Teaching evaluation text in colleges and universities has the characteristics of multiple evaluation dimensions and long text content, which makes it difficult to mine evaluation information. Based on this, this paper designed an unsupervised teaching evaluation text matching algorithm that integrated dimension construction and data enhancement.Firstly, TextRank method was used to extract keywords from the evaluation text, and the evaluation index system was constructed by dimensional induction and recursion based on the keywords. Secondly, the short text was disassembled, and the pre-training model based on the attention mechanism was used to mine the matching features between the short text and the dimensions. Finally, based on each pre-trained model, the SimCSE strategy was adopted for data enhancement, and by compared the experimental data, the best dimension matching result of the short text was obtained. The experimental results show that the models after using this strategy are better than the original training model on the accuracy RAcc and F1 indicators. Among them, the Simcse-Wobert model has the best matching effect, RAcc is 72.50%, and F1 reaches 84.06%, which indicates that the SimCSE model is introduced into evaluation text matching fields can achieve good application effects.This algorithm can realize automatic matching of teaching evaluation content and teaching evaluation dimensions, thereby can more accurately mine the fine-grained information of university evaluation personnel on each evaluation dimension, which is convenient for analyzing the focus points on the teaching links evaluation of evaluation personnel, and can provide theoretical basis for fine-grained emotion mining of teaching evaluation texts.
WANGMeng, FUYaru, MOUZhijia. The research hotspot and frontier perspective of educational text mining [J]. Journal of Open Learning, 2021, 26(3): 17-27. (in Chinese)
WANS X, LANY Y, GUOJ F, et al. A deep architecture for semantic matching with multiple positional sentence representations[DB/OL].(2015-11-26)[2023-07-12].
[11]
WANS X, LANY Y, XUJ, et al. Match-srnn: Modeling the recursive matching structure with spatial RNN[DB/OL].(2016-04-15)[2023-07-12].
[12]
JADHAVS S, THEPADES D. Fake news identification and classification using DSSM and improved recurrent neural network classifier[J]. Applied Artificial Intelligence, 2019, 33(12): 1058-1068.
[13]
WANGS H, JIANGJ. Learning natural language inference with LSTM[DB/OL].(2015-12-30)[2023-07-12].
[14]
PARIKHA P, TÄCKSTRÖMO, DAS D, et al. A decomposable attention model for natural language inference[DB/OL].(2016-06-06)[2023-07-12].
[15]
VASWANIA, SHAZEERN, PARMARN, et al. Attention is all you need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems December, 2017: 6000-6010.
[16]
DEVLINJ, CHANGM W, LEEK, et al. Bert: Pre-training of deep bidirectional transformers for language understanding[DB/OL].(2018-10-11)[2023-07-12].
MAXinyu, FANYixing, GUOJiafeng, et al. An empirical investigation of generalization and transfer in short text matching[J]. Journal of Computer Research and Development, 2022, 59(1): 118-126. (in Chinese)
ZHAOWei, WANGWenjuan, GANYufang. Electric power text matching model based on pre-training model and multi-view recurrent neural network[J]. Journal of Chongqing University of Posts and Telecommunications (Natural Science Edition),2023, 35(3): 545-553. (in Chinese)
[22]
GaoT, YaoX, ChenD. Simcse: Simple contrastive learning of sentence embeddings[DB/OL].(2021-04-18)[2023-07-12].