金融领域的事件句抽取
发布时间:2019-03-31 09:02
【摘要】:事件句抽取是事件抽取中的核心环节,在金融领域中,公司名识别则是事件句抽取中的重点和难点。针对金融领域的事件句抽取,首先充分利用互联网搜索和上市公司名信息进行公司名识别,如果一个N元组是公司名,则进行互联网搜索的结果中包含"公司""集团"等字词多,同时与公司名库中部分公司名有较高的匹配度;其次,综合考虑句子位置信息、包含公司名信息、包含领域动词信息、与标题相似度四个方面特征,构造权值表达式;最终从句子集中选出金融事件句。在数据集上测试,实验结果证明提出的金融领域事件句抽取方法是可行的,公司名识别方法的正确率可达82.28%,召回率达68.93%,事件句抽取的正确率可达66.83%。
[Abstract]:Event sentence extraction is the core of event extraction. In the field of finance, corporate name recognition is the focus and difficulty in event sentence extraction. For the event sentence extraction in the financial field, first of all, make full use of the Internet search and listed company name information to identify the company name, if an N tuple is the company name, Then the result of Internet search contains many words such as "company" and "group", and has a high degree of matching with some of the company names in the company name library; Secondly, considering sentence position information, including company name information, domain verb information and title similarity, weight expression is constructed. Finally, financial event sentence is selected from sentence set. The test results on the data set show that the proposed method is feasible. The correct rate of the company name recognition method is 82.28%, the recall rate is 68.93%, and the correct rate of event sentence extraction is 66.83%.
【作者单位】: 北京信息科技大学网络文化与数字传播北京市重点实验室;首都师范大学北京成像技术高精尖创新中心;
【基金】:2014年度国家社会科学基金委托课题(14@ZH036) 北京成像技术高精尖创新中心资助项目(BAICIT-2016003) 国家自然科学基金资助项目(61271304,61671070)
【分类号】:TP391.1
,
本文编号:2450760
[Abstract]:Event sentence extraction is the core of event extraction. In the field of finance, corporate name recognition is the focus and difficulty in event sentence extraction. For the event sentence extraction in the financial field, first of all, make full use of the Internet search and listed company name information to identify the company name, if an N tuple is the company name, Then the result of Internet search contains many words such as "company" and "group", and has a high degree of matching with some of the company names in the company name library; Secondly, considering sentence position information, including company name information, domain verb information and title similarity, weight expression is constructed. Finally, financial event sentence is selected from sentence set. The test results on the data set show that the proposed method is feasible. The correct rate of the company name recognition method is 82.28%, the recall rate is 68.93%, and the correct rate of event sentence extraction is 66.83%.
【作者单位】: 北京信息科技大学网络文化与数字传播北京市重点实验室;首都师范大学北京成像技术高精尖创新中心;
【基金】:2014年度国家社会科学基金委托课题(14@ZH036) 北京成像技术高精尖创新中心资助项目(BAICIT-2016003) 国家自然科学基金资助项目(61271304,61671070)
【分类号】:TP391.1
,
本文编号:2450760
本文链接:https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2450760.html