References & Citations
Computer Science > Computation and Language
Title: Learning affective meanings that derives the social behavior using Bidirectional Encoder Representations from Transformers
(Submitted on 31 Jan 2022 (v1), last revised 13 Sep 2022 (this version, v2))
Abstract: Predicting the outcome of a process requires modeling the system dynamic and observing the states. In the context of social behaviors, sentiments characterize the states of the system. Affect Control Theory (ACT) uses sentiments to manifest potential interaction. ACT is a generative theory of culture and behavior based on a three-dimensional sentiment lexicon. Traditionally, the sentiments are quantified using survey data which is fed into a regression model to explain social behavior. The lexicons used in the survey are limited due to prohibitive cost. This paper uses a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model to develop a replacement for these surveys. This model achieves state-of-the-art accuracy in estimating affective meanings, expanding the affective lexicon, and allowing more behaviors to be explained.
Submission history
From: Moeen Mostafavi [view email][v1] Mon, 31 Jan 2022 19:58:28 GMT (2053kb,D)
[v2] Tue, 13 Sep 2022 19:26:36 GMT (619kb)
Link back to: arXiv, form interface, contact.