app.py

import json
import os
import torch
from sentence_transformers import SentenceTransformer
import faiss
import numpy as np
import gradio as gr

def load_or_create_model_and_embeddings(model_name, data_file):
    model_path = os.path.join(output_dir, 'saved_model')
    embeddings_path = os.path.join(output_dir, 'corpus_embeddings.pt')
    if os.path.exists(model_path) and os.path.exists(embeddings_path):
        print("載入已保存的模型和嵌入...")
        model = SentenceTransformer(model_path)
        embeddings = torch.load(embeddings_path)
    else:
        model = SentenceTransformer(model_name)
        with open(data_file, 'r', encoding='utf-8') as f:
            data = json.load(f)
        texts = [item['text'] for item in data]
        embeddings = model.encode(texts, convert_to_tensor=True)
    return model, embeddings

# 設置參數
model_name = 'sentence-transformers/all-MiniLM-L6-v2'
data_file = 'labeled_cti_data.json'
output_dir = '.'

# 載入或創建模型和嵌入
model, embeddings= load_or_create_model_and_embeddings(model_name, data_file)

# 創建 Faiss 索引
dimension = embeddings.shape[1]
index = faiss.IndexFlatL2(dimension)
index.add(embeddings.cpu().numpy().astype('float32'))

def semantic_search(query, top_k=3):
    query_vector = model.encode([query], convert_to_tensor=True)
    distances, indices = index.search(query_vector.cpu().numpy().astype('float32'), top_k)
    results = []
    for i, idx in enumerate(indices[0]):
        results.append({
            'text': texts[idx],
            'similarity_score': 1 - distances[0][i] / 2
        })
    return results

def search_and_format(query):
    results = semantic_search(query)
    formatted_results = ""
    for i, result in enumerate(results, 1):
        formatted_results += f"{i}. 相似度分數: {result['similarity_score']:.4f}\n"
        formatted_results += f"   情一: {result['text']}\n\n"
    return formatted_results

# 創建Gradio界面
iface = gr.Interface(
    fn=search_and_format,
    inputs=gr.Textbox(lines=2, placeholder="輸入您的搜索查詢..."),
    outputs=gr.Textbox(lines=10),
    title="語義搜索",
    description="輸入查詢以搜索相關文本。將顯示前3個最相關的結果。"
)

# 啟動Gradio界面
iface.launch()