Task Based Information Retrieval

基于任务的信息检索

• 批准号: EP/P024289/1
• 负责人: Emine Yilmaz
• 金额: $ 106.6万
• 依托单位: University College London
• 依托单位国家: 英国
• 项目类别: Fellowship
• 财政年份: 2017
• 资助国家: 英国
• 起止时间: 2017 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=EP%2FP024289%2F1
• 关键词: Task; Based; Information; Retrieval

The need for search often arises from a person's need to achieve a goal, or a task such as booking travels, buying a house, etc. Contemporary search engines focus on retrieving documents relevant to the query submitted as opposed to understanding and supporting the underlying information needs (or tasks) that have led a person to submit the query. Therefore, search engine users often have to submit multiple queries to the current systems to achieve a single information need. Having identified that users often have to reformulate their queries in order to achieve their final goal, most current search engines attempt to assist users towards a better expression of their needs by suggesting queries to them, other than the query currently issued. However, query suggestions mainly focus on helping the user refine their current query, as opposed to helping them identify and explore aspects related to their current complex tasks. For example, when a user issues the query "flights to Barcelona", it is clear that the user is planning to travel to Barcelona and it is very likely that the user will also need to search for hotels in Barcelona or for shuttles from Barcelona airport. Since query suggestions mainly focus on refining the current query, suggestions provided by commonly used search engines are mostly of the form "flights to Barcelona from LOCATION", or "FLIGHT-CAREER-NAME flights to Barcelona" and the result pages provided by these systems do not contain any information that could help users book hotels or shuttles from the airport. For very common tasks such as arranging travels, it may be possible to manually identify and guide the user through a list of (sub)tasks that need to be achieved in order to complete a certain task (booking a flight, finding a hotel, looking for points of interest, etc. when the user tries to arrange travel). However, given the variety of tasks search engines are used for, this would only be possible for a very small subset of them. Furthermore, quite often search engines are used for achieving such complex tasks that often the searcher themself lacks the task knowledge necessary to decide which step to tackle next. For example, a searcher looking for information about how to maintain a car with no prior knowledge would first need to use the search engine to identify the parts of the car that need maintenance and issue separate queries to learn about maintaining each part. Hence, retrieval systems that can automatically detect the task the user is trying to achieve and guide them through the process are needed, where a search task has been previously defined as an atomic information need that consists of a set of related (sub)tasks. Even though designing information retrieval (IR) systems that can characterize/identify tasks, and can respond to them efficiently is listed as one of the grand challenges in information retrieval, very little progress has been made in this direction. The main goal of this proposal is to devise task oriented retrieval systems, retrieval systems that can automatically detect the task that the user is trying to achieve and guide them through the process (steps) needed to achieve the task.

对搜索的需求通常源于一个人需要实现一个目标或一项任务，例如预订旅行、买房子等。当代搜索引擎侧重于检索与所提交的查询相关的文档，而不是理解和支持导致人们提交查询的潜在信息需求（或任务）。因此，搜索引擎用户常常必须向当前系统提交多个查询才能实现单一信息需求。在认识到用户经常必须重新表述他们的查询以实现他们的最终目标之后，大多数当前的搜索引擎试图通过向用户建议除当前发出的查询之外的查询来帮助用户更好地表达他们的需求。然而，查询建议主要侧重于帮助用户完善当前的查询，而不是帮助他们识别和探索与当前复杂任务相关的方面。例如，当用户发出查询“飞往巴塞罗那的航班”时，很明显该用户计划前往巴塞罗那，并且该用户很可能还需要搜索巴塞罗那的酒店或从巴塞罗那机场出发的班车。由于查询建议主要侧重于细化当前查询，因此常用搜索引擎提供的建议大多采用“从LOCATION飞往巴塞罗那的航班”或“飞往巴塞罗那的FLIGHT-CAREER-NAME航班”的形式，并且这些系统提供的结果页面不包含任何可以帮助用户预订酒店或从机场出发的班车的信息。对于非常常见的任务（例如安排旅行），可以手动识别并引导用户完成需要完成的（子）任务列表，以便完成特定任务（当用户尝试安排旅行时预订航班、寻找酒店、寻找兴趣点等）。然而，考虑到搜索引擎用于执行的任务多种多样，这仅适用于其中的一小部分。此外，搜索引擎常常用于完成如此复杂的任务，以至于搜索者本身通常缺乏决定下一步处理哪一步所需的任务知识。例如，在没有先验知识的情况下寻找有关如何维护汽车的信息的搜索者首先需要使用搜索引擎来识别需要维护的汽车部件，并发出单独的查询以了解如何维护每个部件。因此，需要能够自动检测用户试图实现的任务并引导他们完成整个过程的检索系统，其中搜索任务先前已被定义为由一组相关（子）任务组成的原子信息需求。尽管设计能够表征/识别任务并能够有效响应任务的信息检索（IR）系统被列为信息检索领域的重大挑战之一，但在这个方向上取得的进展却很少。该提案的主要目标是设计面向任务的检索系统，检索系统可以自动检测用户试图实现的任务，并指导他们完成实现任务所需的过程（步骤）。

项目成果

SUM'20: State-based User Modelling

SUM20：基于状态的用户建模

• DOI: 10.1145/3336191.3371883
• 发表时间: 2020
• 作者: Bulathwela S

MultiWOZ 2.4: A Multi-Domain Task-Oriented Dialogue Dataset with Essential Annotation Corrections to Improve State Tracking Evaluation

• DOI: 10.18653/v1/2022.sigdial-1.34
• 发表时间: 2021-04
• 作者: Fanghua Ye;Jarana Manotumruksa;Emine Yilmaz

Analysing Social Media Forums to Discover Potential Causes of Phasic Shifts in Cryptocurrency Price Series

• DOI: 10.3389/fbloc.2020.00001
• 发表时间: 2020-01
• 作者: Andrew Burnie;Emine Yilmaz;T. Aste

A User Model for Query-Sessions

查询会话的用户模型

• 发表时间: 2019
• 作者: Aldo Lipani

Towards an Integrative Educational Recommender for Lifelong Learners

为终身学习者打造综合教育推荐系统

• 发表时间: 2020
• 期刊: AAAI 2020 - 34th AAAI Conference on Artificial Intelligence
• 作者: Bulathwela S.