22、ElasticSearch 实战:倒排索引核心原理
我们看看下面2个doc是怎么建立倒排索引的
doc1:I really liked my small dogs, and I think my mom also liked them.
doc2:He never liked any dogs, so I hope that my mom will not expect me to liked him.
word | doc1 | doc2 |
---|---|---|
I | * | * |
really | * | |
liked | * | * |
my | * | * |
small | * | |
dogs | * | |
and | * | |
think | * | |
mom | * | * |
also | * | |
them | * | |
He | * | |
never | * | |
any | * | |
so | * | |
hope | * | |
that | * | |
will | * | |
not | * | |
expect | * | |
me | * | |
to | * | |
him | * |
此时我们全文检索mother like little dog,是搜索不到结果的
那这是不是我们想要的?绝对不是。因为在我们看来monther和mom有区别吗?都是妈妈的意思,同义词。like和liked有区别吗?没有,都是喜欢的意思,只不过一个是现在时,一个是过去时。little和small有区别吗?都是小的,同义词。dog和dogs有区别吗?够,只不过一个单数一个复数。
所以如果是这样的建立索引和检索的话就是很失败的
因此,es其实在建立倒排索引的时候会进行一个操作(normalization),也就是对拆分出来的各个单词进行相应的处理(时态的转换,单复数的转换,同义词的转换,大小写的转换),以提升后面搜索的时候能够搜索到相关联的文档的概率
针对上面2个doc,建立索引时会进行下面的转换
liked —> like
small —> little
dogs —> dog
加入normalization操作后,真实的倒排索引是这样的
word | doc1 | doc2 | normalization |
---|---|---|---|
I | * | * | |
really | * | ||
like | * | * | liked --> like |
my | * | * | |
little | * | small --> little | |
dog | * | * | dogs --> dog |
and | * | ||
think | * | ||
mom | * | * | |
also | * | ||
them | * | ||
He | * | ||
never | * | ||
any | * | ||
so | * | ||
hope | * | ||
that | * | ||
will | * | ||
not | * | ||
expect | * | ||
me | * | ||
to | * | ||
him | * |
接下来我们进行全文检索mother like little dog,会先进行分词和normalization操作
mother --> mom
like --> like
little --> little
dog --> dog
此时doc1和doc2都会搜索出来了