Textual relation recognition for information retrieval

本研究以情报学论文智能搜索引擎为目标应用，旨在开发一种自动分类搜索查询及其搜索结果之间的关系的方法。通过对实际论文摘要的分析，我们找出了这些论文中普遍存在的文本关系，如方法、目的和结果，并开发了一个由100篇论文摘要组成的标注数据集。通过使用该数据集进行训练/评估，我们开发了一种自动识别这些关系的方法，并达到了实用的准确率水平。我们将该方法应用于3000篇信息科学论文摘要，并开发了一个基于这些关系自动分类搜索结果的原型搜索引擎。

This research aims to develop a method for automatically classifying relationships between a search query and its search results, with an intelligent search engine of information science papers as a target application. By analyzing actual paper abstracts, we identified textual relations that appear in general in those papers, such as method, purpose, and result, and developed an annotated data set that consists of 100 paper abstracts. By using this data set for training/evaluation, we developed a method for automatically recognizing these relations, and achieved a practical level of accuracy. We applied this method to 3000 abstracts of information science papers, and develpoed a prototype search engine that automatically classifies search results based on these relationships.