|Anonymous | Login | Signup for a new account||2019-11-17 12:56 CET|
|Main | My View | View Issues | Change Log | Roadmap|
|View Issue Details|
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0000610||YaCy||[All Projects] General||public||2015-10-23 13:40||2016-03-09 13:25|
|Product Version||YaCy 1.8|
|Target Version||Fixed in Version|
|Summary||0000610: "maximumRecords" returns a random subset of results|
|Description||The following query|
returns relevant results.
Running the same query with a reduced maximumRecords parameter from 100 to 10 (query 0000002) doesn't return the top 10 results with lowest ranking score, as I would expect, instead it apparently returns a random subset of 10 unsorted results with casual rankings, and does not include the most relevant (lowest ranking score) results returned by query 0000001.
|Additional Information||All the results with low ranking score which are returned by 0000001 are in my local index, and thus can be retrieved repeatedly and are always accessible to my queries.|
Let me know if you need more info or if the bug report is invalid.
|Tags||No tags attached.|
0000001 stands for "query number 1"
0000002 stands for "query number 2"
As a first analysis I ran your request in debug mode. I placed conditional breakpoints on URIMetadataNode class, to stop when score proprerty (later assigned to ranking property in JSON) is set with a non zero value. I am not very experienced with YaCy sources, but in this case it seems results are a mix of two Solr queries, and thus sorting should be done again before writing final result.
It needs deeper analysis to confirm that, but if it can help you...
edited on: 2016-03-09 13:25
Thank you luc for your recent works on Yacy.
It'll be interesting to have a high-level understanding of how results are sorted and presented to the API. I have a fear that this part of the code is broken and the API code is severely neglected.
|As you found out, with ranking is a issue, as internally 2 rankings are used, the DHT and the Solr ranking (not really compatible) and mixed in the results.|
Does this defect also affect the results returned via "yacysearch.html", or it only affects the API?
Furthermore, if the defect also affects "yacysearch.html", then the whole ranking algorithm of Yacy would be flawed; if this is the case, then a honest question:
what would be the whole point of keeping using and developing Yacy, if after a decade of development it still suffers fundamental design issues (i.e. the ranking system is broken at the core)?
Even if ranking system is broken in yacysearch.json or even in yacysearch.html, I see at least some reasons to keep using and developing :
- philosophy and ideas behind the whole project are still very good
- existing code base is far from perfect but is certainly the result of a big amount of volunteers work and already give access to a large shared index dataset.
- this kind of issue is not impossible to overcome. You reported it : it is one more step to solve it and improve the whole system, step after step...
edited on: 2015-12-05 13:19
Correct me if I am wrong, but I think ranking compatibility between DHT and Solr is only one part of the problem : even if rankings were compatible it seems items would not be sorted correctly.
Indeed, in yacysearch, result items are gathered concurrently from differents sources, and they are written to output once they are available (see yacysearchitem.java : "final URIMetadataNode result = theSearch.oneResult(item, timeout);").
Finally, re-ranking and sorting of global results list from different sources can not be effective as they are concurrently pushed and pulled one by one.
To solve this issue, shouldn't we consider using SearchEvent.completeResults() method to gather consistent results in yacysearch servlet?
Thank you luc; as you are the only one actively revising the source code at this moment, I'll ask you some turn-point questions, as I am about to decide whether I'll unplug my 700GB Yacy index for a better future, when the ranking code will eventually serve its intended purpose, or to sell the hardware and recover cost.
Do you believe the ranking system offered thru the search API will become functional soon?
Do you believe image search API is accurate at all, and images are appropriately contextualized when indexed from their webpage? If not, do you think the image search API will eventually become usable, as IMHO it currently isn't?
Thank you for your confidence Davide. Difficult questions you are asking, and I am probably not the most appropriate one to answer.
In fact, Reger24 seems to be the person who has been fixing the most bugs last months and has a far stronger background on YaCy.
Did you noticed he made something about ranking : https://github.com/yacy/yacy_search_server/commit/cdb8f3b10d15edc75f68c8170a967289f7c40981. [^]
I had not enough time to check and test if it solves at least partially what you described here. Maybe you could personnaly ask him if he plans to work on the issues who bother you.
Testing takes some time to be truly accurate, especially when running in P2P mode. But I have personnaly good hope it can be solved soon.
Current results in documents or images search also don't fully satisfy me currently. But if we want it to be improved, I believe the job has to be done by ourselves.
I am only gaining some humble skills on YaCy core code which heavily rely on Solr and concurrency. Not so easy to handle... YaCy search system is quite challenging and I think improving results relevance is a big task. It's up to you to decide if you can wait or rather contribute as far as you can.
By the way, I will continue trying to enhance what I can, once I will have some time... after christmas and new year holidays.
|2015-10-23 13:40||Davide||New Issue|
|2015-10-23 13:42||Davide||Note Added: 0001119|
|2015-11-28 00:47||Davide||Note Added: 0001138|
|2015-11-28 19:51||luc||Note Added: 0001139|
|2015-11-29 13:04||Davide||Note Added: 0001141|
|2015-11-29 17:31||BuBu||Note Added: 0001142|
|2015-11-29 17:46||Davide||Note Added: 0001143|
|2015-11-30 00:21||luc||Note Added: 0001144|
|2015-12-05 13:18||luc||Note Added: 0001146|
|2015-12-05 13:19||luc||Note Edited: 0001146||View Revisions|
|2015-12-18 00:22||Davide||Note Added: 0001180|
|2015-12-18 02:01||luc||Note Added: 0001181|
|2016-03-09 13:23||Davide||Note Deleted: 0001138|
|2016-03-09 13:25||Davide||Note Edited: 0001141||View Revisions|
|Copyright © 2000 - 2019 MantisBT Team|