A Text Classification Model for Agricultural Expert Forums Based on Expert Domain Preferences and Deep Semantics

Agricultural expert forums serve as important platforms for knowledge exchange, containing large volumes of text that embody professional expertise and practical experience. However, because experts often focus on different agricultural subfields and category semantics may overlap, traditional text classification methods that rely solely on textual content often show limited discriminative power. To address this issue, this paper proposes a text classification model for agricultural expert forums that integrates expert domain preferences with deep semantic representations. Specifically, the model incorporates an expert domain preference matrix and an expert semantic vector matrix into the deep semantic encoding of forum texts to construct interaction representations between experts and categories, thereby enabling collaborative modeling of textual semantics and expert domain preferences. To further improve classification performance, a category-level enhancement strategy based on domain attention is introduced at the output stage. This strategy uses candidate category masking and probability constraints to guide the model in adaptively aligning its predictions with experts’ long-term domain focus patterns. Experimental results on over 110,000 text sentences show that the proposed model improves average precision by approximately 12.7% and the overall F1 score by around 10.1%, with stable gains in both accuracy and recall. Compared with mainstream encoder-based classification models, the proposed model achieves an average accuracy improvement of 11.67%, demonstrating significant and robust performance advantages. Overall, the proposed model enhances both the accuracy and robustness of text classification in agricultural expert forums while improving interpretability, providing effective technical support for agricultural knowledge management and intelligent information retrieval.