Emotion recognition plays an important role in Human Computer Interaction (HCI) and the evaluation of human behavior based on emotional state is an important research topic. The purpose of emotion recognition is to automatically identify human’s emotional states by analyzing physiological or non-physiological signals. The conventional emotion classification methods cannot comprehensively leverage global and local features which are extracted from Electroencephalogram (EEG) signal generated after being stimulated. Therefore, we propose the graph convolutional neural network based emotion recognition with brain functional connectivity network (GERBN). Firstly, raw EEG data of the public DEAP and SEED datasets is preprocessed and adopted in this study. Secondly, emotion-related brain functional connection pattern is constructed using Phase-Locking Value (PLV) adjacency matrix to measure connectivity between the signals of different EEG channels according to phase synchronization. A novel graph structure is constructed where the EEG electrode channels are defined as the vertex, and the edge is strong connection of the binary brain network. Thirdly, the GERBN model that includes six layers is designed to classify and recognize emotional states on the two-dimensional emotional models of valence and arousal. Finally, extensive experiments are conducted on DEAP and SEED datasets. Experimental results demonstrate that the proposed method can improve classification accuracies, in which average accuracies of 80.43% and 88.47% on DEAP are attained on valence and arousal dimensions, respectively. On the SEED dataset, the accuracy reaches 92.37% higher than some of the other methods.

During a hospital visit, a significant volume of Gastroscopy Diagnostic Text (GDT) data are produced, representing the unstructured gastric medical records of patients undergoing gastroscopy. As such, GDTs play a crucial role in evaluating the patient’s health, shaping treatment plans, and scheduling follow-up visits. However, given the free-text nature of GDTs, which lack a formal structure, physicians often find it challenging to extract meaningful insights from them. Furthermore, while deep learning has made significant strides in the medical domain, to our knowledge, there are not any readily available text-based pre-trained models tailored for GDT classification and analysis. To address this gap, we introduce a Bidirectional Encoder Representations from Transformers (BERT) based three-branch classification network tailored for GDTs. We leverage the robust representation capabilities of the BERT pre-trained model to deeply encode the texts. A unique three-branch decoder structure is employed to pinpoint lesion sites and determine cancer stages. Experimental outcomes validate the efficacy of our approach in GDT classification, with a precision of 0.993 and a recall of 0.784 in the early cancer category. In pinpointing cancer lesion sites, the weighted F1 score achieved was 0.849.