Update app.py

app.py

@@ -8,8 +8,14 @@ from PIL import Image
 from typing import List
 import requests
 
-# Initialize sentence transformer model
-model1 = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
+import torch
+from transformers import BertTokenizer, BertModel
+import torch.nn.functional as F
+
+# Load pre-trained BERT model and tokenizer
+tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
+model = BertModel.from_pretrained('bert-base-uncased')
+
 
 # Initialize Groq client
 client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
@@ -54,21 +60,28 @@ def extract_text_from_image(filepath: str, languages: List[str]):
 
 
 
-import Levenshtein
+# Function to get BERT embeddings
+def get_bert_embedding(text):
+    inputs = tokenizer(text, return_tensors='pt', truncation=True, padding=True)
+    with torch.no_grad():
+        outputs = model(**inputs)
+    # Get the embeddings from the last hidden state
+    embeddings = outputs.last_hidden_state.mean(dim=1)
+    return embeddings
+
+# Function to calculate cosine similarity
+def calculate_cosine_similarity(embedding1, embedding2):
+    similarity = F.cosine_similarity(embedding1, embedding2)
+    return similarity.item()
+
+# Function to compare logic of student and teacher answers
+def compare_answers(student_answer, teacher_answer):
+    student_embedding = get_bert_embedding(student_answer)
+    teacher_embedding = get_bert_embedding(teacher_answer)
+    similarity_score = calculate_cosine_similarity(student_embedding, teacher_embedding)
+    return similarity_score
 
-# Calculate Levenshtein Distance
-def calculate_levenshtein_distance(text1, text2):
-    return Levenshtein.distance(text1, text2)
 
-# Use Levenshtein distance to refine similarity
-def calculate_similarity_with_levenshtein(text1, text2):
-    similarity = calculate_similarity(text1, text2)  # From embeddings
-    edit_distance = calculate_levenshtein_distance(text1, text2)
-    normalized_distance = 1 - (edit_distance / max(len(text1), len(text2)))
-    
-    # Combine similarity score with normalized Levenshtein distance
-    final_similarity = (similarity + normalized_distance) / 2
-    return final_similarity
 
 
 # Assign badges based on the grade
@@ -107,21 +120,12 @@ def get_grade(similarity_score):
     else:
         return 1
 
-def get_embedding(text):
-    return model1.encode(text, convert_to_tensor=True)
-
-# Calculate similarity between two texts focusing on sequence
-def calculate_similarity(text1, text2):
-    embedding1 = get_embedding(text1)
-    embedding2 = get_embedding(text2)
-    similarity = util.pytorch_cos_sim(embedding1, embedding2)
-    return similarity.item()
 
 
 # Function to evaluate student's answer by comparing it to a model answer
 def evaluate_answer(image, languages, model_answer):
     student_answer = extract_text_from_image(image, languages)
-    similarity_score = calculate_similarity_with_levenshtein(student_answer, model_answer)
+    similarity_score = compare_answers(student_answer, model_answer)
     grade = get_grade(similarity_score)
     feedback = f"Student's answer: {student_answer}\nTeacher's answer: {model_answer}"
     # visual_feedback = generate_sequence_feedback(student_answer, model_answer)