Sentiment analysis based on a social media customised dictionary

Abstract

This article presents a methodology to classify the polarity of words from selected Tweets. Usually, social media sentiment (SMS) is lexically determined, manually or by machine learning. However, these methods are either slow or based on a pre-established dictionary, thus not providing a customised analysis. We propose a methodology that, after having mined the topic-related Tweets, filters relevant words based on the mean and standard deviation frequency in positive and negative market days to remove neutral terms. Subsequently, through an ad hoc perceptual mapping, we assign a polarity to the dataset. This method allows the building of a dictionary associated with the investor sentiment customised to that organisation. A practical application was carried out to test the proposed methodology. The results were significant and in line with the behavioural finance theory, confirming that irrational investor feelings—expressed via social media—drive a portion of asset prices. Results also confirm the investor asymmetric behaviour under gain or loss scenarios, with the latter generating more impact than the former because people are risk-averse. The proposed method is expected to identify patterns of behaviour in social media linked to market oscillations, thereby contributing to risk management and optimising decision-making in the stock market. The use of both statistical and perceptual map filters allows a specific asset dictionary to be built; Textual sentiment analysis based on social media; The proposed method efficiently overcomes generic dictionaries and language issues.

Revisiting the Accounting Fraud Components: A Bottom-Up Approach Using the Twitter Platform

“…Therefore, we use a quantitative deductive method (Almeida et al, 2021) to answer our research question differently from previous, exclusively qualitative literature. For our analysis, we used the R Statistical Software.…”

Equivalent sentiment measures for cross-language analysis of corporate communications

“…We follow their steps by using a parallel corpus of English-French reports published by Canadian companies. Third, it is possible to build a sentiment lexicon from the ground up, for example, by finding which words appear more often when companies record positive vs. negative returns in the stock exchange [ 18 ]. In that case, however, cross-language studies are not facilitated.…”

Analysis of public opinion on food safety in Greater China with big data and machine learning

“…Lexicon-based sentiment classification (Taboada et al, 2011) is a classification method based on linguistics that uses a series of dictionaries of pre-marked words, including a sentiment dictionary, degree dictionary, inversion dictionary, and stop word dictionary. A sentiment dictionary is a collection of sentiment words, and its domain specificity is critical to the accuracy of the model (Almeida et al, 2021). We integrated specific food safety vocabulary and authoritative sentiment dictionaries.…”