Electric power marketing entity relationship extraction method based on full-word shielding and multi-feature extraction

The invention provides an entity relationship extraction method oriented to the field of power marketing, and relates to the technical field of natural language processing. The method comprises the following steps: constructing a power marketing field data set; according to the method, data of a data set in the field of power marketing is input into a WmRoBERTa model to vectorize a text, so that the relationship between entity information and sentence information is better mastered, namely, the relationship between an entity and a whole sentence is obtained. Inputting the trained semantic feature information into a PCNN network to learn sentence local feature information; the method comprises the following steps: firstly, extracting a sentence, then inputting an obtained result into an Att-BiLSTM network module to extract front and back context feature information of each word in the sentence, and finally, using a full-connection neural network as a classifier to better screen out a relationship of entity pai