【作者】 张巍；

【导师】 余翔湛； 许保勋；

【作者基本信息】 哈尔滨工业大学 ， 软件工程（专业学位）， 2017， 硕士

【摘要】 在互联网+大数据时代,决策日益基于数据和分析做出,而非经验和直觉。近年来,随着信贷、消费等领域个人“用户画像”的成功应用,如何对资本市场企业和人物对象进行全方位、多角度的模型刻画正在成为金融监管和投融资的一个新热点。本文基于作者在证券交易所的实际开发项目,针对资本市场中证券交易所的监管需求,设计并实现了一个以资本市场人物和企业关系图谱为主要数据模型的信息系统。本文研究的人物和企业关系图谱是知识图谱技术在资本市场这一垂直领域的应用。知识图谱技术自从2012年Google发布以来,其在改进搜索引擎服务质量和效率方面作用明显。本文参考知识图谱技术的通用构建框架,提出了以实体获取和实体关系抽取为主要手段的关系图谱构建方案。在实体获取方面,应用深度学习技术,以长短时记忆学习网络作为语料特征学习模型,以条件随机场为序列标注模型,构建了在文本语料中识别命名实体的方案;在实体关系抽取方面,结合领域知识和业务需求,从公司公告年报等半结构化数据中以规则匹配抽取实体关系。此外,设计并实现了关系图谱在系统中的查询展示功能,提供了良好的可视化及交互性。本文对于资本市场人物和企业关系图谱的设计实现是基于多源异构数据的模型,具有信息价值密度大,抽象层次高以及应用范围广的特点。该业务可以广泛服务于证券交易所的上市公司持续监管、市场监察与执法、以及发行审核与投融资对接等业务,对中国多层次资本市场建设具有重要的支持价值。更多还原

【Abstract】 In the era of Internet + big data,decision making is increasingly based on data and analysis,rather than experience and intuition.In recent years,with the successful application of individual “user portraint” on consumption,credit and other areas,how to characterize and modeling the figures and enterprises in the capital market in all directions and multi-angle is becoming a new hot spot in the field of financial supervision and investment.Based on the author ’s actual project development in the stock exchange,this paper designs and implements an information system based on the capital market figures and the enterpri se relationship graph as the main data model for the regulatory demand of the stock exchange in the capital market.The relation graph of figures and enterprises in capital market in this paper is the application of knowledge map technology in the vertical field of capital market.Knowledge graph technology has been effective in improving the quality and efficiency of search engine service since been first released by Google in 2012.In this paper,by referring to the general construction framework of knowledge graph technology,the authur proposes a relational graph construction scheme based on entity acquisition and entity relation extraction as the main means.In the aspect of entity acquisition,the deep learning technique is used to construct the learning model of the named entity in the text corpus with the short term memory learning network as the corpus feature learning model and the conditional random field as the sequence annotation model.In relation extraction,with domain knowledge and business requirements,from the company’s annual report and other semi-structured data,matching rules are defined to extract the entity relationship.In addition,there is the design and implementation of the relationship graph in the system query and display function,providing a good visual service and interactive functions.The design and implementation of relation graph in this paper is a model based on multi-source heterogeneous data,which has the characteristics of high information density,high level of abstraction and wide application range.The achievement can be widely used in the stock exchange continuously monitoring the listed companies and market monitoring and law enforcement,and the issuance of audit and investment and financing docking business.So the result contains important value to support China’s multi-level capital market construction.更多还原