Spaces:
Sleeping
Sleeping
text_to_summary
Browse files- app.py +169 -0
- requirements.txt +9 -0
app.py
ADDED
@@ -0,0 +1,169 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from collections import Counter
|
3 |
+
import tensorflow as tf
|
4 |
+
import keras
|
5 |
+
from tensorflow.keras.preprocessing import text,sequence
|
6 |
+
from tensorflow.keras.preprocessing.text import Tokenizer
|
7 |
+
|
8 |
+
import nltk
|
9 |
+
nltk.download('punkt')
|
10 |
+
from nltk.tokenize import word_tokenize
|
11 |
+
nltk.download('stopwords')
|
12 |
+
from nltk.corpus import stopwords
|
13 |
+
nltk.download('wordnet')
|
14 |
+
from nltk.stem import WordNetLemmatizer
|
15 |
+
|
16 |
+
from textblob import TextBlob, Blobber
|
17 |
+
from textblob_fr import PatternTagger, PatternAnalyzer
|
18 |
+
|
19 |
+
import spacy.cli
|
20 |
+
spacy.cli.download("fr_core_news_md")
|
21 |
+
|
22 |
+
import torch
|
23 |
+
import sentencepiece as spm
|
24 |
+
from transformers import CamembertTokenizer, CamembertModel
|
25 |
+
from nltk.tokenize import sent_tokenize
|
26 |
+
from sklearn.metrics.pairwise import cosine_similarity
|
27 |
+
|
28 |
+
|
29 |
+
# nombre de mots et de mots uniques
|
30 |
+
def number_words(text):
|
31 |
+
word = text.split()
|
32 |
+
return f'Nombre de mots : {len(word)}', f'Nombre de mots uniques : {len(Counter(word))}'
|
33 |
+
|
34 |
+
|
35 |
+
# polarité
|
36 |
+
def polarity(text):
|
37 |
+
tb = Blobber(pos_tagger=PatternTagger(), analyzer=PatternAnalyzer())
|
38 |
+
if tb(text).sentiment[0] < 0:
|
39 |
+
return f'La polarité de ce texte est {tb(text).sentiment[0]} : ce texte est plus négatif que positif'
|
40 |
+
elif tb(text).sentiment[0] > 0:
|
41 |
+
return f'La polarité de ce texte est {tb(text).sentiment[0]} : ce texte est plus positif que négatif'
|
42 |
+
else :
|
43 |
+
return f'La polarité de ce texte est {tb(text).sentiment[0]} : ce texte est neutre, pas plus négatif que positif'
|
44 |
+
|
45 |
+
|
46 |
+
# subjectivité
|
47 |
+
def subjectivity(text):
|
48 |
+
tb = Blobber(pos_tagger=PatternTagger(), analyzer=PatternAnalyzer())
|
49 |
+
if tb(text).sentiment[1] < 0.5:
|
50 |
+
return f'La subjectivité de ce texte est {tb(text).sentiment[1]} : ce texte est plus subjectif que factuel'
|
51 |
+
elif tb(text).sentiment[1] > 0.5:
|
52 |
+
return f'La subjectivité de ce texte est {tb(text).sentiment[1]} : ce texte est plus subjectif que factuel'
|
53 |
+
else :
|
54 |
+
return f'La subjectivité de ce texte est {tb(text).sentiment[1]} : ce texte est neutre, pas plus subjectif que factuel'
|
55 |
+
|
56 |
+
|
57 |
+
# mots clés
|
58 |
+
def keywords(text):
|
59 |
+
nlp = spacy.load("fr_core_news_md")
|
60 |
+
text2 = nlp(text)
|
61 |
+
text_keywords = [token.text for token in text2 if token.pos_== 'NOUN' or token.pos_== 'PROPN' or token.pos_== 'VERB']
|
62 |
+
counter_words = Counter(text_keywords)
|
63 |
+
most_freq_words = [word for word in counter_words.most_common(10)]
|
64 |
+
most_freq_words_p = []
|
65 |
+
for i in range(len(most_freq_words)):
|
66 |
+
mfwp = most_freq_words[i][0]
|
67 |
+
most_freq_words_p.append(mfwp)
|
68 |
+
return 'mots clés :', ', '.join(most_freq_words_p)
|
69 |
+
|
70 |
+
|
71 |
+
# summary1
|
72 |
+
def summary_1(text):
|
73 |
+
model = CamembertModel.from_pretrained('camembert-base')
|
74 |
+
tokenizer = CamembertTokenizer.from_pretrained('camembert-base')
|
75 |
+
|
76 |
+
## preprocessing
|
77 |
+
sentences = sent_tokenize(text)
|
78 |
+
tokenized_sentences = [tokenizer.encode(sent, add_special_tokens=True) for sent in sentences]
|
79 |
+
|
80 |
+
## padding, encoding
|
81 |
+
max_len = 0
|
82 |
+
for i in tokenized_sentences:
|
83 |
+
if len(i) > max_len:
|
84 |
+
max_len = len(i)
|
85 |
+
|
86 |
+
padded_sentences = []
|
87 |
+
for i in tokenized_sentences:
|
88 |
+
while len(i) < max_len:
|
89 |
+
i.append(0)
|
90 |
+
padded_sentences.append(i)
|
91 |
+
|
92 |
+
input_ids = torch.tensor(padded_sentences)
|
93 |
+
|
94 |
+
## embedding
|
95 |
+
with torch.no_grad():
|
96 |
+
last_hidden_states = model(input_ids)[0]
|
97 |
+
|
98 |
+
sentence_embeddings = []
|
99 |
+
for i in range(len(sentences)):
|
100 |
+
sentence_embeddings.append(torch.mean(last_hidden_states[i], dim=0).numpy())
|
101 |
+
|
102 |
+
## summarizing
|
103 |
+
similarity_matrix = cosine_similarity(sentence_embeddings)
|
104 |
+
|
105 |
+
num_sentences = 2
|
106 |
+
summary_sentences = []
|
107 |
+
for i in range(num_sentences):
|
108 |
+
sentence_scores = list(enumerate(similarity_matrix[i]))
|
109 |
+
sentence_scores = sorted(sentence_scores, key=lambda x: x[1], reverse=True)
|
110 |
+
summary_sentences.append(sentences[sentence_scores[1][0]])
|
111 |
+
|
112 |
+
summary1 = ' '.join(summary_sentences)
|
113 |
+
return summary1
|
114 |
+
|
115 |
+
|
116 |
+
# summary2
|
117 |
+
def summary_2(text):
|
118 |
+
nlp = spacy.load("fr_core_news_md")
|
119 |
+
text2 = nlp(text)
|
120 |
+
text_keywords = [token.text for token in text2 if token.pos_== 'NOUN' or token.pos_== 'PROPN']
|
121 |
+
|
122 |
+
counter_words = Counter(text_keywords)
|
123 |
+
most_freq_words = [word for word in counter_words.most_common(3)]
|
124 |
+
|
125 |
+
most_freq_words_p = []
|
126 |
+
for i in range(len(most_freq_words)):
|
127 |
+
mfwp = most_freq_words[i][0]
|
128 |
+
most_freq_words_p.append(mfwp)
|
129 |
+
|
130 |
+
sentences = sent_tokenize(text)
|
131 |
+
summary2 = []
|
132 |
+
|
133 |
+
for sent in sentences:
|
134 |
+
words_in_sentence = word_tokenize(sent)
|
135 |
+
common_words = set(words_in_sentence).intersection(most_freq_words)
|
136 |
+
|
137 |
+
if common_words:
|
138 |
+
summary2.append(sent)
|
139 |
+
|
140 |
+
return summary2
|
141 |
+
|
142 |
+
|
143 |
+
def analyze_text(text):
|
144 |
+
nb_mots = number_words(text)
|
145 |
+
polarite = polarity(text)
|
146 |
+
subjectivite = subjectivity(text)
|
147 |
+
mots_cles = keywords(text)
|
148 |
+
resume1 = summary_1(text)
|
149 |
+
resume2 = summary_2(text)
|
150 |
+
|
151 |
+
return nb_mots, polarite, subjectivite, mots_cles, resume1, resume2
|
152 |
+
|
153 |
+
|
154 |
+
st.title("Text Analysis and Summary")
|
155 |
+
text = st.text_area("Enter text here:")
|
156 |
+
|
157 |
+
if st.button("Analyze"):
|
158 |
+
if text:
|
159 |
+
nb_mots, polarite, subjectivite, mots_cles, resume1, resume2 = analyze_text(text)
|
160 |
+
|
161 |
+
st.write(f'Nombre de mots : {nb_mots}')
|
162 |
+
st.write(f'Polarité : {polarite}')
|
163 |
+
st.write(f'Subjectivité : {subjectivite}')
|
164 |
+
st.write(f'Mots clés : {", ".join(mots_cles)}')
|
165 |
+
st.write(f'Résumé 1 : {resume1}')
|
166 |
+
st.write(f'Résumé 2 :')
|
167 |
+
for sent in resume2:
|
168 |
+
st.write(sent)
|
169 |
+
|
requirements.txt
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
nltk
|
2 |
+
textblob
|
3 |
+
textblob-fr
|
4 |
+
sentencepiece
|
5 |
+
transformers
|
6 |
+
spacy
|
7 |
+
torch
|
8 |
+
scikit-learn
|
9 |
+
streamlit
|