New DCNN and EResNet model lifts accuracy of human-computer interaction text classification

Researchers report a new deep learning approach for classifying discourse in human-computer interaction, designed to identify user intent from the content of conversational text. The study, published as an open access article in Discover Applied Sciences on 27 October 2026, combines a Deep Convolutional Neural Network (DCNN) with an Enhanced Residual Network (EResNet) and reports an F1 score of 91.70% on an in-house dataset, which the authors state is 2.8% higher than a BERT model.

The work addresses two persistent problems in conversational text classification. Traditional machine learning and standalone deep learning methods, according to the abstract, struggle to adapt to the colloquial nature of conversational text and do not effectively handle common class imbalances, which limits classification accuracy.

To build training and evaluation material, the study integrates publicly available benchmark data, anonymized logs, and synthetically generated samples for under-represented classes. The authors describe constructing a standardized dataset using dual-label arbitration, Bidirectional Encoder Representations from Transformers (BERT) preprocessing, and data augmentation techniques.

The model itself has three components.

The reported result is an experimental finding on the researchers’ in-house dataset rather than a deployment benchmark, and the abstract does not report performance on external test sets or in live systems. The authors frame the contribution as practical: the model, they write, effectively improves the accuracy of voice intent recognition, and the study provides a high-quality dataset and accurate classification models for human-computer interaction scenarios such as intelligent customer service, helping interactive systems respond efficiently to user needs. Because the article is being shared early as a peer-reviewed accepted manuscript, it remains subject to further edits before the final Version of Record is published.

Subject of Research: Deep learning models for classifying user intent in human-computer interaction discourse

Article Title: Text classification of human computer interaction discourse using DCNN and EResNet

Article References: Mao, Z., Wu, X., & Ye, A. (2026). Text classification of human computer interaction discourse using DCNN and EResNet. Discover Applied Sciences. https://doi.org/10.1007/s42452-026-09753-0

Image Credits: AI Generated

DOI: 10.1007/s42452-026-09753-0

Keywords: text classification, human-computer interaction, DCNN, EResNet, BERT, user intent recognition, class imbalance, data augmentation, attention module, intelligent customer service, conversational text, Text