A Hybird Learning-based Framework for Estimating Teaching Ability of Ideological and Political Courses in Modern Chinese University
Lun Li, Sun Xiu-fang · 2021
This paper proposes a hybrid learning-based two-stage framework consisting of an RNNs-based architecture for subjective textual sentiment analysis and a deep decision forest for teaching ability estimation. According to the annotated self-collected dataset and SnowNLP corpus, the connectivity of the subjective items' answers is enhanced to be fed into fixed-sized RNNs with LSTM units for distribution of soft labels. The outputted labels are merged with the objective items' answers to form the learnable vectors for the deep global decision forest framework. We use the top-5 features' variance to weigh the random forests in each level of the cascade forest structure for a fitness voting result. On the self-collected dataset, the F1 score of 0.974 and the CPM score of 0.925 have reached, which are prior to those of the compared methods.