tools to work out the best combination for the lexicon matching task. Next, an Enhanced Sentiment Analyser with a new set of sentiment composition rules is proposed. This set of sentiment composition rules made used of the verb-phrase sentiment composition, the verb-noun phrase sentiment composition, the noun-verb phrase sentiment composition, the conjunction ""but"" sentiment composition, and the negation rule which include more polarity shifters. Finally, this sentiment analyser is further improved and into a Semantic Sentiment Analyser. Three metrics (HSO, LESK, and LIN) were used to find the semantic similarity between input word and matched words as well as to perform polarity tagging and their performances were compared. WordNet was used as the lexical resources in determining the relationship between two words in this task. The best metric found in this task which is HSO was applied to the proposed Semantic Sentiment Analyser to calculate the semantic similarity between words and to perform polarity tagging to the matched pair that yielded the highest semantic similarity value. This task optimized the word with polarity every time a new financial news article is analysed. While analyzing the financial news article, the prior polarity lexicon is expanded as well. The performance of the proposed Semantic Sentiment Analyser was evaluated and showed promising results in classifying positive and negative news.

Item Type: Thesis (Masters)
Uncontrolled Keywords: sentiment analysis, lexicon, semantic similarity techniques, polarity classification, financial news articles
Subjects: Q Science > QA Mathematics > QA76 Computer software
Divisions: FACULTY > Faculty of Computing and Informatics
Depositing User: Munira
Date Deposited: 27 Oct 2017 08:11
Last Modified: 27 Oct 2017 08:11
URI: http://eprints.ums.edu.my/id/eprint/11909

Actions (login required)

View Item View Item