file_loader.py

import os
from tqdm import tqdm
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings
# from langchain_google_genai import GoogleGenerativeAIEmbeddings

# Import từ helpers
from helpers import (
    list_docx_files,  # Lấy danh sách file .docx
    get_splits,  # Xử lý file docx thành splits
    get_json_splits_only,  # Xử lý file JSON (FAQ)  
    get_web_documents,  # Xử lý dữ liệu từ web
)

import json

def get_vectorstore():
    ### Xử lý tất cả các tài liệu và nhét vào database
    folder_path = "syllabus_nct_word_format/"
    docx_files = list_docx_files(folder_path)
    
    all_splits = []  # Khởi tạo danh sách lưu kết quả
    print("Feeding relevent websites' contents")
    # 
    with open('syllabus_nct_word_format/urls.txt', 'r') as f:
        base_urls = [line.strip() for line in f]
    # urls_list
    # base_urls =['https://fda.neu.edu.vn/hoi-nghi-khoa-hoc-cong-nghe-dai-hoc-kinh-te-quoc-dan-nam-2025/']
    # ['https://nct.neu.edu.vn/', 'https://fsf.neu.edu.vn/', 'https://mfe.neu.edu.vn/', 'https://mis.neu.edu.vn/', 'https://fda.neu.edu.vn/', 'https://khoathongke.neu.edu.vn/', 'https://fit.neu.edu.vn/']
    
    
    website_contents = get_web_documents(base_urls=base_urls)
    all_splits += website_contents
    
    print('Feeding .docx files')
    for i, file_path in enumerate(tqdm(docx_files, desc="Đang xử lý", unit="file")):
        output_json_path = f"output_{i}.json"
        splits = get_splits(file_path, output_json_path)
        all_splits += splits
        
    print('Feeding .json files')
    # Xử lý FAQ
    FAQ_path = "syllabus_nct_word_format/FAQ.json"
    FAQ_splits = get_json_splits_only(FAQ_path)
    all_splits += FAQ_splits

    FAQ_path = "syllabus_nct_word_format/FAQ2.json"
    FAQ_splits = get_json_splits_only(FAQ_path)
    all_splits += FAQ_splits
    
    # Lưu vào vectorstore với nhúng từ Google GenAI
    # embedding = GoogleGenerativeAIEmbeddings(model="models/text-embedding-004")
    print('Get embedding model /paraphrase-multilingual-MiniLM-L12-v2')
    embedding = HuggingFaceEmbeddings(model_name="VoVanPhuc/sup-SimCSE-VietNamese-phobert-base")
    print('Set vectorstore FAISS')
    vectorstore = FAISS.from_documents(documents=all_splits, embedding=embedding)
    return vectorstore