Advanced Sentiment and Emotion Analysis

Instructor: Barbara Konat (Adam Mickiewicz University)

Workshop pre-requisites:

• Basic knowledge of Python (writing functions, understanding of data structures, modules, dependencies).
• Familiarity with Google Colaboratory. For those preferring local installation of Jupyter Notebook or another IDE, ensure setup prior to the workshop. Please note that support for installation issues during the workshop will not be possible.

Thursday 29 February 2024, 14:00 – 16:00 (CET) (time TBC)
14:00 – 16:00
Workshop part 1: Introduction to Sentiment Analysis

• Overview of sentiment analysis in NLP.
• Exploration of psychological lexicons and lexical methods.
• Hands-on session: Introduction to basic Transformers using the Pipeline object.
• Hands-on session: Brief overview of building neural networks for NLP.

Friday 1 March 2024, 9:30 – 13:30 (CET)
9:30 – 11:00
Workshop part 2: Advanced Techniques in Sentiment Analysis

• Introduction to fine-tuning and transfer learning in NLP.
• Detailed discussion on models and tokenizers in Transformers.
• Hands-on session: Fine-tuning a Transformer model with a built-in dataset.
• Hands-on session: Fine-tuning a Transformer model with a custom dataset.

12:15 – 13:30
CLES talk: Emotion in Reason. Insights from Argumentation, Psychology, and AI

The presentation will share key findings from the „Computational Pathos” project, spanning two years of interdisciplinary research. Building on the foundations laid by classical theorists and extending through modern developments in psychology and computational linguistics, we look at the complexity and efficacy of emotional appeals.

The dialogue commences with a survey of argumentation theory, revisiting three Aristotelian modes of persuasion and drawing from the work of scholars like Olbrechts-Tyteca, Walton, and van Eemeren, to contextualise our understanding of argument, persuasion, and manipulation.  We propose our own model and operationalisation of pathos in speaker – audience interaction.

Progressing into the psychological domain, we look into the theoretical grounding for the role of emotion in argumentation. On the background of models such as Ekman’s and Plutchik’s categorisations of emotions and the Elaboration Likelihood Model, we present the results of our own three psycholinguistic experiments with positively and negatively framed arguments.

The final segment delves into computational linguistics and AI, presenting our work employing the annotation of Waltonian argumentation schemes and using sentiment analysis to understand emotional appeals in political debates and social media reactions. We also present initial results of using Large Language Models (such as GPT) to compare human and automatic annotation of pathos appeals.

In this lecture, by incorporating empirical evidence and leveraging computational tools, we aim to extend the dialogue on emotions in argumentation theory, contributing a novel interactional model of pathos that reflects the dynamism of modern rhetoric.

Publication
Konat, B., Gajewska, E., Rossa, W., „Pathos in natural language argumentation: emotional appeals and reactions”, Argumentation (to appear)

Speaker bio

Dr. Barbara Konat is an assistant professor at the Faculty of Psychology and Cognitive Science at Adam Mickiewicz University in Poznań. In her work, she bridges cognitive science with computational linguistics, exploring the emotional and dialogical aspects of argumentation. Her research on the relationship between emotions and argumentation benefits from her postdoctoral experience at the Centre for Argument Technology at the University of Dundee, as well as her prior role as the head of a research and development project at the startup „Sentimenti.pl”, which specializes in sentiment and emotion analysis. Currently, she leads the „ComPathos” project funded by the National Science Centre (NCN), aimed at developing a novel computational model of pathos in rhetoric. She collaborates with teams of psychologists, linguists, and information technology experts. Her publications, which appear in journals such as „Informal Logic”, „Metaphor and the Social World”, and „Behavior Research Methods”, focus on creating linguistic models of emotion and applying language analysis in the context of emotions and argumentation. More information can be found at bkonat.home.amu.edu.pl.