Automatically Building Financial Sentiment Lexicons While Accounting for Negation

Abstract

Financial investors make trades based on available information. Previous research has proved that microblogs are a useful source for supporting stock market decisions. However, the financial domain lacks specific sentiment lexicons that could be utilized to extract the sentiment from these microblogs. In this research, we investigate automatic approaches that can be used to build financial sentiment lexicons. We introduce weighted versions of the Pointwise Mutual Information approaches to build sentiment lexicons automatically. Furthermore, existing sentiment lexicons often neglect negation while building the sentiment lexicons. In this research, we also propose two methods (Negated Word and Flip Sentiment) to extend the sentiment building approaches to take into account negation when constructing a sentiment lexicon. We build the financial sentiment lexicons by leveraging 200,000 messages from StockTwits. We evaluate the constructed financial sentiment lexicons in two different sentiment classification tasks (unsupervised and supervised). In addition, the created financial sentiment lexicons are compared with each other and with other existing sentiment lexicons. The best performing financial sentiment lexicon is built by combining our Weighted Normalized Pointwise Mutual Information approach with the Negated Word approach. It outperforms all the other sentiment lexicons in the two sentiment classification tasks. In the unsupervised sentiment classification task, it has, on average, a balanced accuracy of 69.4%, and in the supervised setting, a balanced accuracy of 75.1%. Moreover, the various sentiment classification tasks confirm that the sentiment lexicons could be improved by taking into account negation while building the sentiment lexicons. The improvement could be made by using one of the proposed methods to incorporate negation in the sentiment lexicon construction process.

Negation handling for sentiment analysis task: approaches and performance analysis

“…Experimental results in research [21] show that the performance of the FWL method with a window size of 2 is better when compared to the RoS, FSW, and NNA approaches. Other research [30] uses this approach to build a Sentiment Lexicon by considering negation; experimental results show that considering negation in building a sentiment lexicon can improve the performance of the F-measure and other performance metrics. This research alludes to implicit negation cues but is misinterpreted where implicit negation is considered morphological negation.…”

“…This determination of the negation scope refers to research [19]. Another study [18] also carried out negation handling using the switch negation (SN) approach and modified the sentiment score to a broader scale (-5 to +5), not just on a value scale of 1 and -1. b) Flip sentiment (FS): in another study, Bos and Frasincar [30] proposed an approach that considers negated words to have a sentiment orientation opposite to their sentiment label. Thus, words are negated in sentences with positive sentiment classes as negative words and vice versa.…”

A Knowledge-Based Model for Polarity Shifters

“…A large majority of these studies focus only on negation, and for English, with the intention of specifically targeting its automatic detection and scope so that it can be processed in computational systems (Morante and Daelemans 2009;Wiegand et al 2010;Hogenboom et al 2011;Lapponi, Read and Øvrelid 2012;Asmi and Ishaya 2012;Reitan et al 2015;Sharif et al 2016;Diamantini, Mircoli and Potena 2016;Pandey, Sagnika and Mishra 2018;Mukherjee et al 2021;Morante and Blanco 2021;Bos and Frasincar 2021;Singh and Paul 2021;Barnes, Velldal and Øvrelid 2021). Faced with this fact, several authors have pointed out the convenience of incorporating, in addition to negation, more types of polarity shifters in computational models.…”

“…However, Zhang et al (2011) decided to build a corpus for contextual shifters manually, given the lack of accuracy and presence of noise in the automatic generation models and the need to better understand the problem of contextual polarity shifters. Given the resource requirements and cost associated with the manual creation of training corpora for shifters mentioned above by Ayeste and Noferesti (2022), it is noticeable that more recent studies focus on achieving their automatic construction (Xu et al 2020;Schulder, Wiegand and Ruppenhofer 2021;Bos and Frasincar 2021) or rely on more complex deep learning techniques (Singh and Paul 2021). These studies are based on English, being very scarce for Spanish (Jiménez-Zafra et al 2020;Pabón et al 2022).…”

UTSA: Urdu Text Sentiment Analysis Using Deep Learning Methods

“…However only using unigrams is not sufficient for SA such as a same word may has opposite orientation in different circumstances. Main disadvantage of this approach is unavailability of domain specific lexicons [6]. The solution of dictionary-based is corpus-based approach in which list of seed words (adjectives) is expanded by using the corpus of same domain documents.…”