1
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

SentenceTransformersを使用したXML文書の類似検索のコード

Posted at

SentenceTransformersを使用したXML文書の類似検索のコードです。
XMLの指定したタグについて類似度検索とキーワード検索できるようにしています。
XMLの文書の入ったフォルダの指定と、XMLで検索したいタグを指定すれば動くはずです。
動作にはstreamlit run test.pyとすればブラウザで検索できます。

test.py
import streamlit as st
import os
import xml.etree.ElementTree as ET
import re  # 正規表現モジュールをインポート
from sentence_transformers import SentenceTransformer, util

# XMLファイルから特定の要素のテキスト内容を取得する関数
def get_xml_text_content(xml_file, element_name='<XMLで検索したいタグ>'):
    try:
        tree = ET.parse(xml_file)
        root = tree.getroot()
        text_content = root.find(f'.//{element_name}').text
        return text_content
    except Exception as e:
        return ""

# SentenceTransformersを使用した類似度検索
def search_similar_documents(keyword, xml_directory='<XMLのディレクトリ>'):
#好きなモデルを入れてください
    model = SentenceTransformer('paraphrase-xlm-r-multilingual-v1')
    xml_texts = []
    xml_files = []

    for filename in os.listdir(xml_directory):
        if filename.endswith('.xml'):
            file_path = os.path.join(xml_directory, filename)
            xml_text = get_xml_text_content(file_path)
            if xml_text:
                xml_texts.append(xml_text)
                xml_files.append(filename)

    results = []
    if xml_texts:
        # キーワードとXML文書の埋め込みを計算
        keyword_embedding = model.encode([keyword], convert_to_tensor=True)
        doc_embeddings = model.encode(xml_texts, convert_to_tensor=True)
        # コサイン類似度の計算
        cosine_scores = util.pytorch_cos_sim(keyword_embedding, doc_embeddings)[0]

        similarity_scores = list(enumerate(cosine_scores.tolist()))
        sorted_scores = sorted(similarity_scores, key=lambda x: x[1], reverse=True)[:10]
        for idx, score in sorted_scores:
            results.append((xml_files[idx], score, xml_texts[idx]))

    return results

# 部分一致検索機能は変更なし
def search_documents_by_content(keyword, xml_directory='<XMLのディレクトリ>'):
    results = []
    for filename in os.listdir(xml_directory):
        if filename.endswith('.xml'):
            file_path = os.path.join(xml_directory, filename)
            xml_text = get_xml_text_content(file_path)
            if keyword.lower() in xml_text.lower():
                results.append((filename, xml_text))
    return results[:10]

# Streamlitアプリケーションのメイン関数
def main():
    st.title("XML Search and Similarity Tool")
    
    # キーワード入力
    keyword = st.text_area("Enter keyword for search or similarity search:", height=150)
    
    # 類似度検索ボタン
    if st.button("Search Similar Documents"):
        if keyword:
            results = search_similar_documents(keyword)
            for filename, score, text in results:
                # ファイル名から数字を抽出
                match = re.search(r'\d+', filename)
                if match:
                    trouble_id = match.group()
                    st.write(f"{filename}: Score = {score:.4f} (No ID found)")
                    st.text_area("", text, height=300)
        else:
            st.error("Please enter a keyword.")
    
    # 部分一致検索ボタン
    if st.button("Search Documents by Content"):
        if keyword:
            results = search_documents_by_content(keyword)
            st.write(f"{filename} (No ID found)")
            st.text_area("", text, height=300)

if __name__ == "__main__":
    main()

1
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
1
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?