Text Summarization and Question Generation Models

自然语言处理（NLP）是信息时代最重要的技术之一。理解复杂的语言表达也是人工智能的重要组成部分。NLP的应用无处不在，因为人们几乎用语言交流一切：网络搜索、新闻、广告、电子邮件、客户服务、语言翻译、放射学报告、社交媒体、虚拟助理、会话系统（聊天机器人）、博客、产品评论等，我打算继续关注我之前调查过的领域中的问题，即在文本摘要领域，问题的产生。我感兴趣的是建立一个系统，一方面从文档中概括、综合和抽象信息内容，另一方面捕捉用户的需求、意图和偏好，并生成定制的个性化文档。文本摘要和问题生成是解决信息过载和信息疲劳问题的研究方向。这将导致理论，技术和工具，导致丰富的信息使用和有效的决策。 在这个建议中，我计划研究以下目标：在短期内，（1）开发具有结构和推理的鲁棒表示学习，（2）构建鲁棒的抽象文本摘要模型，（3）设计综合问题生成模型，并构建自动生成答案不可知问题、探究性问题和多跳问题的模型，（4）研究文本摘要和问题生成之间的互补性，从长远来看（1）捕捉用户需求并根据用户需求和意图定制摘要，（2）从信息性摘要转向推理性摘要，以及（3）设计智能机制来处理自然语言（即，推理、综合、概括、浓缩和总结信息）以及用于连贯和定制信息传递的机制。本研究建议将文本摘要和问题生成技术相结合，以形成真正有用的信息传递工具。我正计划开发模型、技术和工具，以便向用户提供与接收用户相关和适当的信息，使用户能够容易地吸收这些信息，从而使他们能够执行任务，通常能够作出更好和更有效的决策。这些工具可用于监视、过滤、提取、抽象和检测大量文本数据中的特定/相关信息和证据，从而为最终用户节省时间。这项工作需要一些感兴趣的本科生和研究生在不同阶段的参与，并将有几个研究贡献。研究结果将改进最先进的文本摘要和问题生成模型，并将在同行评审会议和期刊上传播。

Natural language processing (NLP) is one of the most important technologies of the information age. Understanding complex language utterances is also a crucial part of artificial intelligence. Applications of NLP are everywhere because people communicate mostly everything in language: web searches, news, advertisements, emails, customer services, language translations, radiology reports, social media, virtual assistants, conversational systems (chatbots), blogs, product comments, etc. I intend to continue to focus on problems in the areas that I have previously investigated, namely in the areas of text summarization, and question generation. I am interested in building up systems that on the one hand gist, synthesize, and abstract information content from documents, and on the other hand capture user needs, intents, and preferences and generate customized, personalized documents. Text summarization and question generation are research directions for solving the information overload and the information fatigue problems. This will result in theories, technologies, and tools leading to rich information use and efficient decision-making.  In this proposal, I plan to study the following objectives: in a short-term (1) developing robust representation learning with structure and reasoning, (2) building robust abstractive text summarization models, (3) devising comprehensive question generation models, and constructing models that generate automatically answer-agnostic questions, inquisitive questions, and multiHop questions, (4) studying the complementarity between text summarization and question generation, and in a long-term (1) capturing users' needs and tailoring the summaries as answers to user needs and intents, (2) shifting from the informative summarization to the inferential summarization, and (3) designing intelligent mechanisms to process natural language (i.e., inference, synthesize, gist, condense and summarize information) as well as mechanisms for coherent and tailored information delivery. This research proposal deals with ways of combining text summarization and question generation technologies to form truly useful information delivery tools. I am planning to develop models, technologies and tools to deliver to users information that is relevant and appropriate for the users receiving it, information that they can easily assimilate to enable them to perform their tasks, usually enabling better and more efficient decision-making. These tools can be used to monitor, filter, extract, abstract, and detect specific/relevant information and evidence in large amount of textual data, hence, saving time to the end-user. The work requires the involvement of a number of interested undergraduate and graduate students at different stages, and will have several research contributions. The results will improve the state-of-the-art text summarization and question generation models and will be disseminated in peer-reviewed conferences and journals.