Trying to get a customer to convert to plone with elasticsearch, but they have to be sure that the search is as good as their current solution. Short of doing a content import and a trial, I'm looking for existing installs they can look at to get a feel for how well it works in terms of relevancy.
1 Like
Looking at this a bit more it does seem the system I am trying to compare elasticsearch/solr/lucene to does do more to make its ranking more relavent.
https://www.elastic.co/guide/en/elasticsearch/guide/current/scoring-theory.html
vs
https://docs.funnelback.com/develop/reference-documents/funnelback-ranking-algorithms.html
So I guess my question now is
a) how does relevance in solr/elasticsearch compare to plone? Most discussion here seems to be around speeding up indexing and lower Plones RAM. @zopyx you worked on c.elasticindex, any opinion?
b) If castleCMS uses google analytics data to improve the basic elasticsearch ranking, why? How bad was elasticsearch by itself? How much better does it make it and in what circumstances?
SOLR and ES are based on Lucene - so they basically share the same behaviour.
In general the default search results and ranking are pretty good but there are a lot of knobs that you can or need to adjust in order to make it work for your particular use case in particular when you work with advanced features like wildcards, stemming, phrase search and so on. In general I would not touch much of the default configuration. Term boosting (hits in the title should be more important than hits in the text)...but it really depends on the nature of the content. e.g. in our content a relevant hit is not necessarily relevant when it mentions a particular search term frequently...lots of knobs and wheel for wasting your time...
-aj
Thanks @zopyx, do you know of some good public examples of plone integrated with solr or elasticsearch that perhaps doesn't do a lot of tuning?
@reinhardt and @loechel can possibly provide some Solr based examples.
I have only one intranet running on ES.
-aj
You could have a look at recensio.net (see also recensio.buildout). The LMU also uses solr, but @loechel can probably better tell you about that one. AFAIR solr.buildout and lmu.policy.intranet are the packages where the magic happens.
Hi @djay well, yes we do use a multi instance + multiple different CMS based Intranet at and Extranet at LMU backed by Solr. @reinhardt already pointed to the corresponding repositories.
We did not choose Solr to save RAM or make indexing quicker, but to search accross multiple sources and to have a better support for german. Solr is normally used with a schema, elasticsearch instead schemaless.
Solr is extremely configurable to adapt it to specific use cases.
But it is with all search engines you need someone to take care of the index.
Solr basic template for Plone is better than the Plone search out of the box, but could be improved a lot. Include data from any analytic could help to improve the index, but that is probably more complex than just boost them by those terms / sites.
For each larger setup I would always use an external search engine, today.
Thanks for you help guys.