Abstracting, Gisting, Tailoring, and Delivering Information Contents to User Needs and Intents

在当今的信息社会，我们可以接触到海量的信息。如果没有适当的理论、技术和工具来以明确和有用的方式获取我们所需要的信息，那么信息过载或信息疲劳的风险就会不断增加，导致信息利用不良和决策效率低下。 在自然语言处理和人类语言技术领域，有几个我目前感兴趣的研究问题。我打算继续关注我之前调查过的领域中的问题，即文本摘要、复杂问题回答和问题生成。我感兴趣的是开发技术，以提供与用户适当和相关的信息，避免信息过载，并帮助做出明智的决策。这将导致建立捕获用户需求和偏好的系统，并生成定制的个性化文档。 在这项提案中，我将解决这些新的需求和要求，并计划从与文本摘要、问题回答和问题生成相关的观点来研究这些目标。我认为，它们具有很强的互补性，必须根据以下目标和方面进行研究和探讨。在短期内，开发回答复杂问题的文本摘要和问答模型，建立将复杂问题分解成容易回答的简单问题的分解模型，构建自动生成问题的模型，从摘要摘要转向抽象摘要，研究摘要技术对海量数据的可扩展性。从长远来看，这是一种捕捉用户需求并将摘要调整为对用户需求和意图的回答的机制，具有根据主题生成和定制问题的机制，设计处理自然语言的智能机制，即推理、合成、主旨、浓缩和总结信息，以及连贯和定制的信息交付机制。 使我们能够实现这些目标的科学方法是不同类型的。一方面，我们打算探索基于机器学习技术的方法，特别是涉及深度学习技术的方法，因为这些方法表明，它们能够获得对文本数据的高层抽象进行建模的数据表示，并且这些表示能够捕获含义。另一方面，我们打算探索基于词汇、句法和语义知识的句子结构，目的是压缩、简化、归类和抽象摘要和答案的选定内容。此外，我们还打算探索允许我们处理大量文本数据的框架，以实现摘要技术的可伸缩性。

In today's information society, we have access to an enormous amount of information. Without appropriate theories, technologies, and tools to obtain the information we need in a clear and useful way, there is an ever increasing risk of information overload or information fatigue, leading to poor information use and inefficiencies in decision-making. There are several research problems that are of current interest to me in the fields of natural language processing and human language technologies. I intend to continue to focus on problems in the areas that I have previously investigated, namely text summarization, complex question answering, and question generation. I am interested in developing techniques for delivering information that is appropriate and relevant to the user, avoiding information overload and assisting informed decision making. This will lead to build up systems that capture user needs and preferences and generate customized, personalized documents. In this proposal, I address these new needs and demands and I plan to study these objectives from the points of views that relate text summarization, question answering, and question generation. I contend that they hold a strong complementarity which has to be studied and explored according to the following goals and aspects. In a short-term, developing text summarization and question answering models that answer complex questions, building decomposition models that break down complex questions into simpler questions that are easy to answer, constructing models that generate automatically questions, shifting from extractive summarization to abstractive summarization, studying the scalability of the summarization techniques to large amount of data. In a long-term capturing users' needs and tailoring the summaries as answers to user needs and intents, having mechanisms that generate and tailor questions from topics, and 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. The scientific approaches that will allow us to achieve these goals are of different kinds. On the one hand, we intend to explore methods based on machine learning techniques, especially methods that involve the deep learning techniques since these methods showed that they are able to acquire representations of data that model high-level abstractions of textual data and those representations capture meaning. On the other hand, we intend to explore the structures of sentences based on lexical, syntactic and semantic knowledge for the purpose of compressing, reducing, gisting, and abstracting the selected content for the summaries and the answers. Also we intend to explore frameworks that allow us to process large amounts of textual data for the purpose of the scalability of the summarization techniques.