How would you use the "derived personality" to identify those that you would get on well with? Even if you can come up with some kind of meaningful distance metric, it's definitely not clear that that's useful for matching individuals. For one thing it's completely ignoring the most important factor in dating, which is initial mutual physical attraction. For another, many (I would conjecture most) people are attracted to people who are quite different from themselves.
Could let users pick what type of relationship they're looking for: "forever-marriage", "summer fling", etc. For long-term marriage you could find couples to train a model for true positives. For the others you'd have to find couples who self-identify.
Then determine not just "distance" but "vector"/"clustering" and identify which cluster combinations seem to be compatible for different types of relationships.
This would also entrench present-day dating and marriage regimes.
Then you could handle the physical attractiveness the same way Tinder does. Use the clustering model to pre-filter candidates, and present the user with photos for final thumbsup/down.
