Title: Sentiment analysis model for evaluating foreign language courses in colleges and universities based on improved Bi-LSTM and attention mechanisms

Authors: Ran Li

Addresses: School of English Language and Culture, Xi'an FANYI University, Xi'an, China

Abstract: With growing demand for sentiment analysis in evaluating college foreign language courses, traditional rule- or statistics-based methods struggle with multilingual mixing, cultural differences and domain-specific vocabularies. To overcome these challenges in classifying fine-grained sentiment (positive, neutral, negative) in feedback, this study proposes a Bi-LSTM with CNN sentiment model. Experiments show the model achieves high classification accuracy: 100%, 99.3% and 98.8% for positive, neutral and negative sentiments, respectively. Compared with traditional methods, accuracy improved by 15%. The model also achieved P, R and F1-scores of 98.31%, 97.89% and 98.23% for positive, 97.56%, 97.34% and 98.06% for negative and 95.13%, 95.65% and 95.66% for neutral sentiments. In conclusion, the model provides an efficient, accurate tool for evaluating foreign language courses, enhancing sentiment analysis in higher education.

Keywords: Bi-LSTM; CNN; attention mechanism; course evaluation; sentiment analysis.

DOI: 10.1504/IJWMC.2026.153170

International Journal of Wireless and Mobile Computing, 2026 Vol.30 No.3, pp.290 - 300

Received: 15 Nov 2024
Accepted: 20 Jun 2025
Published online: 25 Apr 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article