First we have each person in an existing relationship answer survey questions plus assess the quality of that relationship. Above couple P25/P26 has a great relationship (96% satisfaction) while couple P47/P48 has a poor relationship (29%). Then we train a machine learning engine to recognize patterns in the survey answers that predict quality of relationships. For an example question “I am religious”, someone answering “Strongly Agree” may not be a good match for someone answering “Strongly Disagree”. For the question “I want to have children”, both answering “Strongly Agree” would be a good sign.
Finally, for someone (A) seeking a relationship we ask those same survey questions and the trained machine learning engine assesses them with answers from others seeking relationships. Pairing A/B76 doesn’t look so good (56%) but pairing A/B65 looks promising (95%), so we recommend candidate B65 to person A.
Patent 11,276,127 provides further details about using photos, videos or social network data as a basis for recommending matches, and a novel neural network architecture to achieve this.