ilsa15 commited on
Commit
fa243e5
·
verified ·
1 Parent(s): 2ccf304

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +53 -233
app.py CHANGED
@@ -1,15 +1,15 @@
1
- import streamlit as st
2
- import pandas as pd
3
- import matplotlib.pyplot as plt
4
- import seaborn as sns
5
- import nltk
6
  import re
7
  import string
8
 
 
 
 
 
 
 
 
9
  from nltk.corpus import stopwords
10
  from nltk.stem import WordNetLemmatizer
11
- from sklearn.model_selection import train_test_split
12
- from sklearn.pipeline import Pipeline
13
  from sklearn.feature_extraction.text import TfidfVectorizer
14
  from sklearn.linear_model import LogisticRegression
15
  from sklearn.metrics import (
@@ -20,37 +20,63 @@ from sklearn.metrics import (
20
  precision_score,
21
  recall_score,
22
  )
 
 
23
 
24
- st.set_page_config(
25
- page_title="Multi-Class Sentiment Classifier",
26
- page_icon="💬",
27
- layout="wide",
28
- )
29
 
30
- @st.cache_resource
31
- def download_nltk_data():
32
- for package in ("stopwords", "wordnet", "omw-1.4"):
33
  nltk.download(package, quiet=True)
34
- return set(stopwords.words("english")), WordNetLemmatizer()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
35
 
36
- STOP_WORDS, LEMMATIZER = download_nltk_data()
37
 
38
  def preprocess_text(text):
39
  text = str(text).lower()
40
- text = re.sub(r"http\S+|www\.\S+", " ", text)
41
  text = re.sub(r"\S+@\S+", " ", text)
42
  text = re.sub(r"\d+", " ", text)
43
  text = text.translate(str.maketrans("", "", string.punctuation))
44
  words = re.findall(r"\b[a-z]+\b", text)
45
- words = [
46
- LEMMATIZER.lemmatize(word)
47
- for word in words
48
- if word not in STOP_WORDS and len(word) > 1
49
- ]
50
- return " ".join(words)
51
 
52
- @st.cache_data
53
- def sample_dataset():
 
 
 
 
 
 
 
 
 
 
 
 
 
54
  examples = {
55
  "positive": [
56
  "I absolutely love this product, it works wonderfully",
@@ -67,210 +93,4 @@ def sample_dataset():
67
  "Beautiful design and impressive performance",
68
  "I enjoyed every moment of the event",
69
  "The results are better than I expected",
70
- "Excellent quality, friendly service, and fair price",
71
- "This book is inspiring and well written",
72
- "The update made the software much better",
73
- "I am satisfied with my purchase",
74
- "A lovely place with a welcoming atmosphere",
75
- "The course is helpful and easy to understand",
76
- "The new feature saves time and works well",
77
- "The staff were polite and made us feel welcome",
78
- "The trip was comfortable and memorable",
79
- "The headphones sound clear and feel comfortable",
80
- "I would happily buy this again",
81
- ],
82
- "negative": [
83
- "I hate this product, it is terrible",
84
- "The service was awful and the staff were rude",
85
- "This was a disappointing experience",
86
- "The movie was boring and poorly made",
87
- "I am unhappy with the quality",
88
- "The food tasted stale and unpleasant",
89
- "Nothing worked and I cannot recommend it",
90
- "The delivery was late and the package was damaged",
91
- "The support team ignored my problem",
92
- "I had a horrible day and feel upset",
93
- "The app is confusing and keeps crashing",
94
- "Poor design and disappointing performance",
95
- "I regret attending this event",
96
- "The results were much worse than expected",
97
- "Bad quality, rude service, and high price",
98
- "This book is dull and badly written",
99
- "The update made the software worse",
100
- "I am dissatisfied with my purchase",
101
- "A noisy place with an unpleasant atmosphere",
102
- "The course is confusing and not helpful",
103
- "The new feature wastes time and fails often",
104
- "The staff were impolite and unwelcoming",
105
- "The trip was uncomfortable and stressful",
106
- "The headphones sound poor and hurt my ears",
107
- "I would never buy this again",
108
- ],
109
- "neutral": [
110
- "The product arrived in a standard box",
111
- "The service started at nine in the morning",
112
- "I watched the movie yesterday",
113
- "The package contains three separate items",
114
- "The food was served on a white plate",
115
- "The app has five tabs on the main screen",
116
- "The delivery took two days",
117
- "The support team replied to my message",
118
- "I attended the event on Friday",
119
- "The book has twelve chapters",
120
- "The software update was released today",
121
- "I purchased the item last week",
122
- "The restaurant is located near the station",
123
- "The course includes six lessons",
124
- "The feature is available in the settings menu",
125
- "The staff asked me to wait outside",
126
- "The trip lasted around four hours",
127
- "The headphones come with a charging cable",
128
- "The meeting is scheduled for tomorrow",
129
- "The report contains several sections",
130
- "The website has a search bar",
131
- "The order includes a receipt",
132
- "The room has two windows",
133
- "The class begins at ten o'clock",
134
- "The device is available in black and white",
135
- ],
136
- }
137
- return pd.DataFrame(
138
- [{"text": text, "sentiment": label}
139
- for label, sentences in examples.items()
140
- for text in sentences]
141
- )
142
-
143
- @st.cache_resource
144
- def train_model(data):
145
- df = data.copy()
146
- df["clean_text"] = df["text"].apply(preprocess_text)
147
- X_train, X_test, y_train, y_test = train_test_split(
148
- df["clean_text"],
149
- df["sentiment"],
150
- test_size=0.25,
151
- random_state=42,
152
- stratify=df["sentiment"],
153
- )
154
- model = Pipeline([
155
- ("tfidf", TfidfVectorizer(ngram_range=(1, 2), sublinear_tf=True)),
156
- ("classifier", LogisticRegression(max_iter=1000, class_weight="balanced")),
157
- ])
158
- model.fit(X_train, y_train)
159
- predictions = model.predict(X_test)
160
- metrics = {
161
- "accuracy": accuracy_score(y_test, predictions),
162
- "f1": f1_score(y_test, predictions, average="macro", zero_division=0),
163
- "precision": precision_score(y_test, predictions, average="macro", zero_division=0),
164
- "recall": recall_score(y_test, predictions, average="macro", zero_division=0),
165
- "y_test": y_test,
166
- "predictions": predictions,
167
- "labels": sorted(df["sentiment"].unique()),
168
- "report": classification_report(
169
- y_test, predictions, output_dict=True, zero_division=0
170
- ),
171
- }
172
- return model, metrics
173
-
174
- st.title("💬 Intelligent Multi-Class Sentiment Classifier")
175
- st.write(
176
- "Classify text as **positive**, **negative**, or **neutral** using text "
177
- "preprocessing, TF-IDF features, and a Scikit-learn machine learning model."
178
- )
179
-
180
- with st.sidebar:
181
- st.header("About the model")
182
- st.write("**Algorithm:** Logistic Regression")
183
- st.write("**Text features:** TF-IDF with unigrams and bigrams")
184
- st.write("**Preprocessing:** lowercasing, punctuation and stop-word removal, lemmatization")
185
- st.info(
186
- "This demo trains on a small built-in sample dataset. It is intended for "
187
- "demonstration and coursework, not as a production-grade sentiment model."
188
- )
189
-
190
- data = sample_dataset()
191
- model, metrics = train_model(data)
192
-
193
- tab_predict, tab_evaluate, tab_data = st.tabs(
194
- ["🔎 Predict Sentiment", "📊 Model Evaluation", "🗂️ Dataset"]
195
- )
196
-
197
- with tab_predict:
198
- st.subheader("Try your own text")
199
- user_text = st.text_area(
200
- "Enter a sentence or review",
201
- placeholder="Example: The product is excellent and easy to use!",
202
- height=130,
203
- )
204
- if st.button("Analyze sentiment", type="primary"):
205
- if not user_text.strip():
206
- st.warning("Please enter some text first.")
207
- else:
208
- cleaned = preprocess_text(user_text)
209
- if not cleaned.strip():
210
- st.warning("After preprocessing, no useful words remain. Try a longer sentence.")
211
- else:
212
- prediction = model.predict([cleaned])[0]
213
- probabilities = model.predict_proba([cleaned])[0]
214
- probability_map = dict(zip(model.classes_, probabilities))
215
- icons = {"positive": "😊", "negative": "😞", "neutral": "😐"}
216
- st.markdown(f"### {icons.get(prediction, '💬')} Predicted sentiment: **{prediction.title()}**")
217
- st.progress(float(probability_map[prediction]))
218
- st.write(f"Model confidence for this class: **{probability_map[prediction]:.1%}**")
219
- st.caption("Confidence is the model's estimated probability, not a guarantee.")
220
- st.write("Class probabilities")
221
- probability_df = pd.DataFrame({
222
- "Sentiment": list(probability_map.keys()),
223
- "Probability": list(probability_map.values()),
224
- }).sort_values("Probability", ascending=False)
225
- st.bar_chart(probability_df.set_index("Sentiment"))
226
-
227
- with st.expander("See preprocessed text"):
228
- st.write(cleaned)
229
-
230
- with tab_evaluate:
231
- st.subheader("Held-out test set performance")
232
- st.caption("Metrics are calculated on a stratified test split of the built-in sample dataset.")
233
- c1, c2, c3, c4 = st.columns(4)
234
- c1.metric("Accuracy", f"{metrics['accuracy']:.1%}")
235
- c2.metric("Macro F1-score", f"{metrics['f1']:.3f}")
236
- c3.metric("Macro precision", f"{metrics['precision']:.3f}")
237
- c4.metric("Macro recall", f"{metrics['recall']:.3f}")
238
-
239
- st.markdown("#### Classification report")
240
- report_df = pd.DataFrame(metrics["report"]).T
241
- st.dataframe(report_df.round(3), use_container_width=True)
242
-
243
- st.markdown("#### Confusion matrix")
244
- cm = confusion_matrix(metrics["y_test"], metrics["predictions"], labels=metrics["labels"])
245
- fig, ax = plt.subplots(figsize=(6, 4))
246
- sns.heatmap(
247
- cm,
248
- annot=True,
249
- fmt="d",
250
- cmap="Blues",
251
- xticklabels=metrics["labels"],
252
- yticklabels=metrics["labels"],
253
- ax=ax,
254
- )
255
- ax.set_xlabel("Predicted label")
256
- ax.set_ylabel("True label")
257
- ax.set_title("Confusion Matrix")
258
- st.pyplot(fig)
259
- plt.close(fig)
260
-
261
- with tab_data:
262
- st.subheader("Sample dataset")
263
- st.write(f"Dataset contains **{len(data)} text examples** across three classes.")
264
- st.dataframe(data, use_container_width=True, hide_index=True)
265
- st.markdown("#### Number of examples per class")
266
- st.bar_chart(data["sentiment"].value_counts())
267
- csv = data.to_csv(index=False).encode("utf-8")
268
- st.download_button(
269
- "Download sample dataset as CSV",
270
- data=csv,
271
- file_name="sentiment_dataset.csv",
272
- mime="text/csv",
273
- )
274
-
275
- st.divider()
276
- st.caption("Task 2 • NLP Text Classification • Built with Python, Streamlit, NLTK and Scikit-learn")
 
 
 
 
 
 
1
  import re
2
  import string
3
 
4
+ import gradio as gr
5
+ import matplotlib
6
+ matplotlib.use("Agg")
7
+ import matplotlib.pyplot as plt
8
+ import pandas as pd
9
+ import nltk
10
+
11
  from nltk.corpus import stopwords
12
  from nltk.stem import WordNetLemmatizer
 
 
13
  from sklearn.feature_extraction.text import TfidfVectorizer
14
  from sklearn.linear_model import LogisticRegression
15
  from sklearn.metrics import (
 
20
  precision_score,
21
  recall_score,
22
  )
23
+ from sklearn.model_selection import train_test_split
24
+ from sklearn.pipeline import Pipeline
25
 
 
 
 
 
 
26
 
27
+ # Download the small NLTK resources required for preprocessing.
28
+ for package in ("stopwords", "wordnet", "omw-1.4"):
29
+ try:
30
  nltk.download(package, quiet=True)
31
+ except Exception:
32
+ pass
33
+
34
+ try:
35
+ STOP_WORDS = set(stopwords.words("english"))
36
+ except LookupError:
37
+ # Fallback list allows the app to start if the NLTK corpus download fails.
38
+ STOP_WORDS = {
39
+ "a", "an", "the", "and", "or", "but", "if", "while", "of", "at",
40
+ "by", "for", "with", "about", "against", "between", "into", "through",
41
+ "during", "before", "after", "above", "below", "to", "from", "up",
42
+ "down", "in", "out", "on", "off", "over", "under", "again", "further",
43
+ "then", "once", "is", "am", "are", "was", "were", "be", "been", "being",
44
+ "have", "has", "had", "do", "does", "did", "i", "you", "he", "she",
45
+ "it", "we", "they", "them", "this", "that", "these", "those", "my",
46
+ "your", "our", "their", "me", "him", "her", "as", "not", "so", "very",
47
+ }
48
+
49
+ try:
50
+ LEMMATIZER = WordNetLemmatizer()
51
+ # Check whether WordNet is available before processing user text.
52
+ LEMMATIZER.lemmatize("tests")
53
+ except LookupError:
54
+ LEMMATIZER = None
55
 
 
56
 
57
  def preprocess_text(text):
58
  text = str(text).lower()
59
+ text = re.sub(r"https?://\S+|www\.\S+", " ", text)
60
  text = re.sub(r"\S+@\S+", " ", text)
61
  text = re.sub(r"\d+", " ", text)
62
  text = text.translate(str.maketrans("", "", string.punctuation))
63
  words = re.findall(r"\b[a-z]+\b", text)
 
 
 
 
 
 
64
 
65
+ cleaned_words = []
66
+ for word in words:
67
+ if word in STOP_WORDS or len(word) <= 1:
68
+ continue
69
+ if LEMMATIZER is not None:
70
+ try:
71
+ word = LEMMATIZER.lemmatize(word)
72
+ except LookupError:
73
+ pass
74
+ cleaned_words.append(word)
75
+
76
+ return " ".join(cleaned_words)
77
+
78
+
79
+ def create_dataset():
80
  examples = {
81
  "positive": [
82
  "I absolutely love this product, it works wonderfully",
 
93
  "Beautiful design and impressive performance",
94
  "I enjoyed every moment of the event",
95
  "The results are better than I expected",
96
+ "Excellent quality, friendly service, and fair price",