我们对BERT的中文问答系统47代码进行了一些调整，以支持根据不同的数据集类型（例如历史类或聊天类）训练和保存不同类型的模型。此外，我们还在加载模型时根据问题的类型自动选择合适的模型。以下是修订后的代码：
python
import os
import json
import jsonlines
import torch
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
from transformers import BertModel, BertTokenizer
import tkinter as tk
from tkinter import filedialog, messagebox, ttk
import logging
from difflib import SequenceMatcher
from datetime import datetime
import requests
from bs4 import BeautifulSoup
# 获取项目根目录
PROJECT_ROOT = os.path.dirname(os.path.abspath(__file__))
# 配置日志
LOGS_DIR = os.path.join(PROJECT_ROOT, 'logs')
os.makedirs(LOGS_DIR, exist_ok=True)
def setup_logging():
log_file = os.path.join(LOGS_DIR, datetime.now().strftime('%Y-%m-%d_%H-%M-%S_羲和.txt'))
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s',
handlers=[
logging.FileHandler(log_file),
logging.StreamHandler()
]
)
setup_logging()
# 数据集类
class XihuaDataset(Dataset):
def __init__(self, file_path, tokenizer, max_length=128):
self.tokenizer = tokenizer
self.max_length = max_length
self.data = self.load_data(file_path)
def load_data(self, file_path):
data = []
if file_path.endswith('.jsonl'):
with jsonlines.open(file_path) as reader:
for i, item in enumerate(reader):
try:
data.append(item)
except jsonlines.jsonlines.InvalidLineError as e:
logging.warning(f"跳过无效行 {i + 1}: {e}")
elif file_path.endswith('.json'):
with open(file_path, 'r') as f:
try:
data = json.load(f)
except json.JSONDecodeError as e:
logging.warning(f"跳过无效文件 {file_path}: {e}")
return data
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
item = self.data[idx]
question = item.get('question', '')
human_answer = item.get('human_answers', [''])[0]
chatgpt_answer = item.get('chatgpt_answers', [''])[0]
try:
inputs = self.tokenizer(question, return_tensors='pt', padding='max_length', truncation=True, max_length=self.max_length)
human_inputs = self.tokenizer(human_answer, return_tensors='pt', padding='max_length', truncation=True, max_length=self.max_length)
chatgpt_inputs = self.tokenizer(chatgpt_answer, return_tensors='pt', padding='max_length', truncation=True, max_length=self.max_length)
except Exception as e:
logging.warning(f"跳过无效项 {idx}: {e}")
return self.__getitem__((idx + 1) % len(self.data))
return {
'input_ids': inputs['input_ids'].squeeze(),
'attention_mask': inputs['attention_mask'].squeeze(),
'human_input_ids': human_inputs['input_ids'].squeeze(),
'human_attention_mask': human_inputs['attention_mask'].squeeze(),
'chatgpt_input_ids': chatgpt_inputs['input_ids'].squeeze(),
'chatgpt_attention_mask': chatgpt_inputs['attention_mask'].squeeze(),
'human_answer': human_answer,
'chatgpt_answer': chatgpt_answer
}
# 获取数据加载器
def get_data_loader(file_path, tokenizer, batch_size=8, max_length=128):
dataset = XihuaDataset(file_path, tokenizer, max_length)
return DataLoader(dataset, batch_size=batch_size, shuffle=True)
# 模型定义
class XihuaModel(torch.nn.Module):
def __init__(self, pretrained_model_name='F:/models/bert-base-chinese'):
super(XihuaModel, self).__init__()
self.bert = BertModel.from_pretrained(pretrained_model_name)
self.classifier = torch.nn.Linear(self.bert.config.hidden_size, 1)
def forward(self, input_ids, attention_mask):
outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask)
pooled_output = outputs.pooler_output
logits = self.classifier(pooled_output)
return logits
# 训练函数
def train(model, data_loader, optimizer, criterion, device, progress_var=None):
model.train()
total_loss = 0.0
num_batches = len(data_loader)
for batch_idx, batch in enumerate(data_loader):
try:
input_ids = batch['input_ids'].to(device)
attention_mask = batch['attention_mask'].to(device)
human_input_ids = batch['human_input_ids'].to(device)
human_attention_mask = batch['human_attention_mask'].to(device)
chatgpt_input_ids = batch['chatgpt_input_ids'].to(device)
chatgpt_attention_mask = batch['chatgpt_attention_mask'].to(device)
optimizer.zero_grad()
human_logits = model(human_input_ids, human_attention_mask)
chatgpt_logits = model(chatgpt_input_ids, chatgpt_attention_mask)
human_labels = torch.ones(human_logits.size(0), 1).to(device)
chatgpt_labels = torch.zeros(chatgpt_logits.size(0), 1).to(device)
loss = criterion(human_logits, human_labels) + criterion(chatgpt_logits, chatgpt_labels)
loss.backward()
optimizer.step()
total_loss += loss.item()
if progress_var:
progress_var.set((batch_idx + 1) / num_batches * 100)
except Exception as e:
logging.warning(f"跳过无效批次: {e}")
return total_loss / len(data_loader)
# 模型评估函数
def evaluate_model(model, data_loader, device):
model.eval()
correct = 0
total = 0
with torch.no_grad():
for batch in data_loader:
input_ids = batch['input_ids'].to(device)
attention_mask = batch['attention_mask'].to(device)
human_input_ids = batch['human_input_ids'].to(device)
human_attention_mask = batch['human_attention_mask'].to(device)
chatgpt_input_ids = batch['chatgpt_input_ids'].to(device)
chatgpt_attention_mask = batch['chatgpt_attention_mask'].to(device)
human_logits = model(human_input_ids, human_attention_mask)
chatgpt_logits = model(chatgpt_input_ids, chatgpt_attention_mask)
human_labels = torch.ones(human_logits.size(0), 1).to(device)
chatgpt_labels = torch.zeros(chatgpt_logits.size(0), 1).to(device)
human_correct = (torch.sigmoid(human_logits) > 0.5).float() == human_labels
chatgpt_correct = (torch.sigmoid(chatgpt_logits) > 0.5).float() == chatgpt_labels
correct += human_correct.sum().item() + chatgpt_correct.sum().item()
total += human_labels.size(0) + chatgpt_labels.size(0)
accuracy = correct / total
return accuracy
# 网络搜索函数
def search_baidu(query):
url = f"https://www.baidu.com/s?wd={query}"
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.3'
}
response = requests.get(url, headers=headers)
soup = BeautifulSoup(response.text, 'html.parser')
results = soup.find_all('div', class_='c-abstract')
if results:
return results[0].get_text().strip()
return "没有找到相关信息"
# 百度百科搜索函数
def search_baidu_baike(query):
url = f"https://baike.baidu.com/item/{query}"
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.3'
}
response = requests.get(url, headers=headers)
soup = BeautifulSoup(response.text, 'html.parser')
meta_description = soup.find('meta', attrs={'name': 'description'})
if meta_description:
return meta_description['content']
return "没有找到相关信息"
# 360百科搜索函数
def search_360_baike(query):
url = f"https://baike.so.com/doc/{query}.html"
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.3'
}
response = requests.get(url, headers=headers)
soup = BeautifulSoup(response.text, 'html.parser')
# 查找 meta description 标签
meta_description = soup.find('meta', attrs={'name': 'description'})
if meta_description:
return meta_description['content']
# 如果 meta description 不存在，查找正文内容
main_content = soup.find('div', class_='lemma-main-content')
if main_content:
paragraphs = main_content.find_all('p')
if paragraphs:
return '\n'.join([p.get_text().strip() for p in paragraphs[:3]]) # 取前三个段落的内容
return "没有找到相关信息"
# GUI界面
class XihuaChatbotGUI:
def __init__(self, root):
self.root = root
self.root.title("羲和聊天机器人")
self.tokenizer = BertTokenizer.from_pretrained('F:/models/bert-base-chinese')
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
self.models = {}
self.current_model_type = None
self.load_models()
self.load_data()
# 历史记录
self.history = []
self.create_widgets()
def create_widgets(self):
# 设置样式
style = ttk.Style()
style.theme_use('clam')
# 顶部框架
top_frame = ttk.Frame(self.root)
top_frame.pack(pady=10)
self.question_label = ttk.Label(top_frame, text="问题:", font=("Arial", 12))
self.question_label.grid(row=0, column=0, padx=10)
self.question_entry = ttk.Entry(top_frame, width=50, font=("Arial", 12))
self.question_entry.grid(row=0, column=1, padx=10)
self.answer_button = ttk.Button(top_frame, text="获取回答", command=self.get_answer, style='TButton')
self.answer_button.grid(row=0, column=2, padx=10)
# 中部框架
middle_frame = ttk.Frame(self.root)
middle_frame.pack(pady=10)
self.chat_text = tk.Text(middle_frame, height=20, width=100, font=("Arial", 12), wrap='word')
self.chat_text.grid(row=0, column=0, padx=10, pady=10)
self.chat_text.tag_configure("user", justify='right', foreground='blue')
self.chat_text.tag_configure("xihua", justify='left', foreground='green')
# 底部框架
bottom_frame = ttk.Frame(self.root)
bottom_frame.pack(pady=10)
self.clear_button = ttk.Button(bottom_frame, text="清空聊天记录", command=self.clear_chat, style='TButton')
self.clear_button.grid(row=0, column=0, padx=10)
self.correct_button = ttk.Button(bottom_frame, text="准确", command=self.mark_correct, style='TButton')
self.correct_button.grid(row=0, column=1, padx=10)
self.incorrect_button = ttk.Button(bottom_frame, text="不准确", command=self.mark_incorrect, style='TButton')
self.incorrect_button.grid(row=0, column=2, padx=10)
self.train_button = ttk.Button(bottom_frame, text="训练模型", command=self.train_model, style='TButton')
self.train_button.grid(row=0, column=3, padx=10)
self.retrain_button = ttk.Button(bottom_frame, text="重新训练模型", command=lambda: self.train_model(retrain=True), style='TButton')
self.retrain_button.grid(row=0, column=4, padx=10)
self.progress_var = tk.DoubleVar()
self.progress_bar = ttk.Progressbar(bottom_frame, variable=self.progress_var, maximum=100, length=200, mode='determinate')
self.progress_bar.grid(row=1, column=0, columnspan=5, pady=10)
self.log_text = tk.Text(bottom_frame, height=10, width=70, font=("Arial", 12))
self.log_text.grid(row=2, column=0, columnspan=5, pady=10)
self.evaluate_button = ttk.Button(bottom_frame, text="评估模型", command=self.evaluate_model, style='TButton')
self.evaluate_button.grid(row=3, column=0, padx=10, pady=10)
self.history_button = ttk.Button(bottom_frame, text="查看历史记录", command=self.view_history, style='TButton')
self.history_button.grid(row=3, column=1, padx=10, pady=10)
self.save_history_button = ttk.Button(bottom_frame, text="保存历史记录", command=self.save_history, style='TButton')
self.save_history_button.grid(row=3, column=2, padx=10, pady=10)
def clear_chat(self):
self.chat_text.delete(1.0, tk.END)
def get_answer(self):
question = self.question_entry.get()
if not question:
messagebox.showwarning("输入错误", "请输入问题")
return
# 自动选择模型
model_type = self.detect_model_type(question)
self.select_model(model_type)
inputs = self.tokenizer(question, return_tensors='pt', padding='max_length', truncation=True, max_length=128)
with torch.no_grad():
input_ids = inputs['input_ids'].to(self.device)
attention_mask = inputs['attention_mask'].to(self.device)
logits = self.model(input_ids, attention_mask)
if logits.item() > 0:
answer_type = "羲和回答"
else:
answer_type = "零回答"
specific_answer = self.get_specific_answer(question, answer_type)
self.chat_text.insert(tk.END, f"用户: {question}\n", "user")
self.chat_text.insert(tk.END, f"羲和: {specific_answer}\n", "xihua")
# 添加到历史记录
self.history.append({
'question': question,
'answer_type': answer_type,
'specific_answer': specific_answer,
'accuracy': None, # 初始状态为未评价
'baidu_baike': None, # 初始状态为无百度百科结果
'360_baike': None # 初始状态为无360百科结果
})
def get_specific_answer(self, question, answer_type):
# 使用模糊匹配查找最相似的问题
best_match = None
best_ratio = 0.0
for item in self.data:
ratio = SequenceMatcher(None, question, item['question']).ratio()
if ratio > best_ratio:
best_ratio = ratio
best_match = item
if best_match:
if answer_type == "羲和回答":
return best_match['human_answers'][0]
else:
return best_match['chatgpt_answers'][0]
return "这个我也不清楚，你问问零吧"
def load_data(self):
self.data = self.load_data_from_file(os.path.join(PROJECT_ROOT, 'data/train_data.jsonl'))
def load_data_from_file(self, file_path):
data = []
if file_path.endswith('.jsonl'):
with jsonlines.open(file_path) as reader:
for i, item in enumerate(reader):
try:
data.append(item)
except jsonlines.jsonlines.InvalidLineError as e:
logging.warning(f"跳过无效行 {i + 1}: {e}")
elif file_path.endswith('.json'):
with open(file_path, 'r') as f:
try:
data = json.load(f)
except json.JSONDecodeError as e:
logging.warning(f"跳过无效文件 {file_path}: {e}")
return data
def load_models(self):
MODELS_DIR = os.path.join(PROJECT_ROOT, 'models')
for model_type in ['history', 'chat']:
model_path = os.path.join(MODELS_DIR, f'xihua_model_{model_type}.pth')
if os.path.exists(model_path):
model = XihuaModel(pretrained_model_name='F:/models/bert-base-chinese').to(self.device)
model.load_state_dict(torch.load(model_path, map_location=self.device))
self.models[model_type] = model
logging.info(f"加载 {model_type} 模型")
else:
logging.info(f"没有找到 {model_type} 模型，将使用预训练模型")
self.models[model_type] = XihuaModel(pretrained_model_name='F:/models/bert-base-chinese').to(self.device)
def select_model(self, model_type):
if model_type in self.models:
self.model = self.models[model_type]
self.current_model_type = model_type
logging.info(f"选择 {model_type} 模型")
else:
logging.warning(f"没有找到 {model_type} 模型，使用默认模型")
self.model = XihuaModel(pretrained_model_name='F:/models/bert-base-chinese').to(self.device)
self.current_model_type = None
def detect_model_type(self, question):
# 这里可以添加更复杂的逻辑来检测问题的类型
if "皇帝" in question or "朝代" in question:
return 'history'
return 'chat'
def train_model(self, retrain=False):
file_path = filedialog.askopenfilename(filetypes=[("JSONL files", "*.jsonl"), ("JSON files", "*.json")])
if not file_path:
messagebox.showwarning("文件选择错误", "请选择一个有效的数据文件")
return
model_type = self.detect_model_type(file_path)
self.select_model(model_type)
try:
dataset = XihuaDataset(file_path, self.tokenizer)
data_loader = DataLoader(dataset, batch_size=8, shuffle=True)
# 加载已训练的模型权重
if retrain:
model_path = os.path.join(PROJECT_ROOT, 'models', f'xihua_model_{model_type}.pth')
self.model.load_state_dict(torch.load(model_path, map_location=self.device))
self.model.to(self.device)
self.model.train()
optimizer = torch.optim.Adam(self.model.parameters(), lr=1e-5)
criterion = torch.nn.BCEWithLogitsLoss()
num_epochs = 30
best_loss = float('inf')
patience = 5
no_improvement_count = 0
for epoch in range(num_epochs):
train_loss = train(self.model, data_loader, optimizer, criterion, self.device, self.progress_var)
logging.info(f'Epoch [{epoch+1}/{num_epochs}], Loss: {train_loss:.10f}')
self.log_text.insert(tk.END, f'Epoch [{epoch+1}/{num_epochs}], Loss: {train_loss:.10f}\n')
self.log_text.see(tk.END)
if train_loss < best_loss:
best_loss = train_loss
no_improvement_count = 0
model_path = os.path.join(PROJECT_ROOT, 'models', f'xihua_model_{model_type}.pth')
torch.save(self.model.state_dict(), model_path)
logging.info("模型保存")
else:
no_improvement_count += 1
if no_improvement_count >= patience:
logging.info("早停机制触发，停止训练")
break
logging.info("模型训练完成并保存")
self.log_text.insert(tk.END, "模型训练完成并保存\n")
self.log_text.see(tk.END)
messagebox.showinfo("训练完成", "模型训练完成并保存")
except Exception as e:
logging.error(f"模型训练失败: {e}")
self.log_text.insert(tk.END, f"模型训练失败: {e}\n")
self.log_text.see(tk.END)
messagebox.showerror("训练失败", f"模型训练失败: {e}")
def evaluate_model(self):
test_data_loader = get_data_loader(os.path.join(PROJECT_ROOT, 'data/test_data.jsonl'), self.tokenizer, batch_size=8, max_length=128)
accuracy = evaluate_model(self.model, test_data_loader, self.device)
logging.info(f"模型评估准确率: {accuracy:.4f}")
self.log_text.insert(tk.END, f"模型评估准确率: {accuracy:.4f}\n")
self.log_text.see(tk.END)
messagebox.showinfo("评估结果", f"模型评估准确率: {accuracy:.4f}")
def mark_correct(self):
if self.history:
self.history[-1]['accuracy'] = True
messagebox.showinfo("评价成功", "您认为这次回答是准确的")
def mark_incorrect(self):
if self.history:
self.history[-1]['accuracy'] = False
question = self.history[-1]['question']
self.show_reference_options(question)
def show_reference_options(self, question):
reference_window = tk.Toplevel(self.root)
reference_window.title("参考答案")
reference_label = ttk.Label(reference_window, text="请选择参考答案来源:", font=("Arial", 12))
reference_label.pack(pady=10)
baidu_button = ttk.Button(reference_window, text="百度百科", command=lambda: self.get_reference_answer(question, 'baidu_baike'), style='TButton')
baidu_button.pack(pady=5)
so_button = ttk.Button(reference_window, text="360百科", command=lambda: self.get_reference_answer(question, '360_baike'), style='TButton')
so_button.pack(pady=5)
def get_reference_answer(self, question, source):
if source == 'baidu_baike':
baike_answer = self.search_baidu_baike(question)
self.chat_text.insert(tk.END, f"百度百科结果: {baike_answer}\n", "xihua")
self.history[-1]['baidu_baike'] = baike_answer
elif source == '360_baike':
baike_answer = self.search_360_baike(question)
self.chat_text.insert(tk.END, f"360百科结果: {baike_answer}\n", "xihua")
self.history[-1]['360_baike'] = baike_answer
messagebox.showinfo("参考答案", f"已获取{source}的结果")
def search_baidu_baike(self, query):
return search_baidu_baike(query)
def search_360_baike(self, query):
url = f"https://baike.so.com/doc/{query}.html"
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.3'
}
response = requests.get(url, headers=headers)
soup = BeautifulSoup(response.text, 'html.parser')
# 查找 meta description 标签
meta_description = soup.find('meta', attrs={'name': 'description'})
if meta_description:
return meta_description['content']
# 如果 meta description 不存在，查找正文内容
main_content = soup.find('div', class_='lemma-main-content')
if main_content:
paragraphs = main_content.find_all('p')
if paragraphs:
return '\n'.join([p.get_text().strip() for p in paragraphs[:3]]) # 取前三个段落的内容
return "没有找到相关信息"
def view_history(self):
history_window = tk.Toplevel(self.root)
history_window.title("历史记录")
history_text = tk.Text(history_window, height=20, width=80, font=("Arial", 12))
history_text.pack(padx=10, pady=10)
for entry in self.history:
history_text.insert(tk.END, f"问题: {entry['question']}\n")
history_text.insert(tk.END, f"回答类型: {entry['answer_type']}\n")
history_text.insert(tk.END, f"具体回答: {entry['specific_answer']}\n")
if entry['accuracy'] is None:
history_text.insert(tk.END, "评价: 未评价\n")
elif entry['accuracy']:
history_text.insert(tk.END, "评价: 准确\n")
else:
history_text.insert(tk.END, "评价: 不准确\n")
if entry['baidu_baike']:
history_text.insert(tk.END, f"百度百科结果: {entry['baidu_baike']}\n")
if entry['360_baike']:
history_text.insert(tk.END, f"360百科结果: {entry['360_baike']}\n")
history_text.insert(tk.END, "-" * 50 + "\n")
def save_history(self):
RECORDS_DIR = os.path.join(PROJECT_ROOT, 'records')
os.makedirs(RECORDS_DIR, exist_ok=True)
file_name = datetime.now().strftime('%Y-%m-%d_%H-%M-%S.txt')
file_path = os.path.join(RECORDS_DIR, file_name)
with open(file_path, 'w', encoding='utf-8') as f:
for entry in self.history:
f.write(f"用户: {entry['question']}\n")
f.write(f"羲和: {entry['specific_answer']}\n")
if entry['baidu_baike']:
f.write(f"百度百科结果: {entry['baidu_baike']}\n")
if entry['360_baike']:
f.write(f"360百科结果: {entry['360_baike']}\n")
f.write("-" * 50 + "\n")
# 保存为JSON格式
json_records = []
for entry in self.history:
record = {
"question": entry['question'],
"human_answers": [entry['specific_answer']] if entry['answer_type'] == "羲和回答" else [],
"chatgpt_answers": [entry['specific_answer']] if entry['answer_type'] == "零回答" else [],
"baidu_baike": entry['baidu_baike'],
"360_baike": entry['360_baike']
}
json_records.append(record)
json_file_name = datetime.now().strftime('%Y-%m-%d_%H-%M-%S.json')
json_file_path = os.path.join(RECORDS_DIR, json_file_name)
with open(json_file_path, 'w', encoding='utf-8') as f:
json.dump(json_records, f, ensure_ascii=False, indent=4)
messagebox.showinfo("保存成功", f"历史记录已保存到 {file_path} 和 {json_file_path}")
# 主函数
if __name__ == "__main__":
# 启动GUI
root = tk.Tk()
app = XihuaChatbotGUI(root)
root.mainloop()主要修改点：
1.多模型支持：增加了 self.models 字典来存储不同类型的模型，并在 load_models 方法中加载这些模型。
2.自动选择模型：在 get_answer 方法中，通过 detect_model_type 方法自动选择合适的模型。
3.模型训练：在 train_model 方法中，根据数据集的类型选择对应的模型进行训练，并保存为特定的文件名。
4.模型评估：在 evaluate_model 方法中，评估当前选择的模型。
这样，您可以根据不同的数据集类型训练和保存不同的模型，并在使用时自动选择合适的模型进行回答