Using technologies to support investigative interviewers in detecting deception

该项目探讨了如何通过提供技术支持来促进调查性访谈（例如，警察，移民，HMRC，福利等）。这种支持采用用户界面的形式，显示与当前访谈阶段相关的事实，例如受访者应该认识的人的姓名。这些事实是从维基百科等知识库以及相关组织的网站中提取出来的。网络信息抽取是本研究的关键课题。挑战在于在整个Web站点上执行命名实体识别和共同引用解析，同时考虑到Web上存在的更丰富的信息，例如格式、位置、标题和标题。然后将为每个Web页面提取的这些实体集成为整个Web站点的实体图，该实体图可由用户界面使用。本文提出了一种半监督的直接在HTML上进行训练的方法，该方法由从维基百科中提取的词典指导。

The project explores how investigative interviews (e.g., police, immigration, HMRC, benefits, etc.) can be facilitated by the provision of technological support. This support takes the form of a user interface that displays factoids relevant to the current stage of the interview, for example the name of a person the interviewee should know. These factoids are extracted from knowledge bases such as Wikipedia and also from the Web sites of relevant organisations. This Web Information Extraction is the key topic of this research. The challenge is to perform Named Entity Recognition and Co-reference Resolution on a whole Web site, taking into account the richer information present on the Web, such as formatting, position, titles and headings. These extracted entities for each Web page are then integrated to form an entity graph for the whole Web site that can be consumed by the user interface. A semi-supervised approach to training direct on HTML is proposed, guided by a lexicon extracted from Wikipedia.