app.py

from transformers import RobertaForSequenceClassification, AutoTokenizer
import torch
import docx2txt
import pandas as pd
import matplotlib.pyplot as plt
import openpyxl
from openpyxl.styles import Font, Color, PatternFill
from openpyxl.styles.colors import WHITE
import gradio as gr
import underthesea

# Load the model and tokenizer
senti_model = RobertaForSequenceClassification.from_pretrained("wonrax/phobert-base-vietnamese-sentiment")
senti_tokenizer = AutoTokenizer.from_pretrained("wonrax/phobert-base-vietnamese-sentiment", use_fast=False)


def segmentation(sentences):
    segmented_sentences = []
    for sentence in sentences:
        segmented_sentence = underthesea.word_tokenize(sentence)
        segmented_sentences.append(' '.join(segmented_sentence))
    return segmented_sentences


def analyze(sentence):
    input_ids = torch.tensor([senti_tokenizer.encode(sentence)])
    with torch.no_grad():
        out = senti_model(input_ids)
        results = out.logits.softmax(dim=-1).tolist()
        return results[0]


def read_file(docx):
    try:
        text = docx2txt.process(docx)
        lines = text.split('\n')
        lines = [line.strip() for line in lines]
        lines = [line for line in lines if line]
        return lines
    except Exception as e:
        print(f"Error reading file: {e}")


def process_file(docx):
    # Read the file and segment the sentences
    sentences = read_file(docx)
    segmented_sentences = segmentation(sentences)

    # Analyze the sentiment of each sentence
    results = []
    for sentence in segmented_sentences:
        results.append(analyze(sentence))

    # Create a DataFrame from the results
    df = pd.DataFrame(results, columns=['Negative', 'Neutral', 'Positive'])
    df['Text'] = read_file(docx)

    # Generate the pie chart and excel file
    pie_chart_name = generate_pie_chart(df)
    excel_file_path = generate_excel_file(df)

    return excel_file_path, pie_chart_name


def analyze_text(text, docx_file):
    if text:
        # Perform analysis on the text
        segmented_text = segmentation([text])
        results = []
        for sentence in segmented_text:
            results.append(analyze(sentence))

        df = pd.DataFrame(results, columns=['Negative', 'Neutral', 'Positive'])
        df['Text'] = [text]
        pie_chart_name = generate_pie_chart(df)
        excel_file_path = generate_excel_file(df)
        return excel_file_path, pie_chart_name

    elif docx_file:
        return process_file(docx_file.name)

    else:
        # No input provided
        return None


def generate_pie_chart(df):
    # Calculate the average scores
    neg_avg = df['Negative'].mean()
    neu_avg = df['Neutral'].mean()
    pos_avg = df['Positive'].mean()

    # Create a new DataFrame with the average scores
    avg_df = pd.DataFrame({'Sentiment': ['Negative', 'Neutral', 'Positive'],
                           'Score': [neg_avg, neu_avg, pos_avg]})

    # Set custom colors for the pie chart
    colors = ['#BDBDBD', '#87CEFA', '#9ACD32']

    # Create a pie chart showing the average scores
    plt.pie(avg_df['Score'], labels=avg_df['Sentiment'], colors=colors, autopct='%1.1f%%')
    plt.title('Average Scores by Sentiment')

    # Save the pie chart as an image file in the static folder
    pie_chart_name = 'pie_chart.png'
    plt.savefig(pie_chart_name)
    plt.close()

    return pie_chart_name


def generate_excel_file(df):
    # Create a new workbook and worksheet
    wb = openpyxl.Workbook()
    ws = wb.active

    # Add column headers to the worksheet
    headers = ['Negative', 'Neutral', 'Positive', 'Text']
    for col_num, header in enumerate(headers, 1):
        cell = ws.cell(row=1, column=col_num)
        cell.value = header
        cell.font = Font(bold=True)

    # Set up cell formatting for each sentiment
    fill_dict = {
        'Negative': PatternFill(start_color='BDBDBD', end_color='BDBDBD', fill_type='solid'),
        'Neutral': PatternFill(start_color='87CEFA', end_color='87CEFA', fill_type='solid'),
        'Positive': PatternFill(start_color='9ACD32', end_color='9ACD32', fill_type='solid')
    }

    # Loop through each row of the input DataFrame and write data to the worksheet
    for row_num, row_data in df.iterrows():
        # Calculate the highest score and corresponding sentiment for this row
        sentiment_cols = ['Negative', 'Neutral', 'Positive']
        scores = [row_data[col] for col in sentiment_cols]
        max_score = max(scores)
        max_index = scores.index(max_score)
        sentiment = sentiment_cols[max_index]

        # Write the data to the worksheet
        for col_num, col_data in enumerate(row_data, 1):
            cell = ws.cell(row=row_num + 2, column=col_num)
            cell.value = col_data
            if col_num in [1, 2, 3]:
                if col_data == max_score:
                    cell.fill = fill_dict[sentiment]
            if col_num == 4:
                fill = fill_dict[sentiment]
                font_color = WHITE if fill.start_color.rgb == 'BDBDBD' else Color('000000')
                cell.fill = fill
                cell.font = Font(color=font_color)
                if col_data == max_score:
                    cell.fill = fill_dict[sentiment]

    # Save the workbook
    excel_file_path = 'result.xlsx'
    wb.save(excel_file_path)

    return excel_file_path


inputs = [
    gr.Textbox(label="Nhập Văn Bản bằng Tiếng Việt để trải nghiệm ngay"),
    gr.File(label="Chọn Tệp File Word(docx) Bạn Muốn Phân Tích")
]

outputs = [
    gr.File(label="Kết Quả Phân Tích Excel"),
    gr.Image(type="filepath", label="Biểu đồ")
]

interface = gr.Interface(
    fn=analyze_text,
    inputs=inputs,
    outputs=outputs,
    title="Phân Tích Cảm xúc thông qua Hội Thoại bằng Tiếng Việt",
    allow_flagging="never"  # Disable flag button
)

if __name__ == "__main__":
    interface.launch()