In this paper, we propose an enhanced version of the vanilla transformer for data-to-text generation and then use it as the generator of a conditional generative adversarial model to improve the semantic quality and diversity of output sentences. Specifically, by adding a diagonal mask matrix to the attention scores of the encoder and using the history of the attention weights in the decoder, this enhanced version of the vanilla transformer prevents semantic defects in the output text. Also, using this enhanced transformer along with a triplet network, respectively, as the generator and discriminator of conditional generative adversarial network, diversity and semantic quality of sentences are guaranteed. To prove the effectiveness of the proposed model, called conditional generative adversarial with enhanced transformer (CGA-ET), we performed experiments on three different datasets and observed that our proposed model is able to achieve better results than the baselines models in terms of BLEU, METEOR, NIST, ROUGE-L, CIDEr, BERTScore, and SER automatic evaluation metrics as well as human evaluation.

Sleep apnea is one of the most common sleep disorders. The consequences of undiagnosed sleep apnea can be very serious, increasing the risk of high blood pressure, heart disease, stroke, and Alzheimer’s disease over a long period of time. However, many people are often unaware of their condition. The gold standard for diagnosing sleep apnea is nighttime polysomnography monitoring in a specialized sleep laboratory. However, these diagnoses are expensive and the number of beds is limited, and there is insufficient monitoring in terms of time dimension. Existing methods for automated detection use no more than three physiological signals, but all other signals are also associated with the patient’s sleep. In addition, the limited amount of medical real annotation data, especially abnormal samples, lead to weak model generalization capability. The gap between model generalization capability and medical field needs still exists. In this paper, we propose a method for integrating medical interpretation rules into a long short-term memory neural network based on self-attention with multichannel respiratory signals as input. We obtain attention weights through a token-level attention mechanism and then extract key rules of medical interpretation to assist the weights, improving model generalization and reducing the dependence on data volume. Compared with the best prediction performance of existing methods, the average improvements of our method in accuracy, precision, and f1-score are 3.26%, 7.03%, and 1.78%, respectively. The algorithm tested the performance of our model on the Sleep Heart Health Study data set and found that the model outperformed existing methods and could help physicians make decisions in their practices.