Relevance of schema fields

[cpojer]

Is your feature request related to a problem? Please describe. It would be great if the schema could allow specifying the relevance for individual fields. For example, if you have a schema with a title and a content field, then in many cases a match on the title implies that the result is more relevant than others where the keyword just matches on the content.

Describe the solution you'd like Here is an example API that makes title 10x more important for ranking than content:

const db = create({
  schema: {
    content: {relevance: 10, type: 'string'},
    title: {relevance: 100, type: 'string'},
  },
});

Describe alternatives you've considered One could issue multiple calls to search, for each property, and then manually merge the results and rank them. That's a lot of work!

[iShibi]

If this feature does get taken under consideration then I suggest specifying relevancy during the search in the properties field is a better option. For eg:

const searchResult = search(db, {
  term: 'foo',
  properties: [{ name: 'content', relevance: 10 }, { name: 'title', relevance: 100 }]
});

For comparison sake, this is the current way of using the properties field during the search:

const searchResult = search(db, {
  term: 'foo',
  properties: ['content', 'title']
});

This will keep the schema simple and let us change the relevancy of properties per search.

[cpojer]

Great idea @iShibi, I like that proposal!

[micheleriva]

@cpojer, I think @iShibi's solution will be doable once we get https://github.com/LyraSearch/lyra/pull/63 (or a similar plugin) merged. We could easily use a relevance value from 0.01 to <1 as a multiplication factor for TF-IDF scores

[micheleriva]

Update: we have the first alpha version for a TF-IDF plugin: https://github.com/LyraSearch/plugin-token-relevance

[micheleriva]

Update: we've moved the plugin as part of the core of Lyra. Would appreciate reviews here: https://github.com/LyraSearch/lyra/pull/169

[micheleriva]

Released in v0.4.2