如何使用 PostgreSQL 全文搜索返回部分短语匹配而不返回太多行？

Question

I am using the pg_search gem to perform full-text search in PostgreSQL, and it is working well. However, some searches are not returning any video results when they should be.

For example, searching for "states of matter" returns 10 results since the videos have a tag named "states of matter." But "3 states of matter" returns zero results. Similarly, "radiocarbon dating" returns 1 video, but "carbon dating" returns nothing.

Here's how I've set up my model:

# app/models/video.rb
class Video < ApplicationRecord
  include PgSearch::Model

  ...

  pg_search_scope(:user_search, {
    against: {
      title: 'C',
      description: 'D'
    },
    associated_against: {
      tags: { name: 'A' }
    },
    using: {
      tsearch: {
        prefix: true,
        dictionary: "english"
      }
    }
  })

  ...

end

To run a search:

query = "3 states of matter"
results = Video.user_search(query)

If I use other options such as trigram or any_word , it returns too many irrelevant results.

How can I improve my full-text search functionality with partial matches and avoid returning too many irrelevant results? I would much prefer a solution that uses pg_search , but if I need to move away from the gem, then I will.

Answer 1

Hope you tried the trigram option with the default threshold(ie 0.3 ). You can try increasing the threshold to look for strict matches. Checkout gem docs for more info - Trigram#threshold .

By default, trigram searches find records which have a similarity of at least 0.3 using pg_trgm's calculations. You may specify a custom threshold if you prefer. Higher numbers match more strictly, and thus return fewer results. Lower numbers match more permissively, letting in more results. Please note that setting a trigram threshold will force a table scan as the derived query uses the similarity() function instead of the % operator.

Check out the below and decide based on your table size,

Please note that setting a trigram threshold will force a table scan as the derived query uses the similarity() function instead of the % operator.

To understand how trigram word similarity is being calculated. Please check this https://www.postgresql.org/docs/9.6/pgtrgm.html and you can set the threshold based on what level of comparison you need.