LARI Score (Local Aggregate Relevance Index)

From Wiki

Jump to: navigation, search

LARI Score (Local Aggregate Relevance Index) is a proprietary algorithm/calculation created by Kelley Reynolds of Inside Sytems Inc.

LARI is a scale of 1 to 100 that indicates how relevant a keyword is to a seed term, with 1 being the least relevent and 100 being the most relevent.  LARI is calculated using a corpus of documents collected based on the seed, and a weighted average of scaled LSI and Bayes values.

In the case of Network Empire's KEYWORDS (Keywords™ By Network Empire), that corpus of documents consists of the top results from Google and Yahoo. LSI and Bayes are completely disparate in how they function, but in our implementation, both are scaled from 0 to 100 and combined with more weight placed on LSI for relevance.

The reason that LARI stands for 'local aggregate relevance index' is because the numbers from 0 to 100 are not comparable between analyses .. it is only relevant within the same analysis. An '80' for one keyword does not necessarily mean the same thing for '80' for another keyword. In particular, the number of terms, stopwords, and different stems will change what this number means, as will the number of different documents retrieved and their contents.

Bayes and LSI can be analyzed independently as they convey different meanings, but LARI is a convenient way of combining those into a single generalized value that quickly finds highly semantically valuable keywords.

LARI is a theme relevance index that uses the bayes and LSI values for the keyword measured against the averages for the entire list of terms returned to determine a relevance score. A score above 60 is highly relevant, a score between 40 and 60 is largely relevant and under 40 is starting to get a bit tangent.

Notes: Our particular Bayes implementation would be considered native and our LSI and Bayes indices both utilize multi-lingual stemming to give locale-specific results, though the quality of those is highly dependent on the corpus and since the search engines don't always get this right, neither do we.

Kelley Reynolds

See our wiki entry on latent semantic indexing for more information

Also See Network Empire's Content Curation Course - Cash Micro-Content Curation, Syndication and Publishing at the Speed of Thought

Personal tools