Emotional Transfer Framework of Short Texts in Chinese
Abstract
The emotion analysis technology based on deep learning is widely studied and applied. However, its dependence on large data set makes this technology unable to play a role in small domains with only light data sets. To break the application barrier of this technology, we combine the knowledge in deep learning and cross-domain transfer and put forward the emotional transfer framework of short texts in Chinese. The framework takes the task domain as the target domain and selects the one that has sufficient text data and semantic feature intersection as the source domain. It extracts semantic features from the source domain and transfers them to the target domain to make up for the lack of sufficient data in the target domain. It helps the text emotion analysis task work effectively when there are less data in the target domain. This paper introduces the principle of the framework from four modules: the data preprocessing module, the semantic feature transfer module, the training module for the words order model, and the monitoring module for model innovation. Finally, the experimental results show that The framework is feasible, the model effect is good, and the accuracy of emotion classification is 89.0%. It is better than other methods in recent years.