Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning to Represent Individual Differences for Choice Decision Making

Yan-Ying Chen, Yue Weng, Alexandre Filipowicz, Rumen Iliev, Francine Chen, Shabnam Hakimi, Yanxia Zhang, Matthew Lee, Kent Lyons, Charlene Wu

TL;DR

The paper addresses the challenge of predicting human decisions amid complex factors and individual differences by introducing representation learning to produce high-dimensional, personalized embeddings. It presents Beh2vec, which learns user-specific representations from past choices and open-ended text, then conditions gamble predictions on these embeddings. The approach yields improved prediction accuracy over non-representation baselines and is competitive with a well-established theory-based model, with text-based signals offering the strongest gains, especially for losses. The work demonstrates the flexibility of multimodal representations for responsible personalization and discusses data-synthesis and future work to extend the method to broader decision-making tasks.

Abstract

Human decision making can be challenging to predict because decisions are affected by a number of complex factors. Adding to this complexity, decision-making processes can differ considerably between individuals, and methods aimed at predicting human decisions need to take individual differences into account. Behavioral science offers methods by which to measure individual differences (e.g., questionnaires, behavioral models), but these are often narrowed down to low dimensions and not tailored to specific prediction tasks. This paper investigates the use of representation learning to measure individual differences from behavioral experiment data. Representation learning offers a flexible approach to create individual embeddings from data that are both structured (e.g., demographic information) and unstructured (e.g., free text), where the flexibility provides more options for individual difference measures for personalization, e.g., free text responses may allow for open-ended questions that are less privacy-sensitive. In the current paper we use representation learning to characterize individual differences in human performance on an economic decision-making task. We demonstrate that models using representation learning to capture individual differences consistently improve decision predictions over models without representation learning, and even outperform well-known theory-based behavioral models used in these environments. Our results propose that representation learning offers a useful and flexible tool to capture individual differences.

Learning to Represent Individual Differences for Choice Decision Making

TL;DR

The paper addresses the challenge of predicting human decisions amid complex factors and individual differences by introducing representation learning to produce high-dimensional, personalized embeddings. It presents Beh2vec, which learns user-specific representations from past choices and open-ended text, then conditions gamble predictions on these embeddings. The approach yields improved prediction accuracy over non-representation baselines and is competitive with a well-established theory-based model, with text-based signals offering the strongest gains, especially for losses. The work demonstrates the flexibility of multimodal representations for responsible personalization and discusses data-synthesis and future work to extend the method to broader decision-making tasks.

Abstract

Human decision making can be challenging to predict because decisions are affected by a number of complex factors. Adding to this complexity, decision-making processes can differ considerably between individuals, and methods aimed at predicting human decisions need to take individual differences into account. Behavioral science offers methods by which to measure individual differences (e.g., questionnaires, behavioral models), but these are often narrowed down to low dimensions and not tailored to specific prediction tasks. This paper investigates the use of representation learning to measure individual differences from behavioral experiment data. Representation learning offers a flexible approach to create individual embeddings from data that are both structured (e.g., demographic information) and unstructured (e.g., free text), where the flexibility provides more options for individual difference measures for personalization, e.g., free text responses may allow for open-ended questions that are less privacy-sensitive. In the current paper we use representation learning to characterize individual differences in human performance on an economic decision-making task. We demonstrate that models using representation learning to capture individual differences consistently improve decision predictions over models without representation learning, and even outperform well-known theory-based behavioral models used in these environments. Our results propose that representation learning offers a useful and flexible tool to capture individual differences.

Paper Structure

This paper contains 13 sections, 6 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset and Problem Definition
  4. Representation Learning
  5. Text Responses for Open-Ended Questions
  6. Behavioral Model
  7. Experiments
  8. Experimental Setting
  9. Prediction Accuracy
  10. Representation Learning and Behavioral Model
  11. Opportunities for Responsible Personalization
  12. Data Synthesis for Cold Start and Future Work
  13. Conclusion

Figures (3)

  • Figure 1: (a) Example of choice questions and demographic attributes and (b) the open-ended questions participant's answered with free text in the behavioral experiment.
  • Figure 2: A user's choice can be different from another user's choice for the same scenario (gambling) features because of individual differences as appeared in the choice dataset - two of five participants chose 1 while the others chose 2. Including individual differences can reduce the confusion for prediction.
  • Figure 3: Model architectures - (a) the architecture of Beh2vec: an embedding layer is added to learn user-specific representations before being concatenated with the scenario features and then fed to the fully connected layers. (b) the architecture for taking inputs of free text responses uses a pre-trained embedding model Word2vec to obtain word/phrase vectors and use the centroid of these vectors as the input layer.