Title: A method for evaluating the credibility of online public opinion based on ERNIE and SHAP

Authors: Xijing Zhang; Zhongjie Shen

Addresses: College of Communication, Xijing University, Xi'an, 710123, China; Faculty of Arts, Communication and Education, Kuala Lumpur University of Science and Technology, Kuala Lumpur, 43000, Malaysia ' College of Communication, Xijing University, Xi'an, 710123, China

Abstract: The current assessment of the credibility of online public opinions has low accuracy and interpretability. To address this issue, the research proposes a new model for online public opinion assessment. The research adopts text segmentation strategies and model compression techniques to enhance the model's ability to handle long texts. Monte Carlo sampling is introduced to replace Shapley value calculation, and feature grouping and embedding techniques are combined to analyse the complex interaction relationships among features. Experimental results show that the classification accuracy of this model is 97.39%. The average response time is 27.98 milliseconds, and the CPU occupancy rate is 29.78%. In practical applications, the F1 value and the area under the curve of the new model are 0.91 and 0.93 respectively. The model can maintain a fast response time with low computational resource consumption and accurately assess the authenticity of online public opinions.

Keywords: online public opinion; credibility assessment; Shapley additive interpretation; text segmentation strategy; Monte Carlo sampling.

DOI: 10.1504/IJWBC.2026.155867

International Journal of Web Based Communities, 2026 Vol.22 No.1, pp.18 - 35

Received: 04 Nov 2025
Accepted: 22 Jun 2026
Published online: 24 Aug 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article