File size: 6,743 Bytes
97420da
 
 
 
7c3be27
 
 
 
97420da
 
 
 
 
7c3be27
 
 
 
97420da
 
 
 
 
 
 
 
 
 
 
 
 
ec9fdfe
97420da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7c3be27
 
97420da
7c3be27
 
 
 
 
 
97420da
 
 
 
7c3be27
97420da
7c3be27
 
97420da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7c3be27
97420da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7c3be27
 
 
97420da
7c3be27
97420da
 
 
 
7c3be27
97420da
7c3be27
97420da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7c3be27
97420da
 
7c3be27
 
 
 
97420da
 
 
7c3be27
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
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
# cluster_news.py
# Clusters news articles using HDBSCAN, labels clusters with TF-IDF n-grams and LDA topics,
# and falls back to a representative summary if the label is too vague.

import numpy as np
import pandas as pd
from collections import defaultdict
from sentence_transformers import SentenceTransformer
from sklearn.feature_extraction.text import TfidfVectorizer, CountVectorizer
from sklearn.metrics.pairwise import cosine_distances
from sklearn.decomposition import LatentDirichletAllocation
import hdbscan
import umap

def generate_embeddings(df, content_column):
    model = SentenceTransformer('all-MiniLM-L6-v2')
    embeddings = model.encode(df[content_column].tolist(), show_progress_bar=True)
    return np.array(embeddings)

def reduce_dimensions(embeddings, n_neighbors=10, min_dist=0.0, n_components=5, random_state=42):
    n_samples = embeddings.shape[0]
    if n_samples < 3:
        return embeddings
    n_components = min(max(2, n_components), n_samples - 2)
    n_neighbors = min(max(2, n_neighbors), n_samples - 1)
    reducer = umap.UMAP(
        n_neighbors=n_neighbors,
        min_dist=min_dist,
        n_components=n_components,
        random_state=random_state,
        n_jobs=1,
        metric='cosine'
    )
    reduced = reducer.fit_transform(embeddings)
    return reduced

def cluster_with_hdbscan(embeddings, min_cluster_size=2, min_samples=1):
    clusterer = hdbscan.HDBSCAN(
        min_cluster_size=min_cluster_size,
        min_samples=min_samples,
        metric='euclidean'
    )
    labels = clusterer.fit_predict(embeddings)
    return labels, clusterer

def extract_tfidf_labels(df, content_column, cluster_labels, top_n=6):
    grouped = defaultdict(list)
    for idx, label in enumerate(cluster_labels):
        if label == -1: continue
        grouped[label].append(df.iloc[idx][content_column])
    tfidf_labels = {}
    for cluster_id, texts in grouped.items():
        vectorizer = TfidfVectorizer(ngram_range=(1, 2), stop_words="english", max_features=50)
        tfidf_matrix = vectorizer.fit_transform(texts)
        avg_tfidf = tfidf_matrix.mean(axis=0).A1
        if len(avg_tfidf) == 0:
            tfidf_labels[cluster_id] = []
            continue
        top_indices = np.argsort(avg_tfidf)[::-1][:top_n]
        top_terms = [vectorizer.get_feature_names_out()[i] for i in top_indices]
        tfidf_labels[cluster_id] = top_terms
    return tfidf_labels

def lda_topic_modeling(texts, n_topics=1, n_words=6):
    vectorizer = CountVectorizer(stop_words='english', ngram_range=(1, 2), max_features=1000)
    X = vectorizer.fit_transform(texts)
    if X.shape[0] < n_topics:
        n_topics = max(1, X.shape[0])
    lda = LatentDirichletAllocation(n_components=n_topics, random_state=42)
    lda.fit(X)
    topic_words = []
    for topic_idx, topic in enumerate(lda.components_):
        top_indices = topic.argsort()[:-n_words - 1:-1]
        words = [vectorizer.get_feature_names_out()[i] for i in top_indices]
        topic_words.extend(words)
    return topic_words

def get_representative_summary(df, cluster_indices, embeddings, centroid):
    cluster_embs = embeddings[cluster_indices]
    dists = cosine_distances(cluster_embs, centroid.reshape(1, -1)).flatten()
    min_idx = np.argmin(dists)
    return df.iloc[cluster_indices[min_idx]]["summary"]

def label_clusters_hybrid(df, content_column, summary_column, cluster_labels, embeddings, tfidf_labels, lda_labels, vague_threshold=15):
    cluster_label_map = {}
    cluster_primary_topics = {}
    cluster_related_topics = {}
    for cluster_id in set(cluster_labels):
        if cluster_id == -1:
            continue
        topics = lda_labels.get(cluster_id, []) or tfidf_labels.get(cluster_id, [])
        topics = [t for t in topics if t]
        primary_topics = topics[:3]
        related_topics = topics[3:]
        label = ", ".join(primary_topics) if primary_topics else ""
        if not label or len(label) < vague_threshold:
            cluster_indices = np.where(cluster_labels == cluster_id)[0]
            centroid = embeddings[cluster_indices].mean(axis=0)
            rep_summary = get_representative_summary(df, cluster_indices, embeddings, centroid)
            label = rep_summary[:80] + "..." if len(rep_summary) > 80 else rep_summary
        cluster_label_map[cluster_id] = label
        cluster_primary_topics[cluster_id] = primary_topics
        cluster_related_topics[cluster_id] = related_topics
    return cluster_label_map, cluster_primary_topics, cluster_related_topics

def cluster_and_label_articles(
    df, 
    content_column="content", 
    summary_column="summary", 
    min_cluster_size=2, 
    min_samples=1, 
    n_neighbors=10, 
    min_dist=0.0, 
    n_components=5, 
    top_n=6,
    lda_n_topics=1,
    lda_n_words=6,
    vague_threshold=15
):
    if df.empty:
        return None

    min_cluster_size = max(2, min(min_cluster_size, len(df) // 2)) if len(df) < 20 else min_cluster_size

    embeddings = generate_embeddings(df, content_column)
    reduced_embeddings = reduce_dimensions(embeddings, n_neighbors, min_dist, n_components)
    cluster_labels, clusterer = cluster_with_hdbscan(reduced_embeddings, min_cluster_size, min_samples)
    df['cluster_id'] = cluster_labels

    tfidf_labels = extract_tfidf_labels(df, content_column, cluster_labels, top_n=top_n)

    lda_labels = {}
    for cluster_id in set(cluster_labels):
        if cluster_id == -1:
            continue
        cluster_texts = df[cluster_labels == cluster_id][content_column].tolist()
        if cluster_texts:
            topics = lda_topic_modeling(
                cluster_texts, n_topics=lda_n_topics, n_words=lda_n_words
            )
            lda_labels[cluster_id] = topics
        else:
            lda_labels[cluster_id] = []

    cluster_label_map, cluster_primary_topics, cluster_related_topics = label_clusters_hybrid(
        df, content_column, summary_column, cluster_labels, embeddings, tfidf_labels, lda_labels, vague_threshold=vague_threshold
    )

    df['cluster_label'] = [
        cluster_label_map.get(cid, "Noise/Other") if cid != -1 else "Noise/Other"
        for cid in cluster_labels
    ]
    df['lda_topics'] = [
        ", ".join(lda_labels.get(cid, [])) if cid != -1 else "" for cid in cluster_labels
    ]

    detected_topics = {
        label: {
            "size": int((df['cluster_label'] == label).sum())
        }
        for label in set(df['cluster_label']) if label != "Noise/Other"
    }

    return {
        "dataframe": df,
        "detected_topics": detected_topics,
        "number_of_clusters": len(detected_topics),
        "cluster_primary_topics": cluster_primary_topics,
        "cluster_related_topics": cluster_related_topics
    }