A1296
Title: Learning preference from observed rankings
Authors: Yu-Chang Chen - National Taiwan University (Taiwan) [presenting]
Chen Chian Fuh - National Taiwan University (Taiwan)
Shang En Tsai - National Taiwan University (Taiwan)
Abstract: Estimating consumer preferences is central to many problems in economics and marketing. A flexible framework is developed for learning individual preferences from partial ranking data by interpreting observed rankings as collections of pairwise comparisons with logistic choice probabilities. Latent utility is modeled as the sum of product attributes, item fixed effects, and a low rank user item factor structure, allowing both interpretability and information sharing across consumers and products. Selection in which comparisons are observed is also addressed. A comparison is recorded only if both items enter the consumer consideration set, which creates exposure bias toward frequently encountered items. Pair observability is modeled as the product of item level propensities and these propensities are estimated using a logistic model for the probability that an item is observable. Preference parameters are then estimated using an inverse probability weighted and ridge regularized log likelihood that reweights observed comparisons toward a target comparison population. To scale computation, a stochastic gradient descent algorithm based on inverse probability resampling is developed. In an application to transaction data from an online wine retailer, the method improves out of sample recommendation performance relative to a popularity benchmark, especially for predicting purchases of previously unconsumed products.
