app.py

# we will take last 8 messages as input and calculate the sentiment of each message
NUM_MESSAGES = 8

from transformers import pipeline
import gradio as gr


pipe = pipeline("text-classification", model="distilbert/distilbert-base-uncased-finetuned-sst-2-english")


def sentiment_analysis(*messages):
    """
    Input will be a list of messages.
    The function calculates the sentiment of each message, and then returns the average sentiment of the messages.
    while calculating the sentiment, also take positive and negative labels into account.
    scores are normalized to 0-100 range.
    """
    # return 0 if no messages are provided
    if len(messages) == 0:
        return 0

    if len(messages) > NUM_MESSAGES:
        messages = messages[-NUM_MESSAGES:]

    # each message should be of same length, so we will pad the messages
    # find longest message
    max_len = max([len(m) for m in messages])
    # pad each message to the length of the longest message
    messages = [m.ljust(max_len) for m in messages]

    output = pipe(messages)
    score = 0
    for i in range(len(output)):
        if output[i]['label'] == 'POSITIVE':
            score += output[i]['score']
        else:
            score -= output[i]['score']

    # shift score to 0-100 range
    score = (score + NUM_MESSAGES) * 50 / NUM_MESSAGES
    return round(score, 2)


demo = gr.Interface(
    fn=sentiment_analysis,
    inputs=["text"] * NUM_MESSAGES,
    outputs=["number"],
    title="Sentiment Analysis",
    description=f"Analyze the sentiment of the last {NUM_MESSAGES} messages"
)

demo.launch()