\u4ece 1970 \u5e74\u4ee5\u540e\u4eba\u4eec\u610f\u8bc6\u5230\u4e0d\u80fd\u57fa\u4e8e\u89c4\u5219\uff0c\u5f97\u57fa\u4e8e\u7edf\u8ba1\u8fd9\u4e2a\u65f6\u95f4\u8282\u70b9\u5f00\u59cb\uff0c\u5c31\u51fa\u73b0\u4e86\u5f88\u591a\u8bed\u8a00\u6a21\u578b\u3002<\/p>\n\n\n\n
\u57fa\u672c\u4ecb\u7ecd<\/strong><\/p>\n\n\n\n N-Gram\u6a21\u578b\u8bde\u751f\u4e8e 1950s\u20131960s\uff0c\u6700\u65e9\u7531\u9999\u519c\uff08Claude Shannon\uff09\u5728\u4fe1\u606f\u8bba\u4e2d\u63d0\u51fa\uff0c\u7528\u4e8e\u8bed\u8a00\u7684\u6982\u7387\u5efa\u6a21\uff0c\u9999\u519c\u5728 1951 \u5e74\u7684\u8bba\u6587\u4e2d\u63d0\u51fa\uff0c\u4f7f\u7528 1-gram\u30012-gram \u7b49\u65b9\u6cd5\u4f30\u8ba1\u82f1\u6587\u6587\u672c\u7684\u6982\u7387\u3002\u524d\u9762\u7684 N \u662f\u4e00\u4e2a\u6570\u5b57\uff0c\u8868\u793a\u4f60\u6bcf\u6b21\u8981\u770b\u51e0\u4e2a\u8bcd\uff0c\u4f8b\u5982\uff1a<\/p>\n\n\n\n \u8be5\u6a21\u578b\u57fa\u4e8e\u9a6c\u5c14\u53ef\u592b\u5047\u8bbe\u3002<\/p>\n\n\n\n \u9a6c\u5c14\u53ef\u592b\u5047\u8bbe\uff1a\u4e00\u4e2a\u8bcd\u51fa\u73b0\u7684\u6982\u7387\uff0c\u4f9d\u8d56\u4e8e\u5b83\u524d\u9762\u7684 N-1 \u4e2a\u8bcd\uff0c\u800c\u4e0d\u662f\u6574\u4e2a\u53e5\u5b50\u5386\u53f2\u3002<\/p>\n\n\n\n \u5728\u4f60\u7684\u5927\u8111\u4e2d\uff0c\u5c31\u4f1a\u81ea\u52a8\u7684\u53bb\u9884\u6d4b\u4e0b\u4e00\u4e2a\u8bcd\u3002<\/p>\n\n\n\n \u4e0d\u4f1a\u5f39\u51fa\u4e0b\u9762\u7684\u8bcd\uff1a<\/p>\n\n\n\n \u5de5\u4f5c\u539f\u7406<\/strong><\/p>\n\n\n\n \u8fd9\u91cc\u6211\u4eec\u4ee5 Bigram\uff082-Gram\uff09\uff1a\u4f1a\u770b\u524d\u9762\u7684\u4e00\u4e2a\u8bcd\u3002<\/p>\n\n\n\n \u5047\u8bbe\u6211\u4eec\u6709\u8fd9\u6837\u4e00\u4e2a\u201c\u8bed\u6599\u5e93\u201d\uff0c\u6211\u4eec\u7528\u8fd9\u4e9b\u6570\u636e\u6765\u8bad\u7ec3\u6a21\u578b\uff1a<\/p>\n\n\n\n Bigram \u6a21\u578b\u4f1a\u6839\u636e\u4e00\u4e2a\u8bcd\u7edf\u8ba1\u4e0b\u4e00\u4e2a\u8bcd\u51fa\u73b0\u7684\u6982\u7387<\/strong>\uff0c\u8fd9\u91cc\u6211\u4eec\u53ef\u4ee5\u6570\u4e00\u6570\u6240\u6709\u8bcd\u5bf9\uff1a<\/p>\n\n\n\n \u6839\u636e\u8fd9\u4e2a\u8bcd\u7684\u7ec4\u5408\uff0c\u5c31\u53ef\u4ee5\u53bb\u9884\u6d4b\u4e00\u4e2a\u8bcd\u7684\u4e0b\u4e00\u4e2a\u8bcd\uff0c\u6bd4\u5982\uff1a<\/p>\n\n\n\n \u6240\u4ee5\uff1a<\/p>\n\n\n\n \u5982\u679c\u4f60\u770b\u5230\u201c\u6211 \u7231 \u5403<\/strong>\u201d\uff0c\u90a3\u4e0b\u4e00\u4e2a\u8bcd\u5927\u6982\u7387\u662f\u201c\u82f9\u679c\u201d\uff01<\/p>\n\n\n\n \u4ee3\u7801\u5b9e\u8df5<\/strong><\/p>\n\n\n\n jieba \u662f python \u91cc\u9762\u7528\u4e8e\u5bf9\u4e2d\u6587\u5206\u8bcd\u7684\u4e00\u4e2a\u5e93\u3002<\/p>\n\n\n\n \u201c\u6211\u559c\u6b22\u5403\u82f9\u679c\u201d<\/p>\n\n\n\n \u6211\u3001\u559c\u6b22\u3001\u5403\u3001\u82f9\u679c<\/p>\n\n\n\n N-Gram \u867d\u7136\u6bd4\u8f83\u7b80\u964b\uff0c\u4f46\u662f\u5b83\u786e\u786e\u5b9e\u5b9e\u662f\u57fa\u4e8e\u7edf\u8ba1\u8bed\u8a00\u5efa\u6a21\u7684\u8d77\u70b9\u3002<\/p>\n\n\n\n N-Gram\u6a21\u578b\u7f3a\u9677<\/strong><\/p>\n\n\n\n \u603b\u7ed3\u4e00\u53e5\u8bdd\uff1a\u53ea\u4f1a\u8bb0\u5f55\u8bcd\u8bed\u642d\u914d\uff0c\u4e0d\u4f1a\u7406\u89e3\u8bed\u8a00\u3002<\/p>\n\n\n\n \u9002\u7528\u4e8e\u65e9\u671f\u7684\u7b80\u5355\u7684\u6587\u672c\u5904\u7406\u4efb\u52a1\u3002<\/p>\n","protected":false},"excerpt":{"rendered":" \u4ece 1970 \u5e74\u4ee5\u540e\u4eba\u4eec\u610f\u8bc6\u5230\u4e0d\u80fd\u57fa\u4e8e\u89c4\u5219\uff0c\u5f97\u57fa\u4e8e..<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[27],"tags":[],"class_list":["post-364","post","type-post","status-publish","format-standard","hentry","category-ai"],"featured_image_urls":{"full":"","thumbnail":"","medium":"","medium_large":"","large":"","1536x1536":"","2048x2048":""},"author_info":{"info":["\u9648 \u67d0\u4eba"]},"category_info":"AI<\/a>","tag_info":"AI","comment_count":"0","_links":{"self":[{"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/posts\/364","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/comments?post=364"}],"version-history":[{"count":1,"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/posts\/364\/revisions"}],"predecessor-version":[{"id":365,"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/posts\/364\/revisions\/365"}],"wp:attachment":[{"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/media?parent=364"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/categories?post=364"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mitongxue.cn\/index.php\/wp-json\/wp\/v2\/tags?post=364"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}\n
\uff08\u53a8\u623f\uff09\u5988\u5988\u8bf4\uff1a\u201c\u5e2e\u6211\u62ff\u4e00\u4e0b\u51b0\u7bb1\u91cc\u9762\u7684....\u201d<\/pre>\n\n\n\n
\n
\n
\u201c\u6211 \u7231 \u5403 \u82f9\u679c\u201d
\u201c\u6211 \u7231 \u5403 \u9999\u8549\u201d
\u201c\u6211 \u559c\u6b22 \u5403 \u82f9\u679c\u201d<\/pre>\n\n\n\n\u524d\u4e00\u4e2a\u8bcd<\/th> \u4e0b\u4e00\u4e2a\u8bcd<\/th> \u6b21\u6570<\/th><\/tr><\/thead> \u6211<\/td> \u7231<\/td> 2 \u6b21<\/td><\/tr> \u6211<\/td> \u559c\u6b22<\/td> 1 \u6b21<\/td><\/tr> \u7231<\/td> \u5403<\/td> 2 \u6b21<\/td><\/tr> \u559c\u6b22<\/td> \u5403<\/td> 1 \u6b21<\/td><\/tr> \u5403<\/td> \u82f9\u679c<\/td> 2 \u6b21<\/td><\/tr> \u5403<\/td> \u9999\u8549<\/td> 1 \u6b21<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n \n
import jieba # \u5bfc\u5165 jieba \u4e2d\u6587\u5206\u8bcd\u5e93
# \u5bfc\u5165 defaultdict \u548c Counter\uff0c\u7528\u4e8e\u6784\u5efa\u9891\u7387\u7edf\u8ba1\u8868
from collections import defaultdict, Counter
\u200b
# \u793a\u4f8b\u8bed\u6599\u5e93\uff08\u6ce8\u610f\uff1a\u8fd9\u4e9b\u662f\u8fde\u7eed\u7684\u4e2d\u6587\u53e5\u5b50\uff0c\u6ca1\u6709\u7a7a\u683c\uff09
corpus = [
\"\u6211\u559c\u6b22\u5403\u82f9\u679c\",
\"\u6211\u559c\u6b22\u5403\u9999\u8549\",
\"\u5979\u559c\u6b22\u5403\u8461\u8404\",
\"\u4ed6\u4e0d\u559c\u6b22\u5403\u9999\u8549\",
\"\u4ed6\u559c\u6b22\u5403\u82f9\u679c\",
\"\u5979\u559c\u6b22\u5403\u8349\u8393\"
]
\u200b
# 1. \u6784\u5efa Bigram \u7edf\u8ba1\u8868\uff0c\u683c\u5f0f\u4e3a bigrams[\u524d\u4e00\u4e2a\u8bcd][\u4e0b\u4e00\u4e2a\u8bcd] = \u51fa\u73b0\u6b21\u6570
bigrams = defaultdict(Counter) # \u81ea\u52a8\u521d\u59cb\u5316\u5d4c\u5957\u7ed3\u6784\uff0c\u907f\u514d\u624b\u52a8\u5224\u65ad\u952e\u662f\u5426\u5b58\u5728
\u200b
# \u904d\u5386\u6bcf\u4e00\u53e5\u8bdd
for sentence in corpus:
words = list(jieba.cut(sentence)) # \u4f7f\u7528 jieba \u5bf9\u53e5\u5b50\u8fdb\u884c\u5206\u8bcd\uff0c\u8fd4\u56de\u8bcd\u5217\u8868
# \u6784\u5efa\u4e8c\u5143\u7ec4\uff08Bigram\uff09\uff1a\u6bcf\u4e24\u4e2a\u8fde\u7eed\u8bcd\u4e4b\u95f4\u7684\u642d\u914d
for i in range(len(words) - 1):
w1, w2 = words[i], words[i + 1] # \u5f53\u524d\u8bcd\u548c\u4e0b\u4e00\u4e2a\u8bcd
bigrams[w1][w2] += 1 # \u51fa\u73b0\u4e00\u6b21\u5c31 +1
\u200b
# 2. \u5c06\u9891\u7387\u8f6c\u6362\u6210\u6982\u7387\uff0c\u6784\u5efa Bigram \u6982\u7387\u6a21\u578b
bigram_probs = {} # \u6700\u7ec8\u4fdd\u5b58\u7684\u662f P(w2 | w1) \u7684\u6982\u7387\u8868
\u200b
# \u904d\u5386\u6240\u6709\u524d\u8bcd\uff08w1\uff09
for w1 in bigrams:
total = sum(bigrams[w1].values()) # w1 \u540e\u9762\u51fa\u73b0\u6240\u6709\u8bcd\u7684\u603b\u6b21\u6570
# \u8ba1\u7b97\u6bcf\u4e00\u4e2a w2 \u7684\u6761\u4ef6\u6982\u7387\uff1aP(w2 | w1) = \u9891\u7387 \/ \u603b\u6570
bigram_probs[w1] = {
w2: count \/ total for w2, count in bigrams[w1].items()
}
\u200b
# 3. \u5b9a\u4e49\u4e00\u4e2a\u51fd\u6570\uff1a\u8f93\u5165\u4e00\u4e2a\u8bcd\uff0c\u9884\u6d4b\u5b83\u540e\u9762\u6700\u53ef\u80fd\u51fa\u73b0\u7684\u8bcd\uff08\u6309\u6982\u7387\u4ece\u5927\u5230\u5c0f\u6392\u5e8f\uff09
\u200b
\u200b
def predict_next_word(word):
if word not in bigram_probs:
return \"\u672a\u77e5\uff08\u6ca1\u6709\u6570\u636e\uff09\" # \u5982\u679c\u8fd9\u4e2a\u8bcd\u4ece\u672a\u51fa\u73b0\u5728\u8bad\u7ec3\u8bed\u6599\u4e2d
# \u6309\u7167\u6982\u7387\u4ece\u9ad8\u5230\u4f4e\u6392\u5e8f\u8fd4\u56de
sorted_probs = sorted(
bigram_probs[word].items(), key=lambda x: x[1], reverse=True
)
return sorted_probs
\u200b
\u200b
# 4. \u6253\u5370\u6240\u6709 Bigram \u7684\u6761\u4ef6\u6982\u7387
print(\"=== Bigram \u6982\u7387 ===\")
for w1 in bigram_probs:
for w2 in bigram_probs[w1]:
prob = bigram_probs[w1][w2]
print(f\"P({w2} | {w1}) = {prob:.2f}\") # \u8f93\u51fa\u5f62\u5982\uff1aP(\u5403 | \u559c\u6b22) = 1.00
\u200b
# 5. \u6d4b\u8bd5\u9884\u6d4b\u67d0\u4e9b\u8bcd\u540e\u9762\u6700\u53ef\u80fd\u51fa\u73b0\u7684\u8bcd
print(\"\\n=== \u9884\u6d4b\u793a\u4f8b ===\")
for test_word in [\"\u6211\", \"\u5403\", \"\u559c\u6b22\", \"\u82f9\u679c\"]:
print(f\"{test_word} \u2192 {predict_next_word(test_word)}\")<\/code><\/pre>\n\n\n\n\n
\n
\n
\n