A Method for Identifying Power Customer Demands Based on Neural Network Language Model

Power customer service plays a pivotal role in providing technical support and consulting services to customers when they encounter difficulties with electricity utilization. At present, electricity customer service primarily relies on telephone channels to manually address customer demands, which necessitates a high level of professional knowledge among customer service personnel. However, with the continuous expansion of customer demands and the increasing complexity of electrical business knowledge, customer service personnel face greater challenges in extracting customer electricity needs, which hinders the improvement of customer satisfaction. This paper proposes a method using speech material recognition algorithms to improve the accuracy of feature extraction from speech materials. A noise removal algorithm is also proposed to perform secondary noise reduction based on expert experience, thus eliminating background noise in voice materials. A customer demand phrase model based on the semantics of the electric power profession is designed, which defines the mapping rules between voice material phrases and customer demands. A neural network-based long short-term prediction algorithm is used to train a semantic extraction classifier for speech materials, which adopts an attention mechanism to increase the network's attention and weight for less frequent text sequences. Finally, the accuracy of the algorithm is trained based on the voice data of power customer service, and the input and hierarchical parameters of the model are optimized to find the parameters corresponding to the best classification performance.