語言模型
台南一中AI選修教材 / 顏永進

 1: text = "我喜歡吃拉麵我喜歡吃牛肉麵我喜歡喝珍珠奶茶我不喜歡吃苦瓜"
 2: 
 3: # 統計每個字後面接什麼字
 4: next_char = {}
 5: for i in range(len(text) - 1):
 6:     current = text[i]
 7:     follow = text[i + 1]
 8:     if current not in next_char:
 9:         next_char[current] = {}
10:     if follow not in next_char[current]:
11:         next_char[current][follow] = 0
12:     next_char[current][follow] += 1
13: 
14: # 看看「我」後面最常接什麼
15: print("「我」後面接的字及出現次數：")
16: for char, count in sorted(next_char["我"].items(), key=lambda x: -x[1]):
17:     print(f"  {char}: {count} 次")
18: 
19: print()
20: print("「喜」後面接的字及出現次數：")
21: for char, count in sorted(next_char["喜"].items(), key=lambda x: -x[1]):
22:     print(f"  {char}: {count} 次")

 1: import random
 2: 
 3: # 用 bigram 來「生成」文字
 4: def build_bigram(text):
 5:     """建立 bigram 模型：統計每個字後面各個字出現的次數"""
 6:     model = {}
 7:     for i in range(len(text) - 1):
 8:         c = text[i]
 9:         n = text[i + 1]
10:         if c not in model:
11:             model[c] = {}
12:         model[c][n] = model[c].get(n, 0) + 1
13:     return model
14: 
15: def predict_next(model, current_char):
16:     """根據 bigram 模型預測下一個字"""
17:     if current_char not in model:
18:         return "？"
19:     candidates = model[current_char]
20:     total = sum(candidates.values())
21:     # 顯示每個候選字的機率
22:     for char, count in sorted(candidates.items(), key=lambda x: -x[1]):
23:         prob = count / total * 100
24:         print(f"  {char}: {prob:.1f}%")
25:     # 回傳機率最高的
26:     return max(candidates, key=candidates.get)
27: 
28: corpus = "我喜歡吃拉麵我喜歡吃牛肉麵我喜歡喝珍珠奶茶我不喜歡吃苦瓜我喜歡打籃球"
29: model = build_bigram(corpus)
30: 
31: print("給定「吃」，模型預測下一個字的機率：")
32: best = predict_next(model, "吃")
33: print(f"→ 模型選擇：{best}")
34: print()
35: print("給定「歡」，模型預測下一個字的機率：")
36: best = predict_next(model, "歡")
37: print(f"→ 模型選擇：{best}")