Solr with RA, accurate and relevant search

Apache Solr** 4.0 with RankingAlgorithm 1.4.2 (build 2012-03-19) is now available for download. See downloads below ...

• **New**, NRT now works with both RankingAlgorithm and Lucene
• NRT does not close the searcher (no commits are needed), so can be fine grained resulting in thread1, thread2, thread3 returning new results
• See how the NRT implemenation is more granular/realtime compared to soft commit, NRT comparison with soft commit
• Very high insert/update performance, 70000 docs / sec
• Very high performance, <10 ms response time (50 million index)
• Scale to 1 billion documents or more

Apache Solr** with RankingAlgorithm enables Solr's search results to be comparable to Google Site Search and much better than Apache Lucene** for searches using the Perl index, see comparison searches. RankingAgorithm is a search library that uses a new scoring algorithm to rank results accurately and relevantly. RankingAlgorithm is very easy to use and uses the very popular Lucene index but scores and ranks on its own.

Solr with RankingAlgorithm also supports NRT and can update documents at about 5000 documents in 498 with concurrent searches without closing the Searchers or clearing the caches, etc., see Near Real Time Search.

Multiple Algorithm are available SIMPLE, SIMPLE1, COMPLEX, COMPLEX1. SIMPLE* are very fast algorithms and can return queries in <50ms on a 10m wikipedia index (complete index). It can also scale to 100m docs or maybe more. COMPLEX is a more complex algorithm so is a little slower compared to the SIMPLE, but can also still return queries in < 100ms on a 10m wikipedia index (complete index). COMPLEX is more accurate and should be able to give you the best rankings as compared to SIMPLE*.

RankingAlgorithm works in two modes, document mode and product mode:

In document mode, it ranks documents such as HTML, Wikipedia, Word/PDF docs relevantly while in Product mode, a term's occurrence is taken into account and scored accordingly. So titles starting with "wii console" are ranked first, and the others rank lower as the occurrence of "wii console" shifts in the title or gets reversed, see below:

• Wii Console and Wii Fit Plus with Balance Board Bundle (Nintendo Wii)
• Wii Console System with Wii Sports Resort Game with TWO MotionPlus Attachments
• Nintendo Wii Console w/ Bonus Wii Sports Resort Bundle, Black
• Pelican Accessories Wii Console Stand - Nintendo Wii
• Grafitti Skin for Nintendo Wii Console
• NEW AC Adapter Cable Cord Power Supply For NINTENDO WII Gaming Console
• Wii Remote Charging Console Stand
• Nintendo Wii Skin - System Console Skin and two Wii Remote Skins - Blue Vortex
• CET Domain 10301901 Console Stand Station for Nintendo Wii

The RankingAlgorithm has been integrated into Solr so that either Lucene or the RankingAlgorithm can be used to do the search. The RankingAlgorithm scoring or working does not break any of the existing functionality, so shrad, faceting, highlighting, etc. still work as before. RankingAlogirhtm only uses Lucene Apis to retrieve the terms from the index and uses its own ranking and scoring to rank the documents. The scoring is very friendly and easy to follow.

Get more information:

• Solr 4.0 with RankingAlgorithm 1.4 paper
• Solr 3.5 with RankingAlgorithm 1.4.1 paper
• Solr 3.5 with RankingAlgorithm 1.3 paper
• Solr 3.4 with RankingAlgorithm 1.3 paper
• Solr with RankingAlgorithm 1.2 paper
• Near Real Time (NRT) paper
• Solr with RankingAlgorithm 1.1 paper

Give it a try Download Apache Solr with RankingAlgorithm

Try the Demos here,

• Autocomplete similar to Google/Yahoo/Bing


• Faceting/Autcomplete demo


• Integration with JamWiki


• Wikipedia/Products/Perl.org

Install Solr as before, no changes to the existing installation steps (see Solr docs for installation). No changes to the way you query or use Solr. The change is when you initiate a query, the search uses RankingAlgorithm instead of Lucene. You can still use Lucene by adding "&lucene=true" to use Lucene as before. You can download Solr with RA from here and follow the steps as in the download docs either on the Solr website or from here.

See examples below:

Searching in document mode (default):
http://localhost:8773/solr/select/?q=california gold rush&fl=score

Searching in product mode:
http://localhost:8773/solr/select/?q=wii console&fl=score&scoring=product

Searching using Lucene library:
http://localhost:8773/solr/select/?q=california gold rush&fl=score&lucene=true

Autocomplete using Solr with RankingAlgorithm, similar to Google/Yahoo/Bing's autocomplete (It is free),  Give it a try

Try the Demos here, Faceting/Autcomplete demo,  Integration with JamWiki,   Wikipedia/Products/Perl.org

1. Search very accurate and relevant. Comparable to Google site search and much better than Lucene, see comparison.
2. Three algorithms, SIMPLE, SIMPLE1 and COMPLEX.
3. Two modes, Document and Product mode. Product mode enables very accurate product/retail/short twitter text searches. Document mode enables relevant search, can be used for product searches too.
4. By default AND/OR combinations.
5. Very easy scoring with a relevancy score.
6. Very easy to use.
7. Score boosting, supports Document, Field, Query & Query term boosts.
8. Uses the very popular Lucene index. No changes to your code or index.
9. NRT search
10. Query a 10m wikipedia index in <50 ms, scale upto 100m docs
11. Supports +- boolean queries, entire Lucene Query Syntax

Browse the javadocs here:

RankingAlgorithm 1.4.1 docs,RankingAlgorithm 1.4.1 Javadocs.
RankingAlgorithm 1.3 docs, RankingAlgorithm 1.3 Javadocs.
RankingAlgorithm 1.2 docs, RankingAlgorithm 1.2 Javadocs.
RankingAlgorithm 1.1/1.0 docs, RankingAlgorithm 1.1 Javadocs.

WIKI, Apache Solr with RankingAlgorithm WIKI.
Browse RankingAlgorithm usage Example.java, SimpleExample.java, LuceneCollectorExample.java

Contact,

• Discussion - community (preferred)
• email -- solr-ra at tgels dot com
• twitter
• facebook