predict_review / app.py
prafulla2121's picture
Upload 4 files
04b70f1 verified
Raw History Blame Contribute Delete
1.67 kB
import gradio as gr
import torch
import onnxruntime as ort
import numpy as np
import pickle
import re
import spacy
from transformers import RobertaTokenizer
# Load metadata and model
with open('top_aspects.pkl', 'rb') as f:
TOP_ASPECTS = pickle.load(f)
ort_session = ort.InferenceSession('mtl_roberta.onnx')
tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
nlp = spacy.load('en_core_web_sm', disable=['parser', 'ner'])
def clean_text(text):
text = re.sub(r'<.*?>', '', text)
text = re.sub(r'[^a-zA-Z\s]', '', text.lower())
doc = nlp(text)
return " ".join([token.lemma_ for token in doc if not token.is_stop and len(token.text) > 2])
def predict(text):
cleaned = clean_text(text)
inputs = tokenizer(cleaned, return_tensors='np', max_length=128, padding='max_length', truncation=True)
ort_inputs = {'input_ids': inputs['input_ids'].astype(np.int64), 'attention_mask': inputs['attention_mask'].astype(np.int64)}
sent_logits, help_score, aspect_logits = ort_session.run(None, ort_inputs)
sentiment_idx = np.argmax(sent_logits)
sentiment_map = {0: '1 Star', 1: '2 Stars', 2: '3 Stars', 3: '4 Stars', 4: '5 Stars'}
asp_probs = 1 / (1 + np.exp(-aspect_logits))
detected = [TOP_ASPECTS[i] for i, p in enumerate(asp_probs[0]) if p > 0.3]
return sentiment_map[sentiment_idx], round(float(help_score[0][0]), 2), ", ".join(detected) if detected else "None"
iface = gr.Interface(
fn=predict,
inputs=gr.Textbox(lines=3),
outputs=[gr.Label(label='Sentiment'), gr.Number(label='Helpfulness'), gr.Textbox(label='Aspects')],
title='Amazon Review Intelligence'
)
iface.launch()