app.py

from transformers import pipeline
import gradio as gr
from nltk.sentiment.vader import SentimentIntensityAnalyzer
import nltk
nltk.download('vader_lexicon')

zero_shot_classifier = pipeline("zero-shot-classification" , model='roberta-large-mnli')

spam_detector = pipeline("text-classification", model="madhurjindal/autonlp-Gibberish-Detector-492513457")

issues = ["Misconduct" , "Negligence" , "Discrimination" , "Corruption" , "Violation of Rights" , "Inefficiency" ,
           "Unprofessional Conduct", "Response Time" , "Use of Firearms" , "Property Damage"]

apprecn = ["Tech-Savvy Staff" , "Co-operative Staff" , "Well-Maintained Premises" , "Responsive Staff"]

def spam_detection(input_text):
    
    return spam_detector(input_text)[0]['label'] == 'clean'

def sentiment_analysis(input_text):
    
    score = SentimentIntensityAnalyzer().polarity_scores(input_text)
    
    if score['neg']>score['pos']:
        
        return "Negative Feedback"
    
    elif score['neg']<score['pos']:
        
        return "Positive Feedback"
    
    else:
        
        return "Neutral Feedback"

def positive_zero_shot(input_text):
    
    return zero_shot_classifier(input_text , candidate_labels = apprecn , multi_label = False)['labels'][0]
    
    
def negative_zero_shot(input_text):
    
    return zero_shot_classifier(input_text , candidate_labels = issues , multi_label = False)['labels'][0]
    
def pipeline(input_text):
    
    if spam_detection(input_text):
        
        if sentiment_analysis(input_text) == "Positive Feedback":
            
            return "Positive Feedback" , positive_zero_shot(input_text)
        
        elif sentiment_analysis(input_text) == "Negative Feedback":
            
            return "Negative Feedback" , negative_zero_shot(input_text)
        
        else:
            
            return "Neutral Feedback" , ""
    else:
        return "Spam" , ""

iface = gr.Interface(fn = pipeline , inputs=['text'] , outputs=['text' , 'text'])
iface.launch(share=True)