NLITS: A Natural Language Interface for Time Series Databases

Yuting Lin, Jianqiu Xu, Xieyang Wang, Yitong Zhang

Abstract: Time series data has become an important part of many fields. The management of time series data is becoming increasingly important due to the growing demand for applications. However, most users lack the ability to use specialized time series databases. Thus, we develop a natural language interface for time series databases called NLITS. The system comprises three components: data preprocessing, natural language understanding, and executable database statement generation. We preprocess natural language queries to extract and normalize time information. To extract the key entity information, we establish a time series data knowledge base and propose a context-based time series data parsing algorithm. Meanwhile, we train a model for query category recognition through a designed time series data corpus by using the LSTM network. Finally, NLITS generates executable database statements. Our demonstration will showcase how NLITS enables users directly querying time series databases with natural language. Experiments show that NLITS has a good performance with an average response time of 1.14 s, a translatability of 91.7% and a translation accuracy of 88.9%.