Give people tools to understand what makes their profile good

[duogenesis]

Suggestions:

• what about seeing certain statistics as a feature? like seeing how many people you interacted with, how many see your profile, etc.
• A ML model which tells people how good their photos are and why

[kero3009destiny]

Private dict key on user about profile completeness, based on how many attributes the user has filled out etc.
Below is OkCupid's frontend implementation, not providing API implementation because their's is a MESS.

[duogenesis]

My vision for this feature is to provide users with in-depth, data-driven feedback specific to their profile, like a digital dating coach. For example, it'd be cool if an ML model could notice that they're not smiling in their pic and suggest that they do, or that they're not looking at the camera. It'd also be cool so provide real stats on how much of a difference it makes. Then when it's appropriate, you can show users messages like, "People who smile at the camera get 42% more replies," or something to that effect. Encouraging users to write more and answer more questions is cool too, and definitely an improvement on what's in Duolicious now.

[kero3009destiny]

Simple solutions should suffice, opencv for decent face detection; redflagged keywords analysis for bio, message return rate analysis etc.

OpenAI induced bankruptcy is a new business model, we should stay away from it probably.

[kero3009destiny]

Made a very comprehensive list, got assistance from Gemini 1.5 Pro for some parsing. https://gist.github.com/kero3009destiny/f3c04532989415a3cace52388257c6c5

[duogenesis]

I guess this ticket was very broadly addressed by this blog article. But it'd still be much better to have feedback customized to individual users. For example, the article's statistics about how the number of images on your profile increases your odds of getting a reply could be made interactive, in the app's UI.

https://gist.github.com/kero3009destiny/f3c04532989415a3cace52388257c6c5

        {
            "keyword": "obey",
            "reason": "Implies a desire for unquestioning obedience and submission, raising concerns about controlling and abusive tendencies.",
            "severity": "high"
        },
...
        {
            "keyword": "ywnbaw",
            "reason": "Stands for \"You Will Never Be A Woman,\" a derogatory and transphobic phrase often used to invalidate and demean transgender women.",
            "severity": "medium"
        }

Gemini's all over the place with these severity ratings lol. Flagging the word "obey" seems pretty wild, but some terms Gemini's listed should still be taken seriously. For example, "ywnbaw" was blocked from profiles with the rollout of #581 about two weeks ago. The feedback users get when writing a bio containing that term is a red message saying "too rude". It might be worthwhile combing through Gemini's list for more terms to add to the block list.

[kero3009destiny]

Oh definitely, the Gemini list was mostly to generate examples I may have missed and test it's reliability; which has shown it's somewhere between "almost useless" and "barely reliable"

As a project separate from Duolicious I would love to actively analyse image boards and online communities to track language narratives.

Also didn't notice the blog or that blog post; it was a very interesting read! Thanks for sharing.