Beyond Text: Leveraging Multi-Task Learning and Cognitive Appraisal Theory for Post-Purchase Intention Analysis

TL;DR

The paper tackles the challenge of predicting post-purchase intentions from text by embedding cognitive appraisal Theory-informed psychology into multi-task learning. It leverages the PEACE-Reviews dataset, which provides first-person emotions, 20 appraisal dimensions, and PCB ratings, to compare a spectrum of models from text-only baselines to theory-informed multi-task and multi-modal architectures. Results show that incorporating appraisal and emotion signals improves PCB prediction, with theory-informed and multi-task configurations offering targeted gains and explainability via Integrated Gradients. This work demonstrates the value of cognitive-emotional constructs in NLP-based consumer behavior modelling and points to practical implications for marketing, product design, and future large-language-model–driven computational psychology applications.

Abstract

Supervised machine-learning models for predicting user behavior offer a challenging classification problem with lower average prediction performance scores than other text classification tasks. This study evaluates multi-task learning frameworks grounded in Cognitive Appraisal Theory to predict user behavior as a function of users' self-expression and psychological attributes. Our experiments show that users' language and traits improve predictions above and beyond models predicting only from text. Our findings highlight the importance of integrating psychological constructs into NLP to enhance the understanding and prediction of user actions. We close with a discussion of the implications for future applications of large language models for computational psychology.

Figures (2)

• Figure 1: Models implemented in our study. Model (12) is the theoretical model.
• Figure 2: Word attribution of two samples that scored high and low in PCBs based on the baseline text -> PCB model, respectively.