Update app.py

app.py

@@ -8,6 +8,7 @@ from typing import List
 import torch
 from transformers import BertTokenizer, BertModel
 import torch.nn.functional as F
+import language_tool_python  # Import LanguageTool for grammar checking
 
 # Load pre-trained models
 tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
@@ -23,6 +24,9 @@ system_prompt = {
     "content": "You are a useful assistant. You reply with efficient answers."
 }
 
+# Initialize grammar checker
+tool = language_tool_python.LanguageTool('en-US')
+
 async def chat_groq(message, history):
     messages = [system_prompt]
     for msg in history:
@@ -99,12 +103,26 @@ def calculate_sentence_similarity(text1, text2):
     embedding2 = sentence_model.encode(text2, convert_to_tensor=True)
     return util.pytorch_cos_sim(embedding1, embedding2).item()
 
+def check_grammar(student_answer):
+    # Check grammar using LanguageTool
+    matches = tool.check(student_answer)
+    errors = len(matches)
+    
+    # Apply a penalty based on the number of grammar errors
+    penalty = 1 - min(0.1 * errors, 0.5)  # Maximum penalty is 50%
+    return penalty
+
 def compare_answers(student_answer, teacher_answer):
     bert_similarity = calculate_cosine_similarity(get_bert_embedding(student_answer), get_bert_embedding(teacher_answer))
     sentence_similarity = calculate_sentence_similarity(student_answer, teacher_answer)
-
+    
     # Use a higher weight for BERT similarity
-    final_similarity = (0.75 * bert_similarity + 0.25 * sentence_similarity)
+    semantic_similarity = (0.75 * bert_similarity + 0.25 * sentence_similarity)
+    
+    # Apply grammar penalty
+    grammar_penalty = check_grammar(student_answer)
+    final_similarity = semantic_similarity * grammar_penalty
+    
     return final_similarity
 
 def extract_keywords(text):