Figure 3
A flowchart illustrating the KeydistilTF model for extracting key phrases from teacher and student responses. The process begins with the input of a reference answer (RA) and a student answer (SA), both with human annotations. In Step 1, preprocessing is done using Stanford CoreNLP and KeyBERT libraries to tokenize candidate words for both RA and SA. Step 2 involves generating embeddings using pretrained DistilBERT embeddings. In Step 3, key phrases are extracted, ensuring they are unique, positive, and negative for both RA and SA. Step 4 involves vectorization using Tf-IDF. Finally, in Step 5, the output similarity is calculated between the key phrases of RA and SA.Example of proposed KeydistilTF model