Scalable Information Extraction in Stratosphere

该项目的主要目标是实现对大量非结构化文本的基于查询的分析。我们设想用户使用平流层查询语言来制定IE任务。这样的查询通过Markl、Freytag和Kao在项目A、B和C中开发的方法在云基础设施上进行解析、优化、并行、执行和重新优化。IE特有的运算符，将文本压缩成结构化表示形式，就是在这个项目中开发的。此外，我们与项目E合作，开发了用于系统地聚合提取的信息的运算符，充分考虑了提取的信息的不确定性。所有IE操作员都将可配置为采用不同的IE策略，要么是面向高吞吐量、高精度/低不确定性，要么是面向高召回率。高级操作符接口必须是独立于域的，而它们的具体实例化需要容易地适应手边的文本域。这些要求需要将简单的IE技术、高级的NLP和机器学习仔细地平衡起来。在这个项目中开发的所有方法都将在生物医学领域的大型和现实的IE任务中进行评估。

The main objective of this project is to enable query-based analysis of large quantities of unstructured text. We envision users to formulate IE tasks with the Stratosphere query language. Such a query is parsed, optimized, parallelized, executed, and re-optimized on a Cloud infrastructure by methods developed in projects A, B, and C by Markl, Freytag, and Kao. The IE-specific operators, which crunch text into structured representations, are developed in this project. Furthermore, we develop, in cooperation with the Project E, operators for a systematic aggregation of extracted information that fully take the uncertainty of extracted information into account. All IE operators will be configurable to embrace different IE strategies, either geared towards high throughput, high precision / low uncertainty, or high recall. The high-level operator interfaces must be domain independent, while their concrete instantiations need to be easily adaptable to the text-domain at hand. These requirements call for a carefully balanced mixture of simple IE techniques, advanced NLP, and Machine Learning. All methods developed within this project will be evaluated on large and realistic IE tasks in the biomedical domain.