Active Machine Learning for Consideration Heuristics

Abstract

W e develop and test an active-machine-learning method to select questions adaptively when consumers use heuristic decision rules. The method tailors priors to each consumer based on a "configurator." Subsequent questions maximize information about the decision heuristics (minimize expected posterior entropy). To update posteriors after each question, we approximate the posterior with a variational distribution and use belief propagation (iterative loops of Bayes updating). The method runs sufficiently fast to select new queries in under a second and provides significantly and substantially more information per question than existing methods based on random, market-based, or orthogonal-design questions.Synthetic data experiments demonstrate that adaptive questions provide close-to-optimal information and outperform existing methods even when there are response errors or "bad" priors. The basic algorithm focuses on conjunctive or disjunctive rules, but we demonstrate generalizations to more complex heuristics and to the use of previous-respondent data to improve consumer-specific priors. We illustrate the algorithm empirically in a Web-based survey conducted by an American automotive manufacturer to study vehicle consideration (872 respondents, 53 feature levels). Adaptive questions outperform market-based questions when estimating heuristic decision rules. Heuristic decision rules predict validation decisions better than compensatory rules.

Consumer Preference Elicitation of Complex Products Using Fuzzy Support Vector Machine Active Learning

“…Because each response error has the potential to set the adaptive question selection to the wrong path and negatively impact selection of all subsequent questions, the presence of such errors poses a long-standing challenge to the adaptive question design literature (e.g., Hauser and Toubia 2005;). To date, response errors have either been neglected (e.g., Toubia, et al 2004;Netzer and Srinivasan 2011) or set as a priori possibility for all individuals and all questions (e.g., Toubia et al 2003;Abernethy et al 2008;Dzyabura and Hauser 2011;Toubia et al 2013). We demonstrate that the proposed method can be used not only to gauge possible response errors on the fly but also to reduce the effects of such noises in adaptive question selection.…”

“…Lastly, inspired by Dzyabura and Hauser (2011) who suggest previous-respondent data may be used to improve elicitation of consideration heuristics, the proposed method utilizes responses from previous respondents via collaborative filtering to facilitate active learning of the focal respondent's product preferences. The concept of collaborative filtering has been applied in various contexts such as prediction of TV show preferences and movie recommendation systems (Breese et al 1998).…”

“…(e.g., Green and Srinivasan 1990;Sattler and Hensel-Borner 2000) Second, despite its high efficiency in uncovering consumers' product preferences, adaptive question selection method is often subject to response errors that could potentially misguide the selection of each subsequent question. Lastly, with the exception of Dzyabura and Hauser (2011) in the context of consideration heuristics elicitation, this line of research has yet to explore the possibility of utilizing other respondents' data to facilitate active learning of the focal respondent's product preferences.…”

“…Our synthetic data experiments further demonstrate that the proposed method is able to rapidly and effectively elicit individual-level preference estimates even when the product category is equipped with more than 100 attribute levels. We also use synthetic data experiments to compare the scalability, parameter recovery, and predictive validity of the proposed algorithm with that of Dzyabura and Hauser (2011) for consideration questions and of Abernethy et al (2008) for choice questions. We show that the proposed question selection algorithm may be used either in conjunction with or as a substitute for the algorithms by Dzyabura and Hauser (2011) and Abernethy et al (2008) to uncover consumers' preferences for complex products.…”

“…We also use synthetic data experiments to compare the scalability, parameter recovery, and predictive validity of the proposed algorithm with that of Dzyabura and Hauser (2011) for consideration questions and of Abernethy et al (2008) for choice questions. We show that the proposed question selection algorithm may be used either in conjunction with or as a substitute for the algorithms by Dzyabura and Hauser (2011) and Abernethy et al (2008) to uncover consumers' preferences for complex products. Overall, our empirical and synthetic studies suggest that the proposed approach offers a promising new method to complement existing preference elicitation methods.…”