Reut
Apel

A Behavior-ML Approach to Persuasion Games
Technion
Reut Apel

Reut
Apel

A Behavior-ML Approach to Persuasion Games
Technion
Reut Apel

Bio

Reut is an MSc. student in the Data Science program of the faculty of Industrial Engineering and Management (IE&M) of the Technion, where she works with Prof. Roi Reichart and Prof. Moshe Tennenholtz. She conducts interdisciplinary research on the intersection of game theory, psychology, and NLP. She was one of the first three students and the first female student of the Technion IE&M’s new Data Science and Engineering BSc program.

 

During her studies she worked as data scientist and product analyst at Advanced Analytics at Intel. Reut established a female-exclusive DS student meeting club, where the participants have the opportunity to meet inspiring women from the field.

Bio

Reut is an MSc. student in the Data Science program of the faculty of Industrial Engineering and Management (IE&M) of the Technion, where she works with Prof. Roi Reichart and Prof. Moshe Tennenholtz. She conducts interdisciplinary research on the intersection of game theory, psychology, and NLP. She was one of the first three students and the first female student of the Technion IE&M’s new Data Science and Engineering BSc program.

 

During her studies she worked as data scientist and product analyst at Advanced Analytics at Intel. Reut established a female-exclusive DS student meeting club, where the participants have the opportunity to meet inspiring women from the field.

Abstract

In recent years there has been a significant development in NLP technology. This development has paved the way for new applications that have only recently been beyond reach. In this research we go one step further and present what is, to the best of our knowledge, the first attempt to build prediction models of human behavior within repeated decision-making interactions, where communication is naturally done by natural language.

 

We have compiled data that represent repeated interactions between a buyer and a seller who communicate in natural language. We then build prediction models of the parties’ decisions and their payoffs based on language representations as well as features derived from the decision-making literature. We were able to extract language-based features that well represent the impact of language on human decision making, and our models’ performance is 10% better than those of strong baselines. Our results show that accounting for language-based communication has a significant impact on human decision prediction, as using only past decisions yields substantially weaker predictions.

Abstract

In recent years there has been a significant development in NLP technology. This development has paved the way for new applications that have only recently been beyond reach. In this research we go one step further and present what is, to the best of our knowledge, the first attempt to build prediction models of human behavior within repeated decision-making interactions, where communication is naturally done by natural language.

 

We have compiled data that represent repeated interactions between a buyer and a seller who communicate in natural language. We then build prediction models of the parties’ decisions and their payoffs based on language representations as well as features derived from the decision-making literature. We were able to extract language-based features that well represent the impact of language on human decision making, and our models’ performance is 10% better than those of strong baselines. Our results show that accounting for language-based communication has a significant impact on human decision prediction, as using only past decisions yields substantially weaker predictions.

Planned Agenda

Planned Agenda