Application Model of Natural Language Processing Technology in Diagnosing Algorithmic Discrimination in Educational Apps

Résumé fourni par la source

With the increasing popularity of educational apps, algorithmic discrimination poses a threat to educational equity. Traditional methods struggle to address implicit discrimination in unstructured text. This study proposes an NLP-based Multi-Level Semantic Matching Algorithm (MSMA). This algorithm achieves discrimination diagnosis through text preprocessing, improved semantic feature extraction from BERT (introducing a semantic attention mechanism for educational purposes), and a multi-level matching diagnosis module. The experiment used 100,000 texts from K-12, higher education, and language learning apps (including 12,000 discriminatory texts). The data was split into training, validation, and test sets in a 7:2:1 ratio. The results were compared with traditional BERT + logistic regression and LSTM + SVM algorithms. Results show that MSMA achieved a test set accuracy of 92.3% (improvements of 5.6% and 10.8% over the two comparison algorithms, respectively), an F1 score of 91.1% (improvements of 6.6% and 11.9%, respectively), and an implicit discrimination detection rate of 88.2% (improvements of 8.7% and 15.6%, respectively). The system is adaptable to all three educational scenarios, with an inference time of 12.3 seconds per 1,000 items. This approach strikes a balance between accuracy and efficiency, providing an effective solution for diagnosing algorithmic discrimination in educational apps.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.