CAREER: Enriching Conversational Information Retrieval via Mixed-Initiative Interactions

该奖项的全部或部分资金来自2021年美国救援计划法案(公法117-2)。很明显，通过自然语言对话提供信息访问将在搜索技术的未来发挥重要作用。这将通过开发高效和有效的对话式搜索引擎来实现。现有系统通常基于查询-响应范例进行设计，在该范例中，用户通过提交键入的单词或短语来发起交互，并且系统用一个或多个文档进行响应。该过程重复进行，直到用户接收到有用的响应或终止搜索会话。这不是交互的最佳设计。一个更好的方法是创建像对话一样运作的搜索系统。例如，在对话式搜索系统中，即使新信息不是对搜索查询的明确响应，系统也可以提出澄清的问题或推荐新信息。作为一种对话式搜索系统，对话应该产生促进用户满意这一最终目标所需的信息。前面提到的查询-响应范例不支持这些自然的对话交互。这个职业奖项旨在通过设想超越这种查询-响应范例的解决方案来推进最先进的技术。为了实现这一目标，该项目研究了在信息寻找对话中生成和处理混合主动交互的理论和机器学习解决方案。更详细地，本项目探索了以下三个研究主题：(1)开发测量混合主动信息寻求对话的理论基础；(2)开发模型来澄清被认为是最常见的混合主动交互类型的用户的信息需求；以及(3)开发主动信息对正在进行的对话的贡献的模型。除了这些算法和模型方面的贡献，该项目还开发了许多宝贵的资源来推进对话式信息检索领域，包括一个对话式学术助理代理，它将被用作在线实验和公共数据创建的工具。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).It has become clear that providing access to information through natural language conversations will play a significant role in the future of search technology. This will be enabled by developing efficient and effective conversational search engines. Existing systems are generally designed based on a query-response paradigm, in which the user initiates the interaction by submitting typing a word or phrase, and the system responds with one or more documents. This process repeats itself until the user either receives a useful response or terminates the search session. This is not an optimal design for interaction. A better approach would be to create search systems that operate like a conversation. In a conversational search systems, for instance, the system may ask a clarifying question or can recommend new information even though it is not an explicit response to the search query. A conversational search system, the conversation should yield the information that is needed to facilitate the ultimate goal of user satisfaction. The mentioned query-response paradigm does not support these natural conversational interactions. This CAREER award aims to advance the state-of-the-art by envisioning solutions that go beyond this query-response paradigm.To achieve this goal, this project studies theoretical and machine learning solutions for generating and handling mixed-initiative interactions in information seeking conversations. In more detail, this project explores the following three research thrusts: (1) developing theoretical foundations for measuring mixed-initiative information seeking conversations; (2) developing models for clarifying the user's information needs which is considered as the most common mixed-initiative interaction type; and (3) developing models for proactive informational contributions to ongoing conversations. In addition to these algorithmic and modeling contributions, this project also develops a number of invaluable resources for advancing the field of conversational information retrieval, including a conversational scholarly assistant agent that will be used as a tool for online experimentation and public data creation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.