java-sim-ast-opt.py

# -*- coding: utf-8 -*-
"""
Abstract Syntax Tree Similarity Detection for Java Code

Martinez-Gil, J. (2024). Source Code Clone Detection Using Unsupervised Similarity Measures. arXiv preprint arXiv:2401.09885.

@author: Jorge Martinez-Gil
"""

import os
import tokenize
import io
import javalang
from difflib import SequenceMatcher

def calculate_similarity2(snippet1, snippet2):
    # Tokenize the code snippets.
    tokens1 = list(tokenize.tokenize(io.BytesIO(snippet1.encode('utf-8')).readline))
    tokens2 = list(tokenize.tokenize(io.BytesIO(snippet2.encode('utf-8')).readline))

    # Calculate the similarity score.
    num_common_tokens = 0
    for token1 in tokens1:
        for token2 in tokens2:
            if token1.string == token2.string:
                num_common_tokens += 1
                break
    total_tokens = len(tokens1) + len(tokens2) - num_common_tokens
    similarity_score = num_common_tokens / total_tokens if total_tokens > 0 else 0.0

    return similarity_score

def get_ast(code):
    tokens = javalang.tokenizer.tokenize(code)
    parser = javalang.parser.Parser(tokens)
    return parser.parse()

def ast_to_string(ast):
    return str(ast).replace('\n', '')

def compare_asts(ast1, ast2):
    str1 = ast_to_string(ast1)
    str2 = ast_to_string(ast2)
    matcher = SequenceMatcher(None, str1, str2)
    return matcher.ratio()

def calculate_similarity(snippet1, snippet2):
    # Get the ASTs for the code snippets.
    ast1 = get_ast(snippet1)
    ast2 = get_ast(snippet2)

    # Compare the ASTs and return the similarity score.
    return compare_asts(ast1, ast2)


# Define the path to the IR-Plag-Dataset folder
dataset_path = os.path.join(os.getcwd(), "IR-Plag-Dataset")

# Define a list of similarity thresholds to iterate over
similarity_thresholds = [0.1, 0.5, 0.52]

# Initialize variables to keep track of the best result
best_threshold = 0
best_accuracy = 0

# Initialize counters
TP = 0
FP = 0
FN = 0

# Loop through each similarity threshold and calculate accuracy
for SIMILARITY_THRESHOLD in similarity_thresholds:
    # Initialize the counters
    total_cases = 0
    over_threshold_cases_plagiarized = 0
    over_threshold_cases_non_plagiarized = 0
    cases_plag = 0
    cases_non_plag = 0

    # Loop through each subfolder in the dataset
    for folder_name in os.listdir(dataset_path):
        folder_path = os.path.join(dataset_path, folder_name)
        if os.path.isdir(folder_path):
            # Find the Java file in the original folder
            original_path = os.path.join(folder_path, 'original')
            java_files = [f for f in os.listdir(original_path) if f.endswith('.java')]
            if len(java_files) == 1:
                java_file = java_files[0]
                with open(os.path.join(original_path, java_file), 'r') as f:
                    code1 = f.read()
                # print(f"Found {java_file} in {original_path} for {folder_name}")

                # Loop through each subfolder in the plagiarized and non-plagiarized folders
                for subfolder_name in ['plagiarized', 'non-plagiarized']:
                    subfolder_path = os.path.join(folder_path, subfolder_name)
                    if os.path.isdir(subfolder_path):
                        # Loop through each Java file in the subfolder
                        for root, dirs, files in os.walk(subfolder_path):
                            for java_file in files:
                                if java_file.endswith('.java'):
                                    with open(os.path.join(root, java_file), 'r') as f:
                                        code2 = f.read()
                                    # print(f"Found {java_file} in {root} for {folder_name}")
                                    similarity_ratio = calculate_similarity(code1, code2)
                                    #print(f"{subfolder_name},{similarity_ratio:.2f}")

                                    # Update the counters based on the similarity ratio
                                    if subfolder_name == 'plagiarized':
                                        cases_plag += 1
                                        if similarity_ratio >= SIMILARITY_THRESHOLD:
                                            over_threshold_cases_plagiarized += 1
                                    elif subfolder_name == 'non-plagiarized':
                                        cases_non_plag += 1
                                        if similarity_ratio <= SIMILARITY_THRESHOLD:
                                            over_threshold_cases_non_plagiarized += 1
                                    total_cases += 1
                                    # Update the counters based on the similarity ratio
                                    if subfolder_name == 'plagiarized':
                                        cases_plag += 1
                                        if similarity_ratio >= SIMILARITY_THRESHOLD:
                                            TP += 1  # True positive: plagiarized and identified as plagiarized
                                        else:
                                            FN += 1  # False negative: plagiarized but identified as non-plagiarized
                                    elif subfolder_name == 'non-plagiarized':
                                        cases_non_plag += 1
                                        if similarity_ratio <= SIMILARITY_THRESHOLD:
                                            over_threshold_cases_non_plagiarized += 1
                                        else:
                                            FP += 1  # False positive: non-plagiarized but identified as plagiarized

            else:
                print(f"Error: Found {len(java_files)} Java files in {original_path} for {folder_name}")

    # Calculate accuracy for the current threshold
    if total_cases > 0:
        accuracy = (over_threshold_cases_non_plagiarized + over_threshold_cases_plagiarized) / total_cases
        if accuracy > best_accuracy:
            best_accuracy = accuracy
            best_threshold = SIMILARITY_THRESHOLD

    # Calculate precision and recall
    if TP + FP > 0:
        precision = TP / (TP + FP)
    else:
        precision = 0

    if TP + FN > 0:
        recall = TP / (TP + FN)
    else:
        recall = 0

    # Calculate F-measure
    if precision + recall > 0:
        f_measure = 2 * (precision * recall) / (precision + recall)
    else:
        f_measure = 0

# Print the best threshold and accuracy
print(f"{os.path.basename(__file__)} - The best threshold is {best_threshold} with an accuracy of {best_accuracy:.2f}, Precision: {precision:.2f}, Recall: {recall:.2f}, F-measure: {f_measure:.2f}")