【摘要】 针对目前自然语言时空问句查询存在语义信息提取不完整、意图识别不准确等问题，提出一种面向具有时空特性的自然语言问句语义理解方法。该方法基于融合注意力机制的BiLSTM-CRF模型，实现时空问句的自动语义编码和语义信息提取。利用预定义的问句类型和语义编码结果对问句进行分类，通过对类内问句的语义解析完成问句意图识别。为满足不同数据库查询的需求，通过构建中间语言的方式实现其到数据库查询语言的转换。结果表明本文方法能准确实现基本查询和复合查询中的语义信息提取及意图理解，且准确度高达93.69%;以数据库语言Cypher为例，时空问句到查询语言的转换精度达到71%。更多还原

【Abstract】 Currently, there exist issues with incomplete semantic information extraction and inaccurate intent recognition in spatiotemporal natural language query processing. This paper proposes a method for semantic understanding of natural language queries with spatiotemporal characteristics. To address the problem of incomplete semantic information extraction in spatiotemporal queries, we propose a BiLSTM-CRF model incorporating attention mechanisms to automatically encode and extract semantic information from such queries. We classify queries using predefined query types and semantic encoding results, and then perform intent recognition by parsing the semantics of queries within each class. To meet the requirements of different database queries, we implement transformations to database query languages via an intermediate language. Results demonstrate that our method accurately achieves semantic information extraction and intent understanding in both basic and composite queries, with an accuracy rate of up to 93.69%. Taking the database language Cypher as an example, the precision of transforming spatiotemporal queries into query language reaches 71%.更多还原