Chinese Named Entity Recognition for a Power Customer Service Intelligent Q&A System

Power customer service intelligent Q&A system can greatly improve the efficiency of power customer service and reduce labor costs. In order to deal with the questions that need to be solved by reasoning, it is necessary to build the power customer service knowledge graph and accurately understand the questions. One of the key tasks is to implement a named entity recognizer using the historical log data of power customer service Q&A. Recently, lattice based neural networks have gained great advantages in Chinese named entity recognition. However, lattice based models rely heavily on an external predetermined dictionary, and the quality of the dictionary may interfere with entity boundary learning. As the power customer service Q&A is a form of oral conversation., it is difficult to build the specialized dictionary, which seriously restricts the application of the original menthod of lattice structure based neural network for Chinese named entity recognition in the field of power customer service. Therefore, this paper proposes a method of using entity boundary locally in lattice based neural networks for Chinese named entity recognition. Through joint learning of entity boundary and entity recognition, without any external dictionary, experiments on data sets in the field of power customer service show that this method has very good potential.