BERT用于心理健康日记：超越情感分析检测认知扭曲 | WellAlly康心伴

pip install transformers datasets torch pandas scikit-learn presidio-analyzer

# src/anonymize.py
from presidio_analyzer import AnalyzerEngine
from presidio_anonymizer import AnonymizerEngine

# 初始化分析器和脱敏器
analyzer = AnalyzerEngine()
anonymizer = AnonymizerEngine()

def anonymize_text(text: str) -> str:
    """
    分析并脱敏文本以移除PII。
    """
    # 分析文本以查找PII
    analyzer_results = analyzer.analyze(text=text, language='en')

    # 脱敏文本
    anonymized_result = anonymizer.anonymize(
        text=text,
        analyzer_results=analyzer_results
    )

    return anonymized_result.text

# 示例用法
journal_entry = "My name is Jane Doe, and I live in New York. I had a tough day at work."
anonymized_entry = anonymize_text(journal_entry)
print(f"原始文本: {journal_entry}")
print(f"脱敏后: {anonymized_entry}")

# 预期输出：
# 原始文本: My name is Jane Doe, and I live in New York. I had a tough day at work.
# 脱敏后: My name is <PERSON>, and I live in <LOCATION>. I had a tough day at work.

# src/data_preparation.py
import pandas as pd
from sklearn.preprocessing import MultiLabelBinarizer
from transformers import BertTokenizer
from datasets import Dataset

# 假设你有一个已脱敏的CSV文件
df = pd.read_csv('anonymized_mental_health_journals.csv')

# 预处理标签
df['labels'] = df['labels'].apply(eval) # 将列表的字符串表示转换为实际列表
mlb = MultiLabelBinarizer()
y = mlb.fit_transform(df['labels'])
labels = pd.DataFrame(y, columns=mlb.classes_)
label_list = list(mlb.classes_)

# 加载分词器
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')

# 创建Hugging Face数据集
hf_dataset = Dataset.from_pandas(pd.concat([df[['text']], labels], axis=1))

def tokenize_data(examples):
    # 分词文本
    tokenized_inputs = tokenizer(examples['text'], truncation=True, padding="max_length", max_length=128)

    # 创建作为浮点数列表的'labels'字段
    labels = [float(examples[label]) for label in label_list]
    tokenized_inputs["labels"] = labels

    return tokenized_inputs

# 应用分词
tokenized_dataset = hf_dataset.map(tokenize_data)

# src/train.py
from transformers import BertForSequenceClassification, Trainer, TrainingArguments
import torch

# 定义模型
model = BertForSequenceClassification.from_pretrained(
    'bert-base-uncased',
    num_labels=len(label_list),
    problem_type="multi_label_classification"
)

# 设置训练参数
training_args = TrainingArguments(
    output_dir='./results',
    num_train_epochs=3,
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    warmup_steps=500,
    weight_decay=0.01,
    logging_dir='./logs',
    logging_steps=10,
    evaluation_strategy="epoch",
    save_strategy="epoch",
    load_best_model_at_end=True,
)

# 初始化Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_dataset, # 假设已进行训练/测试分割
    eval_dataset=tokenized_dataset, # 替换为适当的验证集
)

# 开始训练
trainer.train()

# src/inference.py
from transformers import BertTokenizer, BertForSequenceClassification
import torch
from .anonymize import anonymize_text # 假设 anonymize.py 在同一包中

# 加载微调后的模型和分词器
model_path = './results/checkpoint-XXXX' # 最佳模型的路径
model = BertForSequenceClassification.from_pretrained(model_path)
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')

# 训练中的标签列表
label_list = ['Catastrophizing', 'Mind Reading', 'Overgeneralization', 'Self-blame'] # 示例

def classify_journal_entry(text: str):
    # 1. 脱敏文本
    anonymized_text = anonymize_text(text)

    # 2. 对脱敏文本进行分词
    inputs = tokenizer(anonymized_text, return_tensors="pt", truncation=True, padding=True, max_length=128)

    # 3. 获取模型预测
    with torch.no_grad():
        logits = model(**inputs).logits

    # 4. 将logits转换为概率并获取预测
    probabilities = torch.sigmoid(logits)
    predictions = (probabilities > 0.5).int()

    # 5. 将预测映射回标签
    predicted_labels = [label_list[i] for i, pred in enumerate(predictions[0]) if pred == 1]

    return predicted_labels

# 示例用法
new_entry = "我把项目搞砸了，现在公司里每个人都觉得我无能。"
identified_distortions = classify_journal_entry(new_entry)
print(f"识别到的认知扭曲: {identified_distortions}")

# 预期输出：
# 识别到的认知扭曲: ['Mind Reading', 'Overgeneralization']