question-repeat/test_similarity.py

import json
import numpy as np
from duplicate_checker import QuestionDuplicateChecker

checker = QuestionDuplicateChecker()

user_input = {
    'stem': '如图，两个同心圆中，大圆的半径为$5$，小圆的半径为$3$，若大圆的弦$AB$与小圆有公共点，则$AB$的取值范围是（  ）',
    'options': '{"A": "$8\\le AB\\le10$", "B": "$8<AB\\le10$", "C": "$4\\le AB\\le5$", "D": "$4<AB\\le5$"}',
    'answer': 'A',
    'solution': '如图，过$O$点作$OC\\perp AB$于$C$，连接$OA$，则$AC=BC$。当$AB$与小圆相切时，大圆的弦$AB$与小圆有唯一公共点，$OC$取最大值$3$，此时$AC$取最小值，为$\\sqrt{5^2-3^2}=4$，∴弦$AB$的最小值为$2\\times4=8$；当点$C$与$O$重合时，$AB$的值最大，$AB$为大圆的直径，即$AB$的最大值为$10$，∴$AB$的取值范围是$8\\le AB\\le10$。<image src="https://file.chunsunqiuzhu.com/data/2026/01/17/20260117180759A644.png"/>'
}

# Fetch ID 3094
q3094 = checker.fetch_question_from_db(3094)

if not q3094:
    print("Question 3094 not found")
    exit()

print(f"Comparing user input with ID 3094\n")

# Calculate embeddings for each part to see where it drops
parts = ['stem', 'options', 'answer', 'solution']
for p in parts:
    u_text = user_input.get(p, '')
    db_text = str(q3094.get(p, '') or '')
    
    u_emb = checker.get_embedding(u_text)
    db_emb = checker.get_embedding(db_text)
    
    if u_emb is not None and db_emb is not None:
        # Normalize
        u_emb = u_emb / np.linalg.norm(u_emb)
        db_emb = db_emb / np.linalg.norm(db_emb)
        
        sim = np.dot(u_emb, db_emb)
        print(f"Part {p} similarity: {sim:.4f}")
    else:
        print(f"Part {p} failed to get embedding")

mega_u = checker.get_weighted_embedding(user_input)
mega_db = checker.get_weighted_embedding(q3094)
if mega_u is not None and mega_db is not None:
    overall_sim = np.dot(mega_u, mega_db)
    print(f"\nOverall weighted similarity: {overall_sim:.4f}")
else:
    print("\nMega vector calculation failed")