What is LSI in terms of SEO

SEO Southwest

LSI is a process of pattern recognition within and between texts. This process uses the mathematical method of singular value decomposition, in which a matrix is ‚Äč‚Äčrepresented as the product of three special matrices. With the help of this method, relationships between text patterns and concepts within the texts are to be recognized. Terms that occur in similar contexts allow conclusions to be drawn about the concepts of the respective contexts.

Specifically with regard to search engines, this means that with LSI, a related topic of documents can be recognized even if certain terms are not even present. The mere presence of certain terms in texts can then be used to determine whether a document fits a certain topic. This represents an advance on earlier approaches to relevance assessment, which still relied on specific and precisely fitting keywords.

It is not necessary to know exactly which topic it is about, only the similarity between documents is decisive. If, for example, it is found that a document contains a large number of similar terms as a certain other document, then it can be assumed that both documents are relevant to a common topic.

So-called stop words such as "der", "die", "das" or "ist", which appear in most texts and do not allow any conclusions to be drawn about the relevance for a specific topic, are excluded from the evaluation.

Latent Semantic Indexing is the counterpart to Latent Semantic Optimization, in which one tries to increase the relevance of a document for this topic by placing synonyms and terms from the context of a certain topic.

For the mathematical background of LSI see LSI at Wikipedia and Singular value decomposition at Wikipedia.