Short Text Emotion Classification Based On SimCSE and BERT Hybrid Model
In order to solve the problem that the training effect of the BERT model is affected by the anisotropy of the text vector.This paper combines comparative learning(SimCSE)and BERT to build a model(SimCSE-BERT).The classifier not only expands the amount of training data through the idea of comparative learning,but also obtains text vectors with good ″alignment″and ″uniformity″based on the SimCSE model to optimize the basic BERT model to improve the classification effect.Experimental results are as follows:compared with the basic BERT model,the accu-racy of the hybrid model increases by 0.562,0.584,and 0.734 percentage points in takeout,Ctrip hotel,and Taobao data sets respectively.The classification effect of this model on short text emotional classification data set has been significantly improved,and it has good generalization ability.
NETL
NSTL
万方数据
