TOM/build_topic_model_browser.py at master · AdrienGuille/TOM · GitHub

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
# coding: utf-8
import os
import shutil
import tom_lib.utils as utils
from flask import Flask, render_template
from tom_lib.nlp.topic_model import NonNegativeMatrixFactorization
from tom_lib.structure.corpus import Corpus

__author__ = "Adrien Guille"
__email__ = "adrien.guille@univ-lyon2.fr"

# Flask Web server
app = Flask(__name__, static_folder='browser/static', template_folder='browser/templates')

# Parameters
max_tf = 0.8
min_tf = 4
num_topics = 15
vectorization = 'tfidf'

# Load corpus
corpus = Corpus(source_file_path='input/egc_lemmatized.csv',
                language='french',
                vectorization=vectorization,
                max_relative_frequency=max_tf,
                min_absolute_frequency=min_tf)
print('corpus size:', corpus.size)
print('vocabulary size:', len(corpus.vocabulary))

# Infer topics
topic_model = NonNegativeMatrixFactorization(corpus=corpus)
topic_model.infer_topics(num_topics=num_topics)
topic_model.print_topics(num_words=10)

# Clean the data directory
if os.path.exists('browser/static/data'):
    shutil.rmtree('browser/static/data')
os.makedirs('browser/static/data')

# Export topic cloud
utils.save_topic_cloud(topic_model, 'browser/static/data/topic_cloud.json')

# Export details about topics
for topic_id in range(topic_model.nb_topics):
    utils.save_word_distribution(topic_model.top_words(topic_id, 20),
                                 'browser/static/data/word_distribution' + str(topic_id) + '.tsv')
    utils.save_affiliation_repartition(topic_model.affiliation_repartition(topic_id),
                                       'browser/static/data/affiliation_repartition' + str(topic_id) + '.tsv')
    evolution = []
    for i in range(2012, 2016):
        evolution.append((i, topic_model.topic_frequency(topic_id, date=i)))
    utils.save_topic_evolution(evolution, 'browser/static/data/frequency' + str(topic_id) + '.tsv')

# Export details about documents
for doc_id in range(topic_model.corpus.size):
    utils.save_topic_distribution(topic_model.topic_distribution_for_document(doc_id),
                                  'browser/static/data/topic_distribution_d' + str(doc_id) + '.tsv')

# Export details about words
for word_id in range(len(topic_model.corpus.vocabulary)):
    utils.save_topic_distribution(topic_model.topic_distribution_for_word(word_id),
                                  'browser/static/data/topic_distribution_w' + str(word_id) + '.tsv')

# Associate documents with topics
topic_associations = topic_model.documents_per_topic()

# Export per-topic author network
for topic_id in range(topic_model.nb_topics):
    utils.save_json_object(corpus.collaboration_network(topic_associations[topic_id]),
                           'browser/static/data/author_network' + str(topic_id) + '.json')


@app.route('/')
def index():
    return render_template('index.html',
                           topic_ids=range(topic_model.nb_topics),
                           doc_ids=range(corpus.size),
                           method=type(topic_model).__name__,
                           corpus_size=corpus.size,
                           vocabulary_size=len(corpus.vocabulary),
                           max_tf=max_tf,
                           min_tf=min_tf,
                           vectorization=vectorization,
                           num_topics=num_topics)


@app.route('/topic_cloud.html')
def topic_cloud():
    return render_template('topic_cloud.html',
                           topic_ids=range(topic_model.nb_topics),
                           doc_ids=range(corpus.size))


@app.route('/vocabulary.html')
def vocabulary():
    word_list = []
    for i in range(len(corpus.vocabulary)):
        word_list.append((i, corpus.word_for_id(i)))
    splitted_vocabulary = []
    words_per_column = int(len(corpus.vocabulary)/5)
    for j in range(5):
        sub_vocabulary = []
        for l in range(j*words_per_column, (j+1)*words_per_column):
            sub_vocabulary.append(word_list[l])
        splitted_vocabulary.append(sub_vocabulary)
    return render_template('vocabulary.html',
                           topic_ids=range(topic_model.nb_topics),
                           doc_ids=range(corpus.size),
                           splitted_vocabulary=splitted_vocabulary,
                           vocabulary_size=len(word_list))


@app.route('/topic/<tid>.html')
def topic_details(tid):
    ids = topic_associations[int(tid)]
    documents = []
    for document_id in ids:
        documents.append((corpus.title(document_id).capitalize(),
                          ', '.join(corpus.author(document_id)),
                          corpus.date(document_id), document_id))
    return render_template('topic.html',
                           topic_id=tid,
                           frequency=round(topic_model.topic_frequency(int(tid))*100, 2),
                           documents=documents,
                           topic_ids=range(topic_model.nb_topics),
                           doc_ids=range(corpus.size))


@app.route('/document/<did>.html')
def document_details(did):
    vector = topic_model.corpus.vector_for_document(int(did))
    word_list = []
    for a_word_id in range(len(vector)):
        word_list.append((corpus.word_for_id(a_word_id), round(vector[a_word_id], 3), a_word_id))
    word_list.sort(key=lambda x: x[1])
    word_list.reverse()
    documents = []
    for another_doc in corpus.similar_documents(int(did), 5):
        documents.append((corpus.title(another_doc[0]).capitalize(),
                          ', '.join(corpus.author(another_doc[0])),
                          corpus.date(another_doc[0]), another_doc[0], round(another_doc[1], 3)))
    return render_template('document.html',
                           doc_id=did,
                           words=word_list[:21],
                           topic_ids=range(topic_model.nb_topics),
                           doc_ids=range(corpus.size),
                           documents=documents,
                           authors=', '.join(corpus.author(int(did))),
                           year=corpus.date(int(did)),
                           short_content=corpus.title(int(did)))


@app.route('/word/<wid>.html')
def word_details(wid):
    documents = []
    for document_id in corpus.docs_for_word(int(wid)):
        documents.append((corpus.title(document_id).capitalize(),
                          ', '.join(corpus.author(document_id)),
                          corpus.date(document_id), document_id))
    return render_template('word.html',
                           word_id=wid,
                           word=topic_model.corpus.word_for_id(int(wid)),
                           topic_ids=range(topic_model.nb_topics),
                           doc_ids=range(corpus.size),
                           documents=documents)

if __name__ == '__main__':
    # Access the browser at http://localhost:2016/
    app.run(debug=True, host='localhost', port=2016)