A major problem in the assessment of documents is the recognition of the topic that describes this document. This problem is evident if one takes, for example, words with multiple meanings (so-called "Teekesselchen") as an example. Without additional information, a search engine does not distinguish between the importance attached a word on a page.

Phrase Matching

To work around this problem probably used Google algorithms that calculate the semantic proximity of words. A clue can be found, for example, Automatic taxonomy generation in search results using phrases . Here a method is presented in which groups of words associated with a keyword or combination of keywords will be investigated. Applying this method to the corresponding number of documents, as these documents can be put together on the basis of groups of words in clusters and thereby allow a categorization of the content.

LDA - deferred Semantic Allocation

A similar approach is also known as Latent Dirichlet Allocation (LDA). These systems are quite possible at the present time, which, for example, in the online course information retrieval and text mining is demonstrated.

LSI - Deferred Semantic Indexing

Often one reads of a so-called latent semantic indexing (LSI), which amounts to a similar result. 

Conclusion

Finally, care should be taken in any case that there are various methods already, by means of which the semantic content of documents can also be seen in some way. This data may be used by Google so well for the ranking. SeoBook.com provides some practical examples of how this knowledge as SEO can take advantage of. For a search engine optimized text which generally means that not only the mention of the keyword to the ranking of importance, but also the context of the word is illustrated by semantically similar terms, so that there is more relevant to a search engine.