Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tacit Understanding Game (TUG): Predicting Interpersonal Compatibility

Yueshen Li, Krishnaveni Unnikrishnan, Aadya Agrawal

TL;DR

The paper tackles the challenge of predicting interpersonal compatibility in a privacy preserving, scalable way. It introduces the Tacit Understanding Game, a lightweight two player word association game that yields dense behavioral traces and couples real data with large language model–generated synthetic data. The authors demonstrate a privacy minded data collection pipeline, a hybrid real plus synthetic dataset, and an interactive ML workflow using a Siamese SBERT based architecture that achieves meaningful alignment with human compatibility judgments. The work offers a practical approach to privacy aware social sensing with potential applications in dating, collaboration, and social platforms, while acknowledging limitations and outlining paths for broader validation and fairness checks.

Abstract

Research on relationship quality often relies on lengthy questionnaires or invasive textual corpora, limiting ecological validity and user privacy. We ask whether a sequence of single-word choices made in a playful setting can reveal personality and predict interpersonal compatibility. We introduce the Tacit Understanding Game (TUG), a two-player online word association game. We collect word choice traces, annotate a subset with psychological ground truth scales, and bootstrap a larger synthetic corpus via large language model simulation. TUG demonstrates that minimal, privacy preserving signals can support relationship matching, offering new design space for social platforms.

Tacit Understanding Game (TUG): Predicting Interpersonal Compatibility

TL;DR

The paper tackles the challenge of predicting interpersonal compatibility in a privacy preserving, scalable way. It introduces the Tacit Understanding Game, a lightweight two player word association game that yields dense behavioral traces and couples real data with large language model–generated synthetic data. The authors demonstrate a privacy minded data collection pipeline, a hybrid real plus synthetic dataset, and an interactive ML workflow using a Siamese SBERT based architecture that achieves meaningful alignment with human compatibility judgments. The work offers a practical approach to privacy aware social sensing with potential applications in dating, collaboration, and social platforms, while acknowledging limitations and outlining paths for broader validation and fairness checks.

Abstract

Research on relationship quality often relies on lengthy questionnaires or invasive textual corpora, limiting ecological validity and user privacy. We ask whether a sequence of single-word choices made in a playful setting can reveal personality and predict interpersonal compatibility. We introduce the Tacit Understanding Game (TUG), a two-player online word association game. We collect word choice traces, annotate a subset with psychological ground truth scales, and bootstrap a larger synthetic corpus via large language model simulation. TUG demonstrates that minimal, privacy preserving signals can support relationship matching, offering new design space for social platforms.

Paper Structure

This paper contains 24 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Gamified Crowdsourcing and Games With a Purpose
  4. Personality, Language, and Interpersonal Compatibility
  5. Privacy-Preserving Data Collection and Synthetic Data
  6. Interactive Feedback Loops and Human-in-the-Loop Refinement
  7. Methodology
  8. Word Matrix Design
  9. Game Design
  10. Crowdsourcing Dataset
  11. (1) Pair Questionnaire.
  12. (2) Individual Questionnaire.
  13. Iterative Design Improvements
  14. Synthetic Data Generation
  15. Model Architecture
  16. ...and 9 more sections

Figures (5)

  • Figure 1: Siamese architecture for compatibility prediction from gameplay logs.
  • Figure 2: Game's home page and leaderboard.
  • Figure 3: The word choice interface displays a grid of words from which the user selects based on the given clue, the round result section displays the match rate and highlight matched words between the players.
  • Figure 4: This screen shows how players complete pair questionnaire to provide compatibility labels.
  • Figure 5: This screen shows how players complete individual questionnaire to provide personality traits label.