The Need for an Interdisciplinary Approach: Analyzing Computational Methods in Attitude Detection

The purpose of this talk is to provide a critical analysis of the use of datasets in attitude detection (the equivalent of attitude analysis in social science), focusing on the contrast between computational and psycho-socio-cultural approaches. Although attitude detection has become a prominent field in computer science due to its capacity to examine vast quantities of opinions through machine learning techniques, such methods frequently neglect the intricate socio-cultural contexts that inform human communication. Computational approaches rely heavily on predefined data sets, which tend to simplify complex human interactions into binary categories that may not fully capture the nuances of individual opinions and cultural influences. In contrast, sociological perspectives underscore the necessity for a more comprehensive understanding of opinions as socially and culturally embedded phenomena. In addition, the use of large language models (LLMs) in attitude detection opens up new possibilities due to their ability to analyze complex language patterns, but their effectiveness still depends on the quality of training data, which often overlooks key socio-cultural aspects that influence expression. The issue also calls for sociological research to better understand how socio-cultural context affects the expression of attitudes and how language models can account for these complex interactions, rather than simplifying them into mathematical formulas. This presentation puts forth the proposition that the incorporation of socio-cultural contexts into attitude detection research would not only serve to enhance the precision of these models but also facilitate a more comprehensive understanding of human behavior. Through a comparative analysis of attitude detection techniques and their applications, the paper identifies the shortcomings of current computational approaches and advocates for an interdisciplinary framework that bridges the gap between computational efficiency and cultural complexity.