The goal of this competition is to predict the level of compatibility of two given users to improve yenta’s profile recommendation algorithm. For this purpose, we classify the level of compatibility between user A and user B into 4 categories: • No Match = 0: At least one of either user A or user B swiped left on the other, meaning there is no possibility of match. • Match = 1: Both user A and user B swiped right on each other and matched. • Matched and met but unfavorable review = 2: Both user A and user B swiped right on each other and matched, then met. After the meeting, user A gave user B a review of 1-3 out of 5 (an “unfavorable” review). • Matched and met and favorable review = 3: Both user A and user B swiped right on each other other and matched, then met. After the meeting, user A gave user B a review of 4-5 out of 5 (a “favorable” review). To build this model, we provide subsets of data of 2 different types: user data and interaction data. Note that all of the data is anonymized through the use of IDs and multi-step vectorization models to ensure that user privacy is protected. IDs with low frequency are grouped into “other” categories with an ID of 999999. • User data: These files are connected through the user_id column (e.g. 41245) - user_ages.csv: User's age in years old - user_educations.csv: User's school and (in some cases) degrees - user_works.csv: Companies where user has worked along with information about the company's industry and size - user_skills.csv: User's professional skills - user_strengths.csv: User's strengths based on votes from other users after a meeting - user_purposes.csv: User's reasons for using the yenta app - user_self_intro_vectors_300dims.csv: User's vectorized profile biography (300 dimensions) - user_sessions.csv: User's app session logs • Interaction data: These files are indexed by from-to user_id pairs (e.g. 12345-52462) - interaction_review_comments_300dims.csv: User A’s vectorized opinion of user B after meeting (300 dimensions) - interaction_swipes.csv: User A’s swipe status of user B: 1 = right swipe (interested), -1 = left swipe (not interested) - interaction_review_strengths.csv: User A’s review of the strengths of user B after meeting • Train and test files: In order to train the model, we provide a train.csv file with pairs of user IDs and their corresponding scores - train.csv: Score between (from) user A and (to) user B - test.csv: A list of "from-to" IDs in the form of "userAid-userBid" that have to be predicted The solution file to be provided should follow the form below. The from-to IDs should be the same IDs contained in the test.csv file and they must be in the same order. • submission.csv from-to, score 6280229-6293525, 1 670384-50085, 2 2271906-4685859, 1 ... A few minutes after submitting your solution, you will be able to see the accuracy of your solution on the submission page over a subset of the test data. Final competition results will be based on the Private Leaderboard results, and the winner will be the person(s) at the top of the Private Leaderboard.