app.py

import tensorflow as tf
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
import opendatasets as od
import gradio as gr
import pandas as pd
import plotly.express as plt

tf.get_logger().setLevel("ERROR")

# create tokenizer from pre-trained model
tokenizer = AutoTokenizer.from_pretrained(
    "blanchefort/rubert-base-cased-sentiment-rurewiews"
)
# Load the model
model = AutoModelForSequenceClassification.from_pretrained(
    "blanchefort/rubert-base-cased-sentiment-rurewiews"
)
# Create a pipeline for the model
pipe = pipeline(
    "text-classification", model="blanchefort/rubert-base-cased-sentiment-rurewiews"
)

# load review from open dataset
od.download_kaggle_dataset("vigneshwarsofficial/reviews", data_dir="restaurent_review")
prediction_data = pd.read_csv(
    "restaurent_review/reviews/Restaurant_Reviews.tsv", delimiter="\t"
)
# popping irrelevant coloumn
prediction_data.pop("Liked")
# making a list
data = list(prediction_data["Review"])
# making prediction using pipe
results = pipe(data)

# Categorizing result
positive_counter = 0
negative_counter = 0
neutral_counter = 0
for x in results:
    if x["label"] == "POSITIVE":
        positive_counter = positive_counter + 1
    elif x["label"] == "NEGATIVE":
        negative_counter = negative_counter + 1
    else:
        neutral_counter = neutral_counter + 1

result_data = pd.DataFrame(
    {
        "count": [positive_counter, negative_counter, neutral_counter],
        "sentiment": ["Positive", "Negative", "Neutral"],
    }
)


# create bar chart interface on gradio
def plotly_plot():
    p = plt.bar(
        result_data,
        x="sentiment",
        y="count",
        title="Restaurent Review Analysis",
        color="count",
    )
    return p


# show the results
outputs = gr.Plot()

demo = gr.Interface(
    fn=plotly_plot,
    inputs=None,
    outputs=outputs,
    title="Restaurant Customer Review Sentiment Analysis",
)

demo.launch()