frankly_match 0.2.0

Diversity-maximizing group matching for deliberative conversations. Assigns participants into groups that maximise Hamming distance across answer masks, ensuring each group contains heterogeneous viewpoints.

Frankly Match

Frankly Match is an ongoing research and engineering effort to develop the most effective ways to match people into groups for facilitating constructive dialogue.

Code for the "smart matching" feature that is currently used by the Frankly platform lives here. We welcome contributions, forks, comments, issues, etc. from the community. Thanks for stopping by!

Please visit https://github.com/berkmancenter/frankly-match for our repo, which also contains code for our hosted Frankly Match API.

Repo Contents

• dart/ — frankly_match pub.dev Dart package.
• api/ — Google Cloud Run service implementing the match API. Written in Python for research purposes.

Description

Frankly Match currently consists of one main algorithm which is based on maximizing Hamming distance between participant survey answers.

Each participant's survey answers are encoded as a binary string ("01001" = one bit per question). The endpoints available are as follows:

• bucketMatch — for pairs. Directly maximises pairwise Hamming distance.
• groupMatch — for groups of 3+. BFS-clusters by similarity, then composes groups by taking one member from each cluster.
• randomGroups — no survey data needed. Random assignment baseline.

We aim to develop other matching algorithms which may optimize for different heuristics (e.g. maximizing similarity, or creating other interesting configurations). We are also interested in expanding the diversity-matching algorithm to allow it to ingest and factor in more than simple binary answers (e.g. text, multiple-choice, drawings?!).

Usage

import 'package:frankly_match/frankly_match.dart';

// Participants mapped to their binary answer masks
final responses = {
  'alice': '01001',
  'bob':   '10110',
  'carol': '01010',
  'dave':  '11001',
};

// Match into pairs maximising Hamming distance
final pairs = bucketMatch(samples: responses);
for (final group in pairs) {
  print('Group ${group.id}: ${group.participantIds}');
}
// Group 1: [bob, alice]
// Group 2: [dave, carol]

// Match into groups of 3
final groups = groupMatch(participantResponses: responses, targetGroupSize: 3);
// groups[0].id => '1', groups[0].participantIds => ['alice', 'bob', 'carol']

// Optionally supply your own group IDs (e.g. pre-generated Firestore document IDs)
final groups2 = groupMatch(
  participantResponses: responses,
  targetGroupSize: 3,
  groupIds: ['room_abc', 'room_xyz'],
);

// Random groups (no survey data needed)
final random = randomGroups(responses.keys.toList(), 2);
// random[0].id => '1', random[0].participantIds => ['alice', 'bob']