model_train.py

import os
import joblib
import jieba
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report

# === 配置 ===
MODEL_DIR = 'models'
if not os.path.exists(MODEL_DIR):
    os.makedirs(MODEL_DIR)

# === 1. 构造训练数据 (包含一些厦门大学特有的关键词) ===
# 标签: 1=积极, 0=中立, -1=消极
data_samples = [
    # 积极
    ("厦门大学是最美校园，芙蓉湖太漂亮了", 1),
    ("嘉庚风格的建筑真的很有特色，爱了爱了", 1),
    ("图书馆的学习氛围很好，空调也很足", 1),
    ("勤业餐厅的沙茶面味道一绝，推荐", 1),
    ("感谢志愿者同学，核酸检测很有秩序", 1),
    ("终于抢到选修课了，开心！", 1),
    ("白城沙滩的夕阳太美了", 1),

    # 中立
    ("请问翔安校区怎么去本部？", 0),
    ("明天上午有一节高数课", 0),
    ("谁有二手自行车出吗？", 0),
    ("关于国庆期间图书馆开放时间的通知", 0),
    ("刚刚看到一只猫在路边睡觉", 0),
    ("求问教务处的电话是多少", 0),

    # 消极
    ("游客太多了，根本没法在学校里骑车", -1),
    ("网速太慢了，选课系统根本进不去，崩溃", -1),
    ("宿舍停水了，真的很烦躁", -1),
    ("食堂的菜越来越贵，量还少", -1),
    ("这种形式主义的讲座能不能少一点", -1),
    ("外卖又被偷了，无语", -1),
    ("这学期的课表排得太满了，累死", -1)
]

# 简单扩充数据量以支持训练
X_raw, y_raw = zip(*data_samples)
X_raw = list(X_raw) * 50
y_raw = list(y_raw) * 50

print(f"正在准备训练数据，共 {len(X_raw)} 条样本...")


# === 2. 分词预处理 ===
def clean_text(text):
    words = jieba.cut(text)
    return " ".join(words)


X_processed = [clean_text(text) for text in X_raw]

# === 3. 训练模型 ===
print("开始训练朴素贝叶斯模型...")
X_train, X_test, y_train, y_test = train_test_split(X_processed, y_raw, test_size=0.2, random_state=42)

model = make_pipeline(
    TfidfVectorizer(token_pattern=r"(?u)\b\w+\b"),
    MultinomialNB(alpha=0.1)
)

model.fit(X_train, y_train)

# === 4. 评估与保存 ===
y_pred = model.predict(X_test)
print("模型评估报告：")
print(classification_report(y_test, y_pred, target_names=['消极', '中立', '积极']))

model_path = os.path.join(MODEL_DIR, 'sentiment_nb.pkl')
joblib.dump(model, model_path)
print(f"✅ 模型已保存至: {model_path}")