| 面向查询的观点摘要模型研究:以Debatepedia为数据源 |
| |
| 引用本文： | 余传明,郑智梁,朱星宇,安璐.面向查询的观点摘要模型研究:以Debatepedia为数据源[J].情报学报,2020,39(4):374-386. |
| |
| 作者姓名： | 余传明 郑智梁 朱星宇 安璐 |
| |
| 作者单位： | 中南财经政法大学信息与安全工程学院,武汉 430073;中南财经政法大学信息与安全工程学院,武汉 430073;McKelvey School of Engineering,Washington University in St.Louis,St.Louis,Missouri 63130;武汉大学信息管理学院,武汉 430072 |
| |
| 基金项目： | 国家自然科学基金面上项目“面向跨语言观点摘要的领域知识表示与融合模型研究”(71974202);国家自然科学基金重大课题“国家安全大数据综合信息集成与分析方法”(71790612)。 |
| |
| 摘 要： | 本文系统性地研究面向查询的观点摘要任务,旨在构建一种查询式观点摘要模型框架,探究不同的摘要方法对摘要效果的影响。通过综合考虑情感倾向与句子相似度,从待检文档中抽取出待摘要语句,再结合神经网络和词嵌入技术生成摘要,进而构建面向查询的观点摘要框架。从Debatepedia网站上爬取议题和论述内容构建观点摘要实验数据集,将本文方法应用到该数据集上,以检验不同模型的效果。实验结果表明,在该数据集上,仅使用基于抽取式的方法生成的观点摘要质量更高,取得了最高的平均ROUGE分数、深度语义相似度分数和情感分数,较生成式方法分别提高6.58%、1.79%和11.52%,而比组合式方法提高了8.33%、2.80%和13.86%;同时,本文提出的句子深度语义相似度和情感分数评估指标有助于更好地评估面向查询的观点摘要模型效果。研究结果对于提升面向查询的观点摘要效果,促进观点摘要模型在情报学领域的应用具有重要意义。
|
| 关 键 词： | 观点摘要 信息抽取 话语生成 情感分析 深度学习 |
Query-oriented Opinion Summarization Model Using Debatepedia as Datasource |
| |
| Yu Chuanming,Zheng Zhiliang,Zhu Xingyu,An Lu.Query-oriented Opinion Summarization Model Using Debatepedia as Datasource[J].Journal of the China Society for Scientific andTechnical Information,2020,39(4):374-386. |
| |
| Authors: | Yu Chuanming Zheng Zhiliang Zhu Xingyu An Lu |
| |
| Institution: | (School of Information and Safety Engineering,Zhongnan University of Economics and Law,Wuhan 430073;McKelvey School of Engineering,Washington University in St.Louis,St.Louis,Missouri 63130;School of Information Management,Wuhan University,Wuhan 430072) |
| |
| Abstract: | This study systematically studies query-oriented opinion summarization, aiming to construct a query-oriented opinion summarization framework to explore the impact of different text summarization methods on the opinion summarization. Considering sentiment orientation and the similarity between the sentence and query, we extract the sentences from the original documents and employ neural networks and word embeddings to achieve abstractive summarization. The query-oriented opinion summarization framework is then constructed upon this. We crawl the topics and arguments to build the experimental opinion summarization dataset from the Debatepedia websites and apply the proposed method to the dataset to validate its effect. The experimental results show that on this dataset, the summaries generated by the extractive method are of higher quality;the highest average ROUGE score, deep semantic similarity score, and emotional score are6.58%, 1.79%, and 11.52% higher than the generative method, and 8.33%, 2.80%, and 13.86% higher than the combined method, respectively. Furthermore, the evaluation indicator deep sentence similarity and the sentiment score proposed in this study can better evaluate the effects of the query-oriented opinion summarization model. The research results are of great significance for improving the effects of query-oriented opinion summarization and promoting the application of the opinion summarization model in the field of information science. |
| |
| Keywords: | opinion summarization information extraction discourse generation sentiment analysis deep learning |
| 本文献已被 维普 万方数据 等数据库收录！ |
|
