The "similar records" feature always reports "Your search did not match any records. Please try again."

[SarahBA]

It seems that it can't find the related subjects.

[Edited by emanueldima] This is how to test:

1. Create a "hi.txt" file with "hi" as content
2. Create a new deposit with this file; domain: Eudat, title: "Neptun jupiter pluto", description: "Neptun jupiter pluto"
3. Wait ~10 minutes to make sure the ingestion and ranking services run correctly
4. From the home page, click on "search" to get the list of deposits
5. From the list of deposits, the first one is the latest deposit, the one we just created; Click on its "similar records" link
6. If it works, you should get another search page with 1 deposit, the "Neptun jupiter pluto" object. If it doesn't work, which is what happens currently, you get a "Your search did not match any records. Please try again." message

[emanueldima]

I have been investigating this for the whole day. It seems to work correctly on my clean VM. The record must have multiple words in title and description, these words must be non common and must show up in multiple documents (not just one). A divide by 0 error happens on one code path, but this seems to be unrelated.

[emanueldima]

It also works on the testing server, now that the bibsched services are up and running. See the similar records for https://test-eudatis.csc.fi/record/118

It holds true that this feature behaves a bit weirdly. The record must have more than one word in the header/description, only then you get similar records (and sometimes not even then). This idiosyncrasy is however originating from invenio, and it's very difficult to change it.

I am closing the issue as "invalid".

[emanueldima]

... and now, 10 minutes after, the functionality is not working anymore. I don't understand why, but I'm certain that it used to work for record 118 and now it doesn't.

[emanueldima]

It seems to be a configuration difference between my developer VM and the deployed service. It works on the VM, it doesn't on the test and production servers. I don't know how to solve it, so I am assigning it to Lassi, who can test and see the differences between the VMs and deployment.

@llehtine to check if it works, this is what I did:

1. Create a "hi.txt" file with "hi" as content
2. Create a new deposit with this file; domain: Eudat, title: "Neptun jupiter pluto", description: "Neptun jupiter pluto"
3. Wait ~10 minutes to make sure the ingestion and ranking services run correctly
4. From the home page, click on "search" to get the list of deposits
5. From the list of deposits, the first one is the latest deposit, the one we just created; Click on its "similar records" link
6. If it works, you should get another search page with 1 deposit, the "Neptun jupiter pluto" object. If it doesn't work, which is what happens on the test server, you get a "Your search did not match any records. Please try again." message

[llehtine]

I tried this also with new VM from CSC Cloud, and I was able to reproduce it. After adding the first entry, it seems like it was working, but after adding second entry with same title / descrpition it did break.

[emanueldima]

Ok, we clearly need more time for this one. I am moving it to the Spring milestone.

[stranak]

OK, makes sense.

[emanueldima]

Moving to summer release... :disappointed:

[emanueldima]

To test if valid for the b2share/next

[emanueldima]

Comment from Lars: "To make it work, you need to run "bibrank" in particular you need the word similarity bibrank -w wrd. I however found an issue which I've fixed in https://github.com/inveniosoftware/invenio/pull/2350"