File size: 4,441 Bytes
39cf944
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
08f222a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39cf944
08f222a
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
"""
N-gram analysis for comparing text responses
"""
from sklearn.feature_extraction.text import CountVectorizer
import numpy as np
from collections import Counter
import nltk
from nltk.util import ngrams
from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords


def compare_ngrams(texts, model_names, n=2, top_n=25):
    """
    Compare n-gram representations across multiple texts.

    Args:
        texts (list): List of text responses to compare
        model_names (list): Names of models corresponding to responses
        n (int): Size of n-grams (1 for unigrams, 2 for bigrams, etc.)
        top_n (int): Number of top n-grams to consider

    Returns:
        dict: N-gram analysis results
    """
    # Initialize the results dictionary
    result = {
        "models": model_names,
        "ngram_size": n,
        "important_ngrams": {},
        "ngram_count_matrix": {},
        "differential_ngrams": []
    }

    # Make sure we have texts to analyze
    if not texts or len(texts) < 1:
        return result

    # Convert n to integer if it's a string
    if isinstance(n, str):
        n = int(n)
    
    # Convert top_n to integer if necessary
    if isinstance(top_n, str):
        top_n = int(top_n)

    try:
        # Create n-gram representations using CountVectorizer
        vectorizer = CountVectorizer(
            ngram_range=(n, n),  # Use the specified n-gram size
            max_features=1000,
            stop_words='english'
        )

        X = vectorizer.fit_transform(texts)

        # Get feature names (n-grams)
        feature_names = vectorizer.get_feature_names_out()

        # Create n-gram count matrix
        ngram_counts = {}
        for i, model in enumerate(model_names):
            counts = X[i].toarray()[0]
            ngram_counts[model] = {}

            # Store n-gram frequencies for this model
            for j, ngram in enumerate(feature_names):
                if counts[j] > 0:  # Only store n-grams that appear
                    ngram_counts[model][ngram] = int(counts[j])

                    # Add to n-gram count matrix
                    if ngram not in result["ngram_count_matrix"]:
                        result["ngram_count_matrix"][ngram] = {}
                    result["ngram_count_matrix"][ngram][model] = int(counts[j])

        # Find important n-grams for each model
        for model, ngram_freq in ngram_counts.items():
            # Sort by frequency
            sorted_ngrams = sorted(ngram_freq.items(), key=lambda x: x[1], reverse=True)

            # Store top N n-grams
            result["important_ngrams"][model] = [
                {"ngram": ngram, "count": count}
                for ngram, count in sorted_ngrams[:top_n]
            ]

        # Calculate differential n-grams (n-grams with biggest frequency difference between models)
        if len(model_names) >= 2:
            model1, model2 = model_names[0], model_names[1]

            # Calculate differences
            diff_scores = {}
            for ngram in result["ngram_count_matrix"]:
                count1 = result["ngram_count_matrix"][ngram].get(model1, 0)
                count2 = result["ngram_count_matrix"][ngram].get(model2, 0)

                # Absolute difference
                diff_scores[ngram] = abs(count1 - count2)

            # Sort by difference
            sorted_diffs = sorted(diff_scores.items(), key=lambda x: x[1], reverse=True)
            result["differential_ngrams"] = [ngram for ngram, _ in sorted_diffs[:top_n]]

            # Calculate overlap statistics
            model1_ngrams = set(ngram_counts.get(model1, {}).keys())
            model2_ngrams = set(ngram_counts.get(model2, {}).keys())
            common_ngrams = model1_ngrams.intersection(model2_ngrams)

            # Initialize comparisons if needed
            if "comparisons" not in result:
                result["comparisons"] = {}

            comparison_key = f"{model1} vs {model2}"
            result["comparisons"][comparison_key] = {
                "common_ngram_count": len(common_ngrams)
            }

        return result
    except Exception as e:
        import traceback
        error_msg = f"N-gram analysis error: {str(e)}\n{traceback.format_exc()}"
        print(error_msg)
        # Return basic structure with error
        return {
            "models": model_names,
            "ngram_size": n,
            "error": str(e)
        }