Interaction Experience and Perceived Emotional Understanding of Generative AI: An Exploratory Study Using Explainable Machine Learning
Year: 2026
Type: article
Study design: Observational Study
Abstract: As generative artificial intelligence (GenAI) is increasingly used for emotional expression and support, the relationship between interaction experience and users’ evaluations of AI’s emotional understanding remains to be examined. Based on questionnaire responses from 195 participants with prior experience of emotional disclosure to AI, this study applies logistic regression, random forests, and repeated nested cross-validation, together with feature-group ablation, permutation importance, and SHAP, to examine the contributions of different questionnaire features and the effects of semantically similar items on model results.The results show that the random forest model using interaction experience features alone achieved a ROC-AUC of 0.746, compared with 0.478 for background and usage behaviour features, indicating that interaction experience was the more informative source of classification information in the current sample. Perceived mechanical responses (Q18) and response immediacy (Q15) emerged as prominent interaction experience features. Removing Q18 reduced the random forest AUC to 0.672. Further removal of Q15 from the conservative model excluding Q18 and Q20 reduced the AUC from 0.666 to 0.590, indicating that response immediacy retained a notable classification contribution under the conservative feature setting.The findings indicate that, compared with user background and usage behaviour, interaction experience items contain more information relevant to classifying perceived AI emotional understanding in the current sample. Interpreting key features also requires consideration of overlapping questionnaire content and input feature combinations. Through explainable machine learning, this study identifies the main interaction experience features associated with users’ subjective evaluations and provides empirical evidence for identifying important experience features in emotional interactions with GenAI.
Author: Yiming Huang
Institution: University of the West of England
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225376687