Page MenuHomePhabricator

False positives with dual intitle terms
Closed, DuplicatePublicBUG REPORT

Description

Steps to replicate the issue (include links if applicable):

Background
RW question – how many articles about viruses also have separate articles about the virus infection (e.g., Andes virus *and* Andes virus infection) ?

What happens?:

  • Notice: among the top ten results not involving a redirect, seven do not have both terms in the title
    • viz.: six without 'infection': 'Respiratory syncytial virus', 'B virus', 'Epstein–Barr virus', 'Ross River virus', 'Virus latency', and 'Marburg virus disease'; and one does not have 'virus', i.e. 'Vertically transmitted infection'.

What should have happened instead?:

  • All non-redirect results should have both terms in the title, like the three valid results returned:
    • viz.: 'Epstein–Barr virus infection', 'Nipah virus infection', 'Human papillomavirus infection' (it's not clear if the last one should have been returned, as 'virus' is not a subword of that title)

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript

Adding some subscribers from T424820 – 'Unexplainable .* behavior in intitle regex search' who might be interested in this: @Anomie @Certes @Novem_Linguae @Ponor @EBernhardson.

Note that a query with dual intitle terms where one is minused to exclude the 'infection' term seems to work properly:

Eyeballing these, I don't see any false positives here, i.e., no unwanted 'XYZ virus infection' articles.

@Mathglot thanks for reporting this. T424820 is most likely unrelated since the case you report does not use regular expressions.

All non-redirect results should have both terms in the title, like the three valid results returned

If you rely on the information reported in the results to determine if there's a redirect unfortunately if there's a word match in the title a matching redirect might not be displayed.
For instance:

I did not check them all but I suspect there are redirects with the word infection in them.

Note that redirects handling in CirrusSearch can be misleading and T204089 is related to this.
In short a query like intitle:word1 intitle:word2 does not guarantee that word1 and word2 will be part of the same title.
A page named Some page with two distinct redirects Word1 and Word2 might match the above query.
With the current behavior one way to add more constraint is to use phrase searches:

  • intitle:"word1 word2" will ensure that word1 is followed by word2 at a maximum distance of 1 word in the same title.

But sadly this workaround comes with some side-effects you might not want:

  • word1 and word2 to appear in this particular order.
  • the use of " also instructs the engine to not match the stem field, in short searching for infection might no longer match infections.

In short I think the confusion is caused by two factors:

  • Special:Search does not necessarily highlight a matching redirect, possibly giving the impression that no redirects match
  • Two instances of intitle: in the search query does not necessarily filter pages with redirects that have both terms, the terms can be spread across multiple redirects.

But please let me know if I misunderstood the issue you raised.