Power marketing small sample named entity recognition method based on multiple tasks

The invention provides an electric power marketing small sample named entity recognition method based on multiple tasks. The method comprises the following steps: S1, collecting an electric power marketing named entity recognition corpus and carrying out data processing; s2, inputting the text into a RoBERTa encoder to obtain feature representation; s3, inputting the feature representation obtained in the S2 into an IDCNN layer for expansion convolution; s4, inputting the text data after expansion convolution into a seed module, and predicting seed scores of each single character and each double character: segmenting the text into a seed set containing one Chinese character and two Chinese characters, calculating the score of each seed in the set, and if the score is greater than a seed score threshold omega, taking the score as a candidate seed; s5, expanding the candidate seeds to obtain a complete entity; and S6, inputting the entity into a RoBERTa encoder to obtain a final classification result. According