Searching and Analyzing Big Data: Context-sensitive and Task-aware Approaches

使用谷歌很容易。使用谷歌查找信息则是另一回事。几十年来，信息检索(IR)取得了长足的进步。然而，IR远未成为一个解决的问题，仍然存在许多挑战。首先，大多数Web搜索引擎将简短的文本查询作为输入，并输出排序的文档列表。检索决策主要基于当前的查询和文档集合。给定查询的结果通常是相同的，与用户或用户发出请求的上下文无关。例如，一个人可能在某个时间使用“IRIX”来表示上下文中的信息检索，但在另一个时间使用IRIX操作系统。当前的Web搜索系统不可能区分这两种情况，因为没有考虑用户的搜索上下文。第二，信息检索总体上是一个互动的过程。在当前以文档为中心的检索范式下，交互式检索被视为一系列独立的简单检索决策步骤。然而，人们已经注意到，面向任务的用户会话的分析提供了对用户查询行为的有用洞察。第三，目前包括通用搜索引擎(如Google和Bing)和科学文献搜索引擎(如PubMed和ACM数字图书馆)在内的大多数信息检索系统都使用关键词来查询和索引文档。然而，这种传统的基于关键字的信息检索模型几乎没有为理解用户信息需求提供语义上下文，这给搜索中的查询和文档不匹配带来了严峻的挑战。理想情况下，如果查询和文档在主题上相关或在语义上相近，人们希望看到它们彼此匹配。因此，根据用户对集合中的文档的理解将语义上下文集成到信息检索系统中(例如，通过利用自然语言处理技术)将提高信息检索性能。第四，在每天积累大量数据的情况下，上下文相关和面向任务的方法实施起来更具挑战性。随着深度学习和数据分析技术的出现，上下文敏感和面向任务的方法对于基于语义的匹配变得更加可行。这项研究的长期目标是克服现有信息检索方法的局限性，正式开发一种新的检索范式，称为上下文敏感和任务感知的大数据信息搜索。具体地说，(1)我们将开发新的理论模型来自动分析海量文本数据，以高效地提取知识，并进行准确的语义匹配和精确匹配；(2)我们将开发一个新的检索框架，用于捕获大数据中的用户信息和搜索上下文；(3)我们将为上下文敏感信息检索开发新的基于任务的检索方法，以优化整个检索过程的长期检索效用。

Using Google is easy. Finding information using Google is another matter. Over the decades, significant progress has been made in Information Retrieval (IR). However, IR is far from a solved problem and many challenges remain. First, most Web search engines take a short text query as input and output a ranked list of documents. The retrieval decision is made primarily based on the current query and document collection. The results for a given query are usually identical, independent of the user or the context in which the user made the request. For example, a person may use "IRIX" to mean Information Retrieval in Context at one time, but IRIX operating systems at another time. It is impossible for the current Web search systems to distinguish these two cases because the user's search context is not considered. Second, IR is, in general, an interactive process. With the current document-centered retrieval paradigm, interactive retrieval is treated as a sequence of independent simple retrieval decision making steps. However, it has been brought into attention that analysis of task-oriented user sessions provides useful insight into the query behavior of the users. Third, most of present IR systems including general search engines (e.g. Google and Bing) and scientific literature search engines (e.g. PubMed and ACM Digital Library) use keywords to query and index documents. However, this traditional keyword based IR model provides little semantic context for the understanding of user information needs, which poses a critical challenge of mismatch between query and document in search. Ideally, one would like to see query and document match with each other, if they are topically relevant or semantically close. Thus, the integration of semantic context according to the user's understanding of the documents in the collection into IR systems (e.g. via utilizing natural language processing techniques) will improve the IR performance. Fourth, context-sensitive and task-oriented approaches are even more challenging to implement with huge amounts of data accumulated every day. With the availability of deep learning and data analyzing techniques, context-sensitive and task-oriented approaches become more feasible for semantic-based matching. The long-term objective of the proposed research is to overcome the limitations of the existing IR methods and formally develop a new retrieval paradigm called context-sensitive and task-aware information search for big data. In particular, (1) we will develop novel theoretical models to automatically analyze text data with very large quantity to efficiently extract knowledge and perform both semantic and exact matching accurately; (2) we will develop a new retrieval framework for capturing user information and search context in big data; (3) we will develop novel task-based retrieval methods for context-sensitive information retrieval to optimize the long-term retrieval utility over an entire retrieval session.