app.py

import gradio as gr
from model import DepressionClassifier
import torch
from huggingface_hub import hf_hub_download
import transformers
from transformers import BertModel, BertTokenizer
from PIL import Image
import requests
import os
import tweepy
import pandas as pd

consumer_key = os.environ["API_KEY"]
consumer_secret = os.environ["API_KEY_SECRET"]
access_token = os.environ["ACCESS_TOKEN"]
access_token_secret = os.environ["ACCESS_TOKEN_SECRET"]

auth = tweepy.OAuth1UserHandler(
  consumer_key, 
  consumer_secret, 
  access_token, 
  access_token_secret
)

api = tweepy.API(auth)


class_names = ['Not Depressed', 'Depressed']
pt_file = hf_hub_download(repo_id="liangc40/sentimental_analysis", filename="model.pt")

model = DepressionClassifier(len(class_names), 'bert-base-cased')
model.load_state_dict(torch.load(pt_file, map_location=torch.device('cpu')))
model.eval()


def analyse(text):
    #text = "I'm depressed"
    #model = model.to('cpu')
    tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
    encoding = tokenizer.encode_plus(text, max_length=32, add_special_tokens=True, # Add '[CLS]' and '[SEP]'
                                     return_token_type_ids=False,
                                     pad_to_max_length=True,
                                     return_attention_mask=True,
                                     return_tensors='pt')

    outputs = model(input_ids = encoding['input_ids'], attention_mask = encoding['attention_mask'])
    _, preds = torch.max(outputs, dim=1)
    face_url = "https://raw.githubusercontent.com/liangc40/ID2223_Sentimental_Analysis_Project/main/Image/"+ str(preds.cpu().detach().numpy()[0]) + ".png"
    img = Image.open(requests.get(face_url, stream=True).raw)
    #print(preds)
    return img

def tweets(search_words):
    label_list=['Non-depressed','Depressed']
    tweets = tweepy.Cursor(api.search_tweets, 
                                q=search_words, 
                                lang="en",
                                ).items(10)
    tweet_content = [i.text for i in tweets]
    pred = [i for i in analyse(tweet_content)]
    label = [label_list[j] for j in pred]
    df = pd.DataFrame(list(zip(tweet_content, label)),columns =['Ten tweets'+' on '+search_words, 'sentiment'])
    return df
    
with gr.Blocks() as demo:
    gr.Markdown("<h1><center>Sentiment Analysis with Fine-tuned BERT Model")

    inputs_text=gr.Textbox(placeholder='Type your text for which you want know the sentiment', label='Text')
    text_button = gr.Button('Analyse Sentiment')
    output_text_sentiment = gr.Image(type="pil")
    text_button.click(analyse, inputs = inputs_text, outputs = output_text_sentiment)
    
    search_words=gr.Textbox(placeholder='Type your key word to search the tweets', label='Text')
    text_button_tweet = gr.Button('Analyse Sentiment from 10 realated tweets')
    output_tweet_sentiment = gr.Textbox(placeholder='Sentiment for 10 realated tweets')
    text_button.click(tweets, inputs = search_words, outputs = output_tweet_sentiment)

if __name__ == "__main__":
    demo.launch()