• Resolved SpeedOfLike

    (@speedoflike)


    Thank you for your work on this plugin (which was recommended in a Gemini chat, so you’re doing something right there!). I can see how it can be extremely helpful and hope it can work for my use case.

    I am trying to match similar words like “sad” and “unhappy” or multiple spellings of the same word (which happens with words translated from another language), or even misspellings of words. Is that the right use case for this plugin?

    I’m using it on a custom post type (not woocommerce), which I selected in the setup process. There are 54 posts in that CPT with no relevant custom fields. For testing, I set my threshold to 10%.

    Here’s a sample of what’s happening when I search:

    • “depresssion” shows 4 results, ranging from 16% to 39% match precission. All of the matched posts contain that exact word in the title or the post body
    • “depressed” comes up with one result with 14%
    • “deepressed” and “unhappy” don’t show any results.

    I first generated embeddings with the built-in system, then with my own OpenAI API key (set to text-embedding-3-large), and got pretty much the same results with both. (I did clear all embeddings/cache before regenerating embeddings using OpenAI.)

    Any help would be appreciated!

Viewing 2 replies - 1 through 2 (of 2 total)
  • Plugin Author Samuel Silva

    (@samuelsilvapt)

    Hey @speedoflike, thanks for the kind words. Yes, your use case is exactly what this plugin is built for: semantic matching between “sad”/”unhappy”, handling typos, cross-language variations. So the fact that “unhappy” returns nothing while exact-word matches do is the opposite of what you’d expect.

    Howerver, this expectation is something I would like to clarify: Right now, the plugin doesn’t compare the search query against single words (actually it’s on the roadmap for the next versions). Instead, we compare the search query with the page content.

    So if you have the word “sad” in a long text, “unhappy” won’t match the page at all. It works more like: if your page is about “sadness”, the “unhappy” keyword will be close to your post.

    To double-check that everything is set up correctly on your side, I recommend creating a new page with only the word “sad” and then searching for “unhappy”. It should work as expected; if not, please let me know so we can investigate further.

    Best,

    Samuel

    Plugin Author Samuel Silva

    (@samuelsilvapt)

    @speedoflike what do you think about creating dedicated text embeddings to categories, terms and tags? In this way we could have more accuracy in the cases like you mentioned.

Viewing 2 replies - 1 through 2 (of 2 total)

You must be logged in to reply to this topic.