app.py

import gradio as gr
from transformers import pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import train_test_split
from sklearn import metrics
import pandas as pd

# Load the provided dataset
file_path = 'data.csv'
df = pd.read_csv(file_path)

# Split data into training and test sets
X_train, X_test, y_train, y_test = train_test_split(df['Sentence'], df['Sentiment'], test_size=0.2, random_state=42)

# Define models
nb_model = make_pipeline(TfidfVectorizer(), MultinomialNB())
svm_model = make_pipeline(TfidfVectorizer(), SVC(probability=True))
rf_model = make_pipeline(TfidfVectorizer(), RandomForestClassifier())

# Train models
nb_model.fit(X_train, y_train)
svm_model.fit(X_train, y_train)
rf_model.fit(X_train, y_train)

# Define sentences to choose from
sentences = [
    "The announced restructuring will significantly decrease the company's indebtedness.",
    "UPM-Kymmene upgraded to `in-line' from `underperform' by Goldman Sachs.",
    "$AAPL shares are breaking out of the recent resistance level.",
    "Profitability (in EBIT %) was 13.6%, compared to 14.3% in Q2 2009.",
    "The Finnish bank has issued a profit warning.",
    "TeliaSonera's underlying results however included 457 mln SKr in positive one-offs, hence the adjusted underlying EBITDA actually amounts to 7.309 bln SKr, clearly below expectations, analysts said."
]

# Function to map BERT labels
def map_bert_label(label):
    if label in ["1 star", "2 stars"]:
        return "negative"
    elif label == "3 stars":
        return "neutral"
    elif label in ["4 stars", "5 stars"]:
        return "positive"


# Function to map RoBERTa labels
def map_roberta_label(label):
    label_mapping = {"LABEL_0": "negative", "LABEL_1": "neutral", "LABEL_2": "positive"}
    return label_mapping[label]


# Function to analyze sentiment
def analyze_sentiment(sentence):
    # Define model paths
    model_paths = {
        "FinBert": "ProsusAI/finbert",
        "BERT": "nlptown/bert-base-multilingual-uncased-sentiment",
        "RoBERTa": "cardiffnlp/twitter-roberta-base-sentiment"
    }

    # Analyze sentiment using transformers models
    results = {}
    for model_name, model_path in model_paths.items():
        sentiment_analyzer = pipeline("sentiment-analysis", model=model_path)
        result = sentiment_analyzer(sentence[:512])[0]  # Analyze first 512 characters for brevity

        if model_name == "BERT":
            result['label'] = map_bert_label(result['label'])
        elif model_name == "RoBERTa":
            result['label'] = map_roberta_label(result['label'])

        results[model_name] = result

    # Analyze sentiment using sklearn models
    results["Naive Bayes"] = {"label": nb_model.predict([sentence])[0],
                              "score": nb_model.predict_proba([sentence]).max()}
    results["SVM"] = {"label": svm_model.predict([sentence])[0], "score": svm_model.predict_proba([sentence]).max()}
    results["Random Forest"] = {"label": rf_model.predict([sentence])[0],
                                "score": rf_model.predict_proba([sentence]).max()}

    return sentence, results


# Create Gradio interface
dropdown = gr.Dropdown(choices=sentences, label="Select Sentence")
text_output = gr.Textbox(label="Selected Sentence", lines=2)
sentiment_output = gr.JSON(label="Sentiment Scores")

gr.Interface(
    fn=analyze_sentiment,
    inputs=[dropdown],
    outputs=[text_output, sentiment_output],
    title="Compare Sentiment Analysis Across Models",
    description="Select a sentence to see sentiment analysis results from multiple models."
).launch(share=True)