Financial markets sentiment analysis: developing a specialized Lexicon

  • PDF / 1,865,222 Bytes
  • 20 Pages / 439.642 x 666.49 pts Page_size
  • 24 Downloads / 246 Views

DOWNLOAD

REPORT


Financial markets sentiment analysis: developing a specialized Lexicon Mehdi Yekrangi1 · Neda Abdolvand2 Received: 29 February 2020 / Revised: 29 August 2020 / Accepted: 3 November 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract Natural language processing in specific domains such as financial markets requires the knowledge of domain ontology. Therefore, developing a domain-specific lexicon to improve financial context sentiment analysis is noteworthy. In this paper, by exploring a wide related corpus along with using lexical resources, a hybrid approach is proposed to build a lexicon specialized for financial markets sentiment analysis. The lexicon is applied on a large dataset gathered from Twitter during nine months. Experimental results demonstrate a significant correlation between extracted sentiments from the corpus and market trends which indicates lexicon’s superior efficiency in measuring market sentiment compared with general-purpose dictionaries. Keywords Natural language processing · Financial market · Sentiment analysis · Text mining · Lexicon

1 Introduction Today, a large amount of textual data is unstructured which cannot be easily processed. In order to extract information from such data, text mining methods can be used. Text mining refers to the area at the intersection of data mining, natural language processing, information retrieval, and machine learning contexts (Mittermayer and Knolmayer 2007). It may be applied on the texts containing different attitudes towards various topics. Extracting useful knowledge about the sentiments behind such texts is called sentiment analysis. Sentiment analysis refers to the systematic investigations (Wilson et al. 2005b) of people’s opinions, attitudes, sentiments, and emotions towards entities such as services, products, companies,  Neda Abdolvand

[email protected] Mehdi Yekrangi [email protected] 1

Department of Computer Science and Information Systems, University of Limerick, Limerick, Ireland

2

Faculty of Social and Economic Sciences, Alzahra University, Tehran, Iran

Journal of Intelligent Information Systems

individuals, events, etc., and their different aspects (Liu 2012). Two main approaches for systematic investigation of the texts about attitudes are based on lexicon or machine learning methods (Taboada et al. 2011). While lexicon based methods consider predefined words and phrases in the text and the orientations behind them Turney (2002), machine learning methods are based on the classifiers built from labeled instances of texts or sentences (Bo et al. 2002). Sentiment analysis is applicable in many different domains, including financial markets. Generally, there are two predictive measures in the financial markets which are technical or fundamental. While the former is based on historical trading data (Investopedia 2019b), the latter deals with various kinds of information about a country or company. Fundamental analysis is based on macroeconomic factors (Investopedia 2019a) including structu