This paper studies the extraction method of symptom entity attributes from electronic medical records. By analyzing the characteristics of the language description of symptom entity segments, a dictionary-assisted multi-feature fusion symptom entity attribute extraction model SWET is proposed. The model uses domain dictionaries to assist in generating word set features that reflect the word-formation positions of text characters, and fuses them with character features to generate text expressions with richer semantics, which effectively promotes the model’s ability to correctly learn professional vocabulary. At the same time, the model improves the Transformer Encoder framework at the feature extraction layer. On the one hand, it introduces multiple convolutional layers of different scales to capture different ranges of contextual information in the text, helping the model better understand the structure of the text. On the other hand, it designs a dynamic adjustment mechanism for the number of multi-head attention heads according to the complexity of the text, improving the model’s ability to capture text information. Experimental results show that this method can better extract the defined symptom entity attributes.