A lattice LSTM-based framework for knowledge graph construction from power plants maintenance reports

Historical experience plays a significant role in the intelligent maintenance of power plants. While maintaining power equipment, engineers would record the experience in maintenance documents called status reports. Through decades of maintenance, massive status reports have been accumulated. These text data contains rich knowledge about power equipment, and they can be a strong support for intelligent maintenance. However, to fully utilize the knowledge from these reports is not easy because of two main reasons. First, there are a huge amount of data, making it difficult to find the specific knowledge we want. Second, the knowledge contained in reports is unorganized, and few previous works have been attempted to automatically mine the knowledge from these text data. To address this problem, we propose an innovative framework for automatic construction and reasoning of Chinese knowledge graph toward intelligent maintenance of power plants. In this framework, the lattice LSTM and multi-grained lattice framework (MG lattice) are adopted to extract entities and relations respectively from text data. What’s more, we present a dataset for Chinese Named Entity Recognition, which contains four categories of entities and consists of 864 sentences from status reports. Comprehensive experiments are carried out on this dataset. The experimental results show that the lattice LSTM method is significantly superior to classic LSTM-CRF model on power plant maintenance data, implying the effectiveness and potential of our proposed framework.