3 ms·
the biggest issue with postgres search is the inability to use TF-IDF or BM25 (the current default and state of the art on elasticsearch). The current ranking s
by sandGorgon 8y ago
the biggest issue with postgres search is the inability to use TF-IDF or BM25 (the current default and state of the art on elasticsearch). The current ranking system is not very relevant.
Anyone who is familiar with PG internals - is there something in the internal data structure that prevents a BM25 or TF-IDF style rank generation ?
The work on RUM seems to have stagnated (and TF-IDF was a todo anyhow here) https://github.com/postgrespro/rum https://github.com/postgrespro/rum
I have a theory that if they incorporate these algorithms, it makes postgres potent enough that a lot of people may choose not to use elasticsearch/lucene.
- 1996 8y agoThen by your own theory, unless there is a conspiracy within the devel community, this will be made - if only as extensions with commercial support.
- threeseed 8y agoI had never heard of these algorithms until now. But I would still use Elasticsearch since (a) everyone else does so lots of support, (b) ridiculous amount of plugins, drivers and rivers, (c) PostgreSQL has a far weaker scalability story, (d) Kibana is a great UI to help with debugging, (e) you can use ELK for your monitoring as well as search. I could definitely come with a few more but Elasticsearch is a product that does search really well.
- aisofteng 8y agoThese are standard information retrieval algorithms that have been known and used for decades.