A Study of Policy Will Recognition Model for Public Opinion Texts Based on Multi-level Feature Fusion with BERT

A Study of Policy Will Recognition Model for Public Opinion Texts Based on Multi-level Feature Fusion with BERT

by

WENG Ke-ruiZHOU Ya-jieYU Shi-wei A Study of Policy Will Recognition Model for Public Opinion Texts Based on Multi-level Feature Fusion with BERT

Abstract: Traditional policy needs research has gradually shifted to the use of social media for policy needs intelligence discovery due to cost and time factors. Although social media provides rich public policy will, capturing policy views in it is challenged by semantic ambiguity and complex comment network relationships. To address the above issues, this paper proposes the ConTextBERT-CNN model to identify public policy intentions on social media. The model combines the optimised BERT pre-training model and the improved TextCNN architecture, enhances the Chinese semantic understanding through the full word masking technique, and fuses the outputs of different layers of decoding layers to achieve fine extraction of multi-layer semantic information. The experimental results show that the ConTextBERT-CNN model achieves classification accuracies of 86.4%, 82.0%, and 82.5% when dealing with the datasets on the topics of new energy vehicles, carbon neutrality, and time-sharing tariff policies, respectively, which are significantly better than the traditional deep learning methods, demonstrating that it has high efficiency and accuracy in capturing and parsing the public’s policy intentions.
Key words: social media；policy need；BERT；public opinion policy text
A Study of Policy Will Recognition Model for Public Opinion Texts Based on Multi-level Feature Fusion with BERT

引用本文： WENG Ke-rui, ZHOU Ya-jie, YU Shi-wei. A Study of Policy Will Recognition Model for Public Opinion Texts Based on Multi-level Feature Fusion with BERT[J]. Zhongguo Dizhi Daxue Xuebao (Social Science Edition), 2025, 25(01): 131-140.

作者简介：WENG Ke-rui, China University of Geosciences (Wuhan), School of Economics and Management (Hubei Wuhan 430078); ZHOU Ya-jie, China University of Geosciences (Wuhan), School of Economics and Management; YU Shi-wei (corresponding author), China University of Geosciences (Wuhan), School of Economics and Management, [email protected]

基金信息： National Natural Science Foundation of China, Major Project “Research on Ecological Modeling of Complex Policy Decision-making Scenarios” (72293572); National Natural Science Foundation of China, “Research on Policy Demand Mining and Dissemination Evolution Model Linked to Appeals and Hotspots” (72474201); Ministry of Education Humanities and Social Sciences Research Planning Fund Project “Research on Identification and Dissemination Intervention of False Information on Social Media Based on Multi-granularity Features” (24YJA630101)

中图分类号： D63-3

文章编号： 1671-0169 (2025) 01-0131-10

文献标识码： A

出版日期： 2025-01-20

网刊发布日期： 2025-01-20

责任编辑：孙洁

《中国地质大学学报(社会科学版)》 投稿须知 《中国地质大学学报(社会科学版)》是教育部主管、中国地质大学(武汉)主办的社会科学综合性学术刊物，为国家社会科学基金资助期刊、中文社会科学引文索引(CSSCI)来源期刊、全国中文核心期刊、中国人文社会科学核心期刊。根据《中国学术期刊影响因子年报（人文社会科学·2024版）》，本刊复合影响因子5.014，位于Q1区。 本刊常设栏目有资源环境研究（中宣部首批哲学社会科学期刊重点专栏、教育部“名栏工程”建设栏目，下设“资源环境经济与管理”和“环境资源法”两个子栏目）、马克思主义研究、新闻与传播、管理与经济等，同时不定期推出高等教育、生态哲学等栏目。 现面向社会征稿，稿件要求是尚未公开发表的原创性论文，版权清晰，不存在一稿多投。请按要求在本刊网站(http://ddxs.cbpt.cnki.net)实名注册并完善相关信息后，以非匿名形式Word文件在线投稿，请勿直接寄给栏目编辑。若有多个作者也请一一注册。投稿后，原则上不得变更作者或者作者顺序，请通讯作者在投稿前核对各位作者信息及作者排序，并征得各位作者同意，逐一签署《著作权许可使用协议》，签名必须本人手写。 本刊实行国际通行的双向匿名审稿制度，稿件处理周期为三个月，三个月内未收到编辑部回复的可自行处理稿件并第一时间通知编辑部撤稿。本刊不收取任何形式的审稿费或版面费，亦不与任何中介机构合作。稿件录用后，编辑部将尽快安排排版和刊期，刊出后稿酬从优。

A Study of Policy Will Recognition Model for Public Opinion Texts Based on Multi-level Feature Fusion with BERT

Leave a Comment