CVproject.py

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import re
import nltk
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.metrics import classification_report, accuracy_score
from sklearn.preprocessing import LabelEncoder
import os

# 確保NLTK資料已下載。如果首次運行，請取消註釋以下行並執行。
# nltk.download('stopwords')
# nltk.download('wordnet')
# nltk.download('punkt')

class ResumeScreener:
    """
    一個用於履歷篩選的類別，包含資料載入、預處理、特徵工程、
    模型訓練和評估等功能。
    """
    def __init__(self, data_path='SomeoneCV.csv', tfidf_max_features=5000):
        """
        初始化履歷篩選器。

        參數:
            data_path (str): 履歷資料集CSV檔案的路徑。
            tfidf_max_features (int): TF-IDF向量化器中最大特徵數。
        """
        self.data_path = data_path
        self.tfidf_max_features = tfidf_max_features
        self.df = None
        self.tfidf_vectorizer = None
        self.label_encoder = None
        self.model = None
        self.X_train = None
        self.X_test = None
        self.y_train = None
        self.y_test = None

        # 初始化NLTK工具
        self.stop_words = set(stopwords.words('english'))
        self.lemmatizer = WordNetLemmatizer()
        print("ResumeScreener 已初始化。")

    def load_data(self):
        """
        載入履歷資料集到DataFrame中。
        """
        try:
            self.df = pd.read_csv(self.data_path)
            print(f"資料 '{self.data_path}' 載入成功！")
            print("資料集形狀 (行, 列)：", self.df.shape)
            print("資料集前5行：\n", self.df.head())
        except FileNotFoundError:
            print(f"錯誤：找不到檔案在 {self.data_path}。請確保檔案路徑正確。")
            self.df = None # 確保df為None，以便後續檢查
            return False
        return True

    def visualize_category_distribution(self):
        """
        視覺化履歷類別的分佈。
        """
        if self.df is None:
            print("請先載入資料。")
            return

        print("\n各類別履歷數量：")
        print(self.df['Category'].value_counts())

        plt.figure(figsize=(15, 8))
        sns.countplot(y='Category', data=self.df, palette='viridis')
        plt.title('各類別履歷分佈')
        plt.xlabel('數量')
        plt.ylabel('類別')
        plt.show()

    def clean_resume_text(self, text):
        """
        清理單個履歷文本：
        - 移除URLs、RT、cc、哈希標籤、提及、標點符號、非ASCII字元和數字。
        - 轉換為小寫。
        - 移除停用詞。
        - 進行詞形還原。
        """
        text = re.sub('http\S+\s*', ' ', text)  # 移除URLs
        text = re.sub('RT|cc', ' ', text)  # 移除RT和cc
        text = re.sub('#\S+', '', text)  # 移除哈希標籤
        text = re.sub('@\S+', ' ', text)  # 移除提及
        text = re.sub('[%s]' % re.escape("""!"#$%&'()*+,-./:;<=>?@[\]^_`{|}~"""), ' ', text)  # 移除標點符號
        text = re.sub(r'[^\x00-\x7f]', r' ', text) # 移除非ASCII字符
        text = re.sub('\s+', ' ', text)  # 移除多餘的空格
        text = re.sub('\d+', '', text) # 移除數字
        text = text.lower() # 轉換為小寫

        # 移除停用詞並進行詞形還原
        text = ' '.join([self.lemmatizer.lemmatize(word) for word in text.split() if word not in self.stop_words])
        return text

    def preprocess_and_vectorize(self):
        """
        對履歷文本進行預處理（清理）和特徵化（TF-IDF向量化）。
        """
        if self.df is None:
            print("請先載入資料。")
            return None
        
        print("\n開始文本預處理和特徵化...")
        self.df['cleaned_resume'] = self.df['Resume'].apply(self.clean_resume_text)
        print("文本清理完成。")
        print("清理後履歷的前5行：\n", self.df[['Resume', 'cleaned_resume']].head())

        # 初始化TF-IDF向量化器並進行擬合和轉換
        self.tfidf_vectorizer = TfidfVectorizer(stop_words='english', max_features=self.tfidf_max_features)
        X = self.tfidf_vectorizer.fit_transform(self.df['cleaned_resume'])
        print(f"TF-IDF 特徵矩陣形狀：{X.shape}")
        return X

    def train_model(self, X, test_size=0.2, random_state=42):
        """
        訓練機器學習模型。

        參數:
            X (sparse matrix): TF-IDF特徵矩陣。
            test_size (float): 測試集佔比。
            random_state (int): 隨機種子，用於確保結果可復現。
        """
        if self.df is None:
            print("請先載入資料。")
            return

        print("\n開始模型訓練...")
        # 將類別標籤編碼為數值
        self.label_encoder = LabelEncoder()
        y = self.label_encoder.fit_transform(self.df['Category'])

        # 將資料分割為訓練集和測試集
        self.X_train, self.X_test, self.y_train, self.y_test = train_test_split(
            X, y, test_size=test_size, random_state=random_state, stratify=y
        )
        print(f"訓練集形狀 (X_train): {self.X_train.shape}, (y_train): {self.y_train.shape}")
        print(f"測試集形狀 (X_test): {self.X_test.shape}, (y_test): {self.y_test.shape}")

        # 選擇並訓練多項式朴素貝葉斯模型
        self.model = MultinomialNB()
        self.model.fit(self.X_train, self.y_train)
        print("模型訓練完成！")

    def evaluate_model(self):
        """
        評估訓練好的模型在測試集上的性能。
        """
        if self.model is None or self.X_test is None or self.y_test is None:
            print("模型尚未訓練或測試資料不可用。")
            return

        print("\n開始模型評估...")
        y_pred = self.model.predict(self.X_test)

        accuracy = accuracy_score(self.y_test, y_pred)
        print(f"模型準確度：{accuracy:.4f}")

        # 將數值標籤轉換回原始類別名稱以顯示分類報告
        target_names = self.label_encoder.inverse_transform(sorted(np.unique(self.y_test)))
        print("\n分類報告：")
        print(classification_report(self.y_test, y_pred, target_names=target_names))

    def predict_new_resume(self, new_resume_text):
        """
        使用訓練好的模型預測新履歷的類別。

        參數:
            new_resume_text (str): 新履歷的文本內容。

        返回:
            str: 預測的履歷類別名稱。
        """
        if self.model is None or self.tfidf_vectorizer is None or self.label_encoder is None:
            print("模型尚未訓練，或TF-IDF向量化器/標籤編碼器不可用。")
            return "無法預測"

        print(f"\n預測新履歷：'{new_resume_text[:50]}...'")
        # 1. 清理新履歷文本
        cleaned_new_resume = self.clean_resume_text(new_resume_text)

        # 2. 使用之前訓練好的 TF-IDF 向量化器轉換文本
        # 注意：這裡只用 transform，不能用 fit_transform
        new_resume_features = self.tfidf_vectorizer.transform([cleaned_new_resume])

        # 3. 使用訓練好的模型進行預測
        predicted_category_encoded = self.model.predict(new_resume_features)

        # 4. 將預測的數值標籤轉換回原始類別名稱
        predicted_category_name = self.label_encoder.inverse_transform(predicted_category_encoded)

        return predicted_category_name[0]

# 主執行區塊
if __name__ == "__main__":
    # 創建 ResumeScreener 實例，並指定新的資料檔案名稱
    screener = ResumeScreener(data_path='SomeoneCV.csv')

    # 1. 載入資料
    if not screener.load_data():
        print("資料載入失敗，程式終止。")
    else:
        # 2. 視覺化類別分佈
        screener.visualize_category_distribution()

        # 3. 預處理和特徵化
        X_features = screener.preprocess_and_vectorize()

        if X_features is not None:
            # 4. 訓練模型
            screener.train_model(X_features)

            # 5. 評估模型
            screener.evaluate_model()

            # 6. 預測新履歷
            sample_resume_1 = "Highly skilled software developer with expertise in Python, Java, and C++. Seeking a challenging role in backend development."
            predicted_1 = screener.predict_new_resume(sample_resume_1)
            print(f"範例履歷 1 預測類別：{predicted_1}")

            sample_resume_2 = "Experienced HR professional with strong background in talent acquisition, employee relations, and compensation & benefits."
            predicted_2 = screener.predict_new_resume(sample_resume_2)
            print(f"範例履歷 2 預測類別：{predicted_2}")

            sample_resume_3 = "Data scientist with a master's degree in statistics. Proficient in machine learning, deep learning, and data visualization using R and Python."
            predicted_3 = screener.predict_new_resume(sample_resume_3)
            print(f"範例履歷 3 預測類別：{predicted_3}")