JMIR Publications

ABSTRACT

Background:

With the prevalence of Online Consultation, many dialogues information are accumulated, which in authentic language environment are of significant value to the research and development of intelligent question answering and auxiliary diagnosis.Contextual data which contains patient-doctor dialogues has been utilized in recent Natural Language Processing (NLP) studies.

Objective:

Find a more effective and simple model to achieve named entity automatic classification in Patient-Doctor dialogues with Chinese RoBERTa-WWM-ext-CNN

Methods:

In this paper, our task is named entity automatic annotation and classification in patient-doctor dialogues. We adapt downstream architecture in Chinese RoBERTa-WWM-ext, which combines text convolutional neural network (CNN). We use RoBERTa-WWM-ext to express sentence semantics as a text vector, then extract the local features of the sentence through CNN, which is our new fusion model. To verify its knowledge learning ability, we choose kNowledge IntEgration (ERNIE)、original Bidirectional Encoder Representations from Transformers (BERT) and Chinese BERT with Whole Word Masking (WWM) to do the same task, then compare these models’ results.

Results:

From the scoring results, it shows that our model outperforms the other models on this task.

Conclusions:

The fine-tuning in the downstream task of the model and the integration of multi-model can be well optimized.