Spaces:
Paused
Paused
change the main language
Browse files
app.py
CHANGED
|
@@ -4,15 +4,19 @@ import gradio as gr
|
|
| 4 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| 5 |
from bertopic import BERTopic
|
| 6 |
from sentence_transformers import SentenceTransformer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
-
# --- 1. CẤU HÌNH ĐƯỜNG DẪN ---
|
| 9 |
ROBERTA_PATH = "roberta_base_checkpoint"
|
| 10 |
BERTOPIC_PATH = "bertopic_checkpoint"
|
| 11 |
|
| 12 |
-
# --- 2.
|
| 13 |
-
print("---
|
| 14 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 15 |
-
print(f"
|
| 16 |
|
| 17 |
# Load RoBERTa
|
| 18 |
tokenizer = AutoTokenizer.from_pretrained(ROBERTA_PATH)
|
|
@@ -23,12 +27,12 @@ roberta_model.eval()
|
|
| 23 |
embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
|
| 24 |
topic_model = BERTopic.load(BERTOPIC_PATH, embedding_model=embedding_model)
|
| 25 |
|
| 26 |
-
print("---
|
| 27 |
|
| 28 |
-
# --- 3. LOGIC
|
| 29 |
def analyze_review(text):
|
| 30 |
if not text.strip():
|
| 31 |
-
return "
|
| 32 |
|
| 33 |
# --- Sentiment Analysis (RoBERTa) ---
|
| 34 |
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=256).to(device)
|
|
@@ -38,61 +42,62 @@ def analyze_review(text):
|
|
| 38 |
probs = F.softmax(outputs.logits, dim=1)
|
| 39 |
pred_idx = torch.argmax(probs, dim=1).item()
|
| 40 |
|
|
|
|
| 41 |
sentiment_labels = ['Negative 😡', 'Neutral 😐', 'Positive 😊']
|
| 42 |
sentiment = sentiment_labels[pred_idx]
|
| 43 |
confidence = probs[0][pred_idx].item()
|
| 44 |
|
| 45 |
# --- Topic Modeling (BERTopic) ---
|
| 46 |
-
# BERTopic transform nhận vào 1 list
|
| 47 |
new_topics, _ = topic_model.transform([text])
|
| 48 |
topic_id = int(new_topics[0])
|
| 49 |
|
| 50 |
if topic_id != -1:
|
| 51 |
keywords_data = topic_model.get_topic(topic_id)
|
| 52 |
-
#
|
| 53 |
keywords_list = ", ".join([word for word, score in keywords_data[:5]])
|
| 54 |
else:
|
| 55 |
-
keywords_list = "Outlier (
|
| 56 |
|
| 57 |
return sentiment, f"{confidence:.2%}", topic_id, keywords_list
|
| 58 |
|
| 59 |
-
# --- 4.
|
| 60 |
-
with gr.Blocks(title="
|
| 61 |
-
gr.Markdown("# 📊
|
| 62 |
-
gr.Markdown("
|
| 63 |
|
| 64 |
with gr.Row():
|
| 65 |
with gr.Column():
|
| 66 |
input_text = gr.Textbox(
|
| 67 |
-
label="
|
| 68 |
-
placeholder="
|
| 69 |
lines=5
|
| 70 |
)
|
| 71 |
-
btn = gr.Button("
|
| 72 |
|
| 73 |
with gr.Column():
|
| 74 |
-
out_sentiment = gr.Label(label="
|
| 75 |
-
out_confidence = gr.Textbox(label="
|
| 76 |
with gr.Row():
|
| 77 |
out_topic_id = gr.Number(label="Topic ID")
|
| 78 |
-
out_keywords = gr.Textbox(label="
|
| 79 |
|
| 80 |
-
#
|
| 81 |
btn.click(
|
| 82 |
fn=analyze_review,
|
| 83 |
inputs=input_text,
|
| 84 |
outputs=[out_sentiment, out_confidence, out_topic_id, out_keywords]
|
| 85 |
)
|
| 86 |
|
| 87 |
-
#
|
| 88 |
gr.Examples(
|
| 89 |
examples=[
|
| 90 |
["The food was amazing, very delicious!"],
|
| 91 |
-
["Service was very slow and the waiter was rude."]
|
|
|
|
| 92 |
],
|
| 93 |
inputs=input_text
|
| 94 |
)
|
| 95 |
|
| 96 |
-
#
|
| 97 |
if __name__ == "__main__":
|
| 98 |
demo.launch()
|
|
|
|
| 4 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| 5 |
from bertopic import BERTopic
|
| 6 |
from sentence_transformers import SentenceTransformer
|
| 7 |
+
import os
|
| 8 |
+
|
| 9 |
+
# --- 1. CONFIGURATION ---
|
| 10 |
+
# Tránh xung đột luồng khi chạy tokenizer
|
| 11 |
+
os.environ["TOKENIZERS_PARALLELISM"] = "false"
|
| 12 |
|
|
|
|
| 13 |
ROBERTA_PATH = "roberta_base_checkpoint"
|
| 14 |
BERTOPIC_PATH = "bertopic_checkpoint"
|
| 15 |
|
| 16 |
+
# --- 2. INITIALIZATION (Runs once at startup) ---
|
| 17 |
+
print("--- System Initializing ---")
|
| 18 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 19 |
+
print(f"Device detected: {device.type.upper()}")
|
| 20 |
|
| 21 |
# Load RoBERTa
|
| 22 |
tokenizer = AutoTokenizer.from_pretrained(ROBERTA_PATH)
|
|
|
|
| 27 |
embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
|
| 28 |
topic_model = BERTopic.load(BERTOPIC_PATH, embedding_model=embedding_model)
|
| 29 |
|
| 30 |
+
print("--- All models loaded successfully! ---")
|
| 31 |
|
| 32 |
+
# --- 3. CORE LOGIC ---
|
| 33 |
def analyze_review(text):
|
| 34 |
if not text.strip():
|
| 35 |
+
return "Please enter some text", None, None, ""
|
| 36 |
|
| 37 |
# --- Sentiment Analysis (RoBERTa) ---
|
| 38 |
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=256).to(device)
|
|
|
|
| 42 |
probs = F.softmax(outputs.logits, dim=1)
|
| 43 |
pred_idx = torch.argmax(probs, dim=1).item()
|
| 44 |
|
| 45 |
+
# Matching common sentiment output formats
|
| 46 |
sentiment_labels = ['Negative 😡', 'Neutral 😐', 'Positive 😊']
|
| 47 |
sentiment = sentiment_labels[pred_idx]
|
| 48 |
confidence = probs[0][pred_idx].item()
|
| 49 |
|
| 50 |
# --- Topic Modeling (BERTopic) ---
|
|
|
|
| 51 |
new_topics, _ = topic_model.transform([text])
|
| 52 |
topic_id = int(new_topics[0])
|
| 53 |
|
| 54 |
if topic_id != -1:
|
| 55 |
keywords_data = topic_model.get_topic(topic_id)
|
| 56 |
+
# Extracting top 5 keywords
|
| 57 |
keywords_list = ", ".join([word for word, score in keywords_data[:5]])
|
| 58 |
else:
|
| 59 |
+
keywords_list = "Outlier (No specific topic detected)"
|
| 60 |
|
| 61 |
return sentiment, f"{confidence:.2%}", topic_id, keywords_list
|
| 62 |
|
| 63 |
+
# --- 4. GRADIO INTERFACE (WEB & API) ---
|
| 64 |
+
with gr.Blocks(title="Food Review Analysis Tool") as demo:
|
| 65 |
+
gr.Markdown("# 📊 Food Review Analysis System")
|
| 66 |
+
gr.Markdown("An integrated pipeline using **RoBERTa** for Sentiment and **BERTopic** for Topic Extraction.")
|
| 67 |
|
| 68 |
with gr.Row():
|
| 69 |
with gr.Column():
|
| 70 |
input_text = gr.Textbox(
|
| 71 |
+
label="Review Content",
|
| 72 |
+
placeholder="e.g., The pizza was delicious but the delivery took two hours!",
|
| 73 |
lines=5
|
| 74 |
)
|
| 75 |
+
btn = gr.Button("Analyze Now", variant="primary")
|
| 76 |
|
| 77 |
with gr.Column():
|
| 78 |
+
out_sentiment = gr.Label(label="Sentiment Prediction")
|
| 79 |
+
out_confidence = gr.Textbox(label="Confidence Score")
|
| 80 |
with gr.Row():
|
| 81 |
out_topic_id = gr.Number(label="Topic ID")
|
| 82 |
+
out_keywords = gr.Textbox(label="Main Keywords")
|
| 83 |
|
| 84 |
+
# Event binding
|
| 85 |
btn.click(
|
| 86 |
fn=analyze_review,
|
| 87 |
inputs=input_text,
|
| 88 |
outputs=[out_sentiment, out_confidence, out_topic_id, out_keywords]
|
| 89 |
)
|
| 90 |
|
| 91 |
+
# Examples for quick testing
|
| 92 |
gr.Examples(
|
| 93 |
examples=[
|
| 94 |
["The food was amazing, very delicious!"],
|
| 95 |
+
["Service was very slow and the waiter was rude."],
|
| 96 |
+
["While the truffle-infused risotto was a masterpiece, the deafening noise made it hard to enjoy."]
|
| 97 |
],
|
| 98 |
inputs=input_text
|
| 99 |
)
|
| 100 |
|
| 101 |
+
# Launch the app
|
| 102 |
if __name__ == "__main__":
|
| 103 |
demo.launch()
|