Efficient Semantic Search on Big Data

这个项目是关于大数据的高效语义搜索，特别是非常大的文本集合和非常大的知识库。在第一轮SPP中，我们做出了以下贡献：一个新的搜索引擎，用于文本和知识库的交互式组合搜索；一种新的可扩展文本语义连贯分解算法一种新的知识库三元组关联分数计算框架与算法一种自动将自然语言问题转换为知识库问题的自学习问答系统广泛的文本和知识库语义搜索领域综述。在SPP的下一轮中，我们计划为其中一些问题开发改进的解决方案，以及在我们第一轮工作中突然出现的新问题的解决方案：一个全功能的SPARQL+文本引擎（现有的SPARQL引擎只有中等功能的文本搜索扩展，我们的引擎从第一轮开始只支持树形查询，并依赖于它们的增量结构）；我们的问答系统的扩展，以查询更复杂的模式，可能涉及文本搜索组件；自动完成自然语言问题的系统；改进大规模命名实体识别和语义搜索消歧。对于所有命名的问题，我们的目标是（和第一轮一样）：可证明的高效算法和数据结构；对其效率和质量进行广泛的实验评估；开源软件和可公开访问的演示或原型；通过提供所有相关材料（如果可能）或专门的web应用程序来完全再现我们的结果。

This project is about efficient semantic search on big data, notably very large text collections and very large knowledge bases. In the first round of this SPP we have made the following contributions: a new search engine for interactive combined search on text and knowledge bases; a new scalable algorithm for decomposing text into its semantically coherent units; a new framework and algorithm for computing relevance scores for knowledge base triples; a self-learning question answering system for automatically translating natural-language questions into knowledge base queries;a comprehensive survey on the vast field of semantic search on text and knowledge bases.In the next round of this SPP, we plan to develop improved solutions for some of these problems, as well as solutions for new problems that cropped up during our work in the first round: a full-featured SPARQL+Text engine (existing SPARQL engines have only moderately powerful text-search extensions, our engine from the first round supports only tree-shaped queries and relies on their incremental construction); an extension of our question answering system to query patterns that are more complex and may involve a text-search component; a system for automatic completion of natural-language questions; improved large-scale named entity recognition and disambiguation for semantic search.For all the named problems, our goals are (like in the first round): provable efficient algorithms and data structures; an extensive experimental evaluation of their efficiency and quality; open-sourced software and a publicly accessible demonstrator or prototype;full reproducibility of our results by providing all relevant materials (if possible) or a dedicated web application.