Dear all, Dr. Joanna Michalak (Nicolaus Copernicus University in Torun, Poland) will give a seminar entitled "Integrated Approach to Sentiment Analysis: Optimizing Short Text Analysis Efficiency in the Context of Economics and Finance" on November 14th at 09:30 am in Room 5 at Macro Teaching Hub Engineering Area., Via del Politecnico, 1, University of Roma "Tor Vergata", Roma. The seminar can also be followed online by sending an email to roberto.monte@uniroma2.eu

Abstract:

Big social data (BSD) is widely used in the social sciences, particularly in economics and finance, where it is valued both for its explanatory insights and its predictive power. Predictive

analyses leveraging BSD often focus on modeling the relationship between social media activity and stock market movements. The quality of these analyses relies on two main

dimensions: (1) models that capture short-term dynamics and account for observed time-series properties, and (2) sentiment analysis methods that are specifically adapted to the

unique vocabulary of finance and economics.

This presentation addresses these two dimensions. In the first part, we explore a case study of Apple Inc., comparing the effectiveness of VAR, GARCH, and LSTM models for describing

short-term interdependencies between social media sentiment and stock market behavior. The second part of the presentation focuses on the performance of various sentiment

analysis methods selected to construct sentiment-based indices. This section aims to compare the efficiency and accuracy of different sentiment analysis methods when applied to

short text data.

The study emphasizes optimization of analysis efficiency by evaluating widely used datasets, such as Senti140. Several research questions guide this investigation:

1. Which sentiment analysis method is most effective in terms of accuracy and processing speed in financial data analysis?

2. Do advanced methods, such as deep learning models (e.g., BERT), outperform traditional lexicon-based techniques and classical machine learning algorithms?

3. What are the differences in sentiment analysis accuracy between general datasets like Senti140 and finance-specific datasets such as FinBERT?

The issue of optimizing text data analysis methods remains an open aspect of this research. The ultimate objective of this study is to construct a dataset or lexicon specifically designed

and recommended for financial data analysis.