生成所有最长公共子串的列表和变体列表

时间:2014-01-24 01:11:40

标签: c# algorithm set-theory longest-substring

高级

我试图在句子列表中折叠常见的子字符串,并仅显示它们不同的区域。所以采取这个:

Please don't kick any of the cats
Please do kick any of the cats
Please don't kick any of the dogs
Please do kick any of the dogs
Please don't kick any of the garden snakes
Please do pet any of the garden snakes

然后回复:

Please [don't|do] [kick|pet] any of the [cats|dogs|garden snakes]

更多详情

  • 我一直在研究最长公共子串算法,但似乎只比较两个字符串。
  • 我只对比较字符串中的整个单词感兴趣。
  • 只想从左到右评估字符串。
  • 不常见的子串的长度不会是相同的字数(“猫”与“花园蛇”)

我正在寻找有关算法的帮助。我相信这是LCS问题的变种,我认为某种后缀树的处理。可能解释和实现的伪代码将是理想的。

另一个例子

Please join thirteen of your friends at the Midnight Bash this Friday
Don't forget to join your friend John at the Midnight Bash tomorrow
Don't forget to join your friends John and Julie at the Midnight Bash tonight

变成:

[Please|Don't forget to]
join
[thirteen of your friends|your friend John|your friends John and Julie]
at the Midnight Bash
[this Friday|tomorrow|tonight]

也许这种方法

这种做法怎么样......

for an array of sentences
  loop with the remaining sentence
    find the "first common substring (FCS)"
    split the sentences on the FCS
    every unique phrase before the FCS is part of the set of uncommon phrases
    trim the sentence by the first uncommon phrase
  end loop

2 个答案:

答案 0 :(得分:0)

将每个唯一的单词映射到单个对象。然后构建条件概率表(参见Markov chains)以枚举一个单词跟随每个序列的次数。

答案 1 :(得分:-1)

有趣的是,我一直在考虑在很久以前创造像你这样的东西,直到我意识到这实际上是一种人工智能。需要考虑的因素太多:语法,语法,情境,错误等等。但是如果你的输入总是如此固定,就像"请[A1 | A2 | ..] [B1 | B2 | ..]任何一个[C1 | C2 | ..]"然后可能是一个简单的正则表达式模式:" ^请\ s *(?(不要| t)执行)\ s *(?\ w +)+ \ s *任何\ s *(?。)* $"。