Update from Lucene
May 10, 2016
It has been awhile since we heard about our old friend Apache Lucene, but the open source search engine has something new, says Open Source Connections in the article, “BM25 The Next Generation Of Lucene Relevance.” Lucene is added BM25 to its search software and it just might improve search results.
“BM25 improves upon TF*IDF. BM25 stands for “Best Match 25”. Released in 1994, it’s the 25th iteration of tweaking the relevance computation. BM25 has its roots in probabilistic information retrieval. Probabilistic information retrieval is a fascinating field unto itself. Basically, it casts relevance as a probability problem. A relevance score, according to probabilistic information retrieval, ought to reflect the probability a user will consider the result relevant.”
Apache Lucene formerly relied on TF*IDF, a way to rank how users value a text match relevance. It relied on two factors: term frequency-how often a term appeared in a document and inverse document frequency aka idf-how many documents the term appears and determines how “special” it is. BM25 improves on the old TF*IDF, because it gives negative scores for terms that have high document frequency. IDF in BM25 solves this problem by adding a 1 value, therefore making it impossible to deliver a negative value.
BM25 will have a big impact on Solr and Elasticsearch, not only improving search results and accuracy with term frequency saturation.
Whitney Grace, May 10, 2016
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
Comments
One Response to “Update from Lucene”
-
- Subscribe to Beyond Search
Feature archive
News archive
-
For the record, Elasticsearch has had BM25 for a couple years now. Good that Solr can continue to play catch-up. 😉