Cross-Platform Cultural Content Classification and Influence Evaluation Using Enhanced TextCNN with Multi-Scale Feature Fusion
Year: 2026
Type: conference-paper
Abstract: The rapid proliferation of cultural and tourism content across heterogeneous social media platforms necessitates efficient cross-domain text mining techniques. Traditional deep learning models and massive pre-trained architectures often struggle with the semantic distribution shifts between different platforms, such as the concise hashtags of Weibo and the extensive descriptive paragraphs of Xiaohongshu. To address this critical challenge, this paper proposes an Enhanced TextCNN model designed specifically for the multi-task classification and influence evaluation of cross-platform cultural content. The proposed architecture integrates a multi-scale feature fusion module to capture complex linguistic structures and a novel platform-aware feature normalization layer to dynamically align domain-specific semantic spaces. Extensive experiments were conducted on a custom dataset comprising four thousand eight hundred and fifty posts collected from Weibo and Xiaohongshu. The empirical results demonstrate that the Enhanced TextCNN achieves a superior classification accuracy of 89.5 percent and a Macro-F1 score of 89.1 percent across five cultural categories. Furthermore, the model records an exceptionally low mean squared error of 0.042 in the continuous influence prediction task. Notably, the proposed lightweight model successfully outperforms the standard unaligned BERT architecture in both cross-domain accuracy and computational efficiency. These findings validate that explicitly modeling platform-specific linguistic traits provides a highly robust, low-latency, and scalable solution for real-time cultural trend monitoring and digital humanities research.
Author: Qiu Chen
Institution: Dalian Art College
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7208716736