app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import pandas as pd
import numpy as np
import googleapiclient.discovery
from dotenv import load_dotenv
import os

load_dotenv()
api_key = os.getenv("DEVELOPER_KEY")
# Initialize the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(
    "nlptown/bert-base-multilingual-uncased-sentiment"
)
model = AutoModelForSequenceClassification.from_pretrained(
    "nlptown/bert-base-multilingual-uncased-sentiment"
)


# Function to get comments from a YouTube video
def get_comments(youtube, **kwargs):
    comments = []
    results = youtube.commentThreads().list(**kwargs).execute()

    while results:
        for item in results["items"]:
            comment = item["snippet"]["topLevelComment"]["snippet"]["textDisplay"]
            comments.append(comment)

        # Check if there are more comments
        if "nextPageToken" in results:
            kwargs["pageToken"] = results["nextPageToken"]
            results = youtube.commentThreads().list(**kwargs).execute()
        else:
            break

    return comments


# Function to get sentiment score
def sentiment_score(comment):
    sentiment = model(tokenizer.encode(comment, return_tensors="pt"))
    return torch.argmax(sentiment.logits).item() + 1


# Gradio function to analyze video comments
def analyze_video(video_id):

    video_id = video_id.split("v=")[1]
    youtube = googleapiclient.discovery.build("youtube", "v3", developerKey=api_key)
    comments = get_comments(
        youtube, part="snippet", videoId=video_id, textFormat="plainText"
    )

    if not comments:
        return "No comments found."

    df = pd.DataFrame(np.array(comments[:200]), columns=["comments"])
    print(df.head())
    df["sentiment"] = df["comments"].apply(lambda x: sentiment_score(x[:512]))
    print(df.head())

    sentiment_counts = df["sentiment"].value_counts().sort_index()
    sentiment_dict = {
        f"Sentiment {index}": count for index, count in sentiment_counts.items()
    }

    # Returning as a dictionary, where the key is the sentiment score (1-5) and value is the count
    return sentiment_dict


# Creating the Gradio Interface
gr_interface = gr.Interface(
    fn=analyze_video,
    inputs=[gr.Textbox(label="YouTube Video ID")],
    outputs="json",
    title="YouTube Comment Sentiment Analysis",
    description="Input a YouTube video ID and your API key to analyze the sentiment of the comments.",
)

# Launch the Gradio app
gr_interface.launch()