Option to only return results that match all tokens

For the purposes of this example you can assume I have set the fuzziness threshold to 0.

Given I have a list of items like this:

var items = [
  'large red shirt',
  'large green shirt',
  'large blue shirt',
  'medium red shirt',
  'medium green shirt',
  'medium blue shirt',
  'small red shirt',
  'small green shirt',
  'small blue shirt',
  'large red trousers',
  'large green trousers',
  'large blue trousers',
  'medium red trousers',
  'medium green trousers',
  'medium blue trousers',
  'small red trousers',
  'small green trousers',
  'small blue trousers',
  'large red socks',
  'large green socks',
  'large blue socks',
  'medium red socks',
  'medium green socks',
  'medium blue socks',
  'small red socks',
  'small green socks',
  'small blue socks'
];

I would like a search of large shirt to return the 3 results that match both words in the input:

[
  'large red shirt',
  'large green shirt',
  'large blue shirt'
]

A default search returns 0 results.

Enabling thetokenize option to search for individual words does successfully return these 3 results, however it also returns the other 12 large or shirt items. I am only interested in items that match both search tokens.

Could a matchAllTokens option be added to Fuse to achieve this?

https://jsbin.com/hifegi/17/edit?js,console

[keeganstreet]

Hey @krisk,

Thanks for taking the time to reply (and for building the library of course!)

I do appreciate that Fuse is all about string approximation, and I am using it for this feature. In my actual implementation I have not set the threshold to 0. I have just used that as an example in this ticket because I thought taking fuzziness out of the equation would make this issue clearer.

Maybe a more real-world example would better explain what I mean.

Say I have this list of companies, and I want to filter the list down based on a user entered search term:

AustralianSuper - Corporate Division
Aon Master Trust - Corporate Super
Promina Corporate Superannuation Fund
Workforce Superannuation Corporate
IGT (Australia) Pty Ltd Superannuation Fund

A search input of “Australia” will return 2 results.

And a search input of “corporate” will return 4 results.

A search input of “Australia corporate”, which in the user’s mind is a more specific search term, will return all 5 results. It seems counter-intuitive for a more specific search term (“Australia corporate”) to return more results than a less specific search term (“Australia”).

I understand that this is useful behaviour in some use-cases, because we may want Fuse to uncover more results even if some of the tokens don’t match. But in this use case, we want to reduce the number of results as more tokens are provided as input.

Here’s a demo: https://jsbin.com/zoposik/4/edit?js,console

Also its not really important, but you may have been mistaken about the threshold parameter not having an effect when tokenize is true. On https://jsbin.com/pehixamoba/edit?js,console, this returns 15 results:

fuse = new Fuse(items, {
  threshold: 0,
  tokenize: true
});
result = fuse.search('large shirt');

And this returns 27:

fuse = new Fuse(items, {
  threshold: 1,
  tokenize: true
});
result = fuse.search('large shirt');

[krisk]

@keeganstreet, your example illustrates the problem quite nicely. All the other matched results might be superfluous.

Very well, I’m sold. I’ll add some logic + option to Fuse.js which would address the issue. I will post updates on this thread.

And about this:

Also it’s not really important, but you may have been mistaken about the threshold parameter not having an effect when tokenize is true.

You’re absolutely right. Mea culpa. I had made this change a while ago, and had forgotten about it😅