app.py

import io
from pdfminer.converter import TextConverter
from pdfminer.pdfinterp import PDFPageInterpreter, PDFResourceManager
from pdfminer.pdfpage import PDFPage
import gradio as gr
import re
import groq
import os
from dotenv import load_dotenv
from sumy.parsers.html import HtmlParser
from sumy.parsers.plaintext import PlaintextParser
from sumy.nlp.tokenizers import Tokenizer
from sumy.summarizers.lsa import LsaSummarizer

load_dotenv(verbose=True)
client = groq.Client(api_key=os.environ.get("GROQ_API_KEY"))

# PDFからテキストを抽出する関数
def extract_text_from_pdf(pdf_path):
    resource_manager = PDFResourceManager()
    fake_file_handle = io.StringIO()
    converter = TextConverter(resource_manager, fake_file_handle)
    page_interpreter = PDFPageInterpreter(resource_manager, converter)

    with open(pdf_path, 'rb') as fh:
        for page in PDFPage.get_pages(fh, caching=True, check_extractable=True):
            page_interpreter.process_page(page)

        text = fake_file_handle.getvalue()

    converter.close()
    fake_file_handle.close()

    if text:
        return text

def split_text(pdf_text):
    full_text = pdf_text.strip()
    chunks = []
    while len(full_text) > 0:
        first_part = full_text[:5000]
        last_period_index = first_part.rfind(".")
        if last_period_index != -1:
            first_part = first_part[:last_period_index+1]
        
        chunks.append(first_part)
        full_text = full_text[len(first_part):]
    
    return chunks

# Gradioのインターフェース
def summarize_pdf(pdf_path):
    pdf_text = extract_text_from_pdf(pdf_path.name)
    print("pdftexts:",pdf_text)
    chunks = split_text(pdf_text)
    '''summaries = []
    for chunk in chunks:
        summary = f"日本語で要約してもらいます。\n{chunk}\n要約したものは\n・を必ず先頭につけて箇条書きにしてください。\n"
        summaries.append(summary)
    return "\n\n".join(summaries)'''

    parser = PlaintextParser.from_string(pdf_text, Tokenizer("japanese")) 
    #parser = PlaintextParser.from_file(pdf_path, Tokenizer("japanese"))  
    #parser = HtmlParser.from_url(url, Tokenizer("japanese"))
    summarizer = LsaSummarizer()
    summary = summarizer(parser.document, 10000)  # 5文で要約
        
    return "\n".join(str(sentence) for sentence in summary)

    '''
    # 分割されたテキストデータを要約して出力する
    summary = ""
    for i in range(len(chunks)):
        tmp = chunks[i]
        print("tmp:",tmp)

        completion = client.chat.completions.create(
            model="llama3-70b-8192",
            temperature=0,
            messages=[
                {"role": "system", "content": "貴方は優秀なアナリストです。"},
                {"role": "user", "content": tmp+"を日本語に要約してください。要約したものは\n・を必ず先頭につけて箇条書きにしてください。 必ず、日本語で答えてください。"}
            ],
        )
    
        print("trans:",completion.choices[0].message.content)

        # 要約結果をsummaryに追加する
        summary += re.sub('[\n.]', '', completion.choices[0].message.content+ '\n')
        summary += "\n" 

    return summary'''


def summarize_html(html_url):
    #html_text = extract_text_from_pdf(pdf_path.name)
    #print("pdftexts:",pdf_text)
    #chunks = split_text(pdf_text)
    

    #parser = PlaintextParser.from_string(pdf_text, Tokenizer("japanese")) 
    #parser = PlaintextParser.from_file(pdf_path, Tokenizer("japanese"))  
    print("myurl:",html_url)
    parser = HtmlParser.from_url(html_url, Tokenizer("japanese"))
    summarizer = LsaSummarizer()
    summary = summarizer(parser.document, 10000)  # 5文で要約
    print("summary:",summary)
        
    return "\n".join(str(sentence) for sentence in summary)

# GradioのBlockを設定
with gr.Blocks(css="footer {visibility: hidden;} #custom_button {width: 400px; margin: 0 auto; background-color: #E0E7FF;}", theme=gr.themes.Soft(), title="ハイブリッド・サマリー・エージェント") as smry:
    gr.HTML('''<div style="display: flex; justify-content: center; align-items: center; font-size: 20px; font-weight: bold; font-family: 'Noto Sans JP', 'Yu Gothic', 'ヒラギノ角ゴシック', 'メイリオ', sans-serif;">サマリー・エージェント</div>''')

    gr.Markdown("PDF SUMMARY")
    with gr.Row():
        pdf_input = gr.File(label="PDFファイルをアップロード")
    with gr.Row():
        pdf_output = gr.Textbox(label="要約結果", lines=20)

    pdf_input.change(fn=summarize_pdf, inputs=[pdf_input], outputs=[pdf_output])

    gr.Markdown("HTML SUMMARY")
    with gr.Row():
        html_input = gr.Textbox(label="URLを入力",value="https://jp.reuters.com/world/us/VTXXEXTCYRIJFM73HAWI6ZBFKM-2025-05-09/")
    with gr.Row():
        html_output = gr.Textbox(label="要約結果", lines=20)

    html_btn = gr.Button("HTML要約",elem_id="custom_button")
    html_btn.click(summarize_html, inputs=[html_input], outputs=[html_output])
    #html_input.change(fn=summarize_html, inputs=[html_input], outputs=[html_output])

smry.launch()