面向复杂查询的异质媒体搜索理论与方法研究

Over the past few years, the area of search engine has witnessed the development of keyword search, entity search, unstructured data search, and vertical search, meeting the ever-increasing needs of search for information. The next breakthrough may very well be in the heterogeneous media search, through a comprehensive use of multiple media information types to effectively answer complex queries. This project builds on the research results on a variety of search techniques, and focuses on an increased use of semantic information at various points during the search process. The search queries themselves may be expressed in a cross-media form, and the search objects combine many different media types. In this way, the search technology becomes broader, deeper, and smarter. The key research questions addressed in this project include (1) how to bridge the high-level semantics in the search queries and the low-level semantics needed to find the search objects, (2) how to extract useful semantic features for specific search queries from the unstructured data, (3) how to collaboratively analyze the cross media data for semantic information, as well as (4) how to efficiently perform the search on massive amounts of data. The projects uses four components in its approaches, namely (1) knowledge graph to support cross-media semantic knowledge integration, (2) semantic templates to support intelligent query decomposition and transformation, (3) demand-driven heterogeneous media analysis to deepen the semantic mining for specific cross media queries, and (4) cloud computing architecture to support fast search and knowledge extraction. The project is expected to gain insights in complex query understanding and cross media mining of semantic information, resulting in innovative theories and methods. With Baidu’s participation, the project expects to put the theories and methods to practical tests to build a solid basis for future wide-spread adoption of complex heterogeneous media search tasks.

近年来，搜索技术经历了关键词搜索、实体搜索、非结构化数据搜索以及垂直搜索等，满足不断提升的搜索需求。异质媒体搜索综合利用异质媒体的交叉信息，解答复杂查询，可能是搜索技术的下一突破口。本项目建立在各类搜索研究成果的基础上，更多地利用语义层面的信息，支持搜索问题本身及搜索对象的跨媒体化，加强搜索的广度、深度和智能程度。所涉及的关健科学问题包括搜索层高语义到操作层低语义的转化，非结构化数据面向个性化搜索的语义特征抽取，跨媒体数据的协同语义分析，以及大规模数据的高效搜索。在研究方法上，以知识图谱为知识组织方法，用于跨媒体语义综合；以语义模版做为基础构件，实现智能化查询分解及转化；以异质媒体搜索需求为驱动，深化媒体个性化交叉语义挖掘；以云计算构架，实现快速知识抽取及搜索。项目预期在复杂查询理解、语义知识交叉挖掘利用等方面提出创新性理论和方法，并与百度公司合作，为异质媒体复杂搜索的理论与实践奠定基础。

近年来，搜索技术经历了关键词搜索、实体搜索、非结构化数据搜索以及垂直搜索等，以满足不断提升的搜索需求。异质媒体搜索综合利用异质媒体的交叉信息，解答复杂查询，探索搜索技术的下一突破口。本项目建立在各类搜索研究成果的基础上，更多地利用语义层面的信息，支持搜索问题本身及搜索对象的跨媒体化，加强搜索的广度、深度和智能程度。所涉及的关健科学问题包括搜索层高语义到操作层低语义的转化，非结构化数据面向个性化搜索的语义特征抽取，跨媒体数据的协同语义分析，以及大规模数据的高效搜索。在研究方法上，以知识图谱为知识组织方法，用于跨媒体语义综合；以语义模版做为基础构件，实现智能化查询分解及转化；以异质媒体搜索需求为驱动，深化媒体个性化交叉语义挖掘；以云计算构架，实现快速知识抽取及搜索。在研究成果方面，项目在异质媒体处理和分析、知识获取和知识库构建、快速搜索方面形成了一系列创新型成果，在国内外权威会议和期刊上累计发表高水平学术论文*篇，开发了一个异质媒体的搜索系统原型，并将科学研究成果应用于新浪新闻和微博的基于搜索的异质媒体聚合系统中。