A lexicographic rule is a non-compensatory decision rule. The consumer who uses this rule to make choices has a lexicographic preference. Dierent from compensatory rules, the consumer orders alternatives in the consideration set over dierent attribute levels. The lexicographic preference is a sequence of the orders on these attribute levels.  The consumer uses this sequence to evaluate alternatives rst on the most important attribute level, and then the second, and so forth. Some previous studies prove that most consumers apply a lexicographic rule to make choices when they face a lot of alternatives.  Online searching is one of these cases.  Consumers browse lots of results and need to choose one of them to purchase.  However, few literature in consumer search considers the use of lexicographic rules.

In this paper, we propose a sequential search model with Dirichlet learning based on this characterization of consumer behavior.  Our approach has three distinct advantages.  First, we inference consumers' lexicographic preference in the sequential search.  Second, we detect how consumers learn about the distribution of the products' attribute levels.  As the learning accumulates, consumers have better knowledge about the probabilities of whether unobserved products will have their favourite attribute levels and therefore decide when to stop.  Third, we endogenize a consumer's response to the default ranking with search renement tools, such as ltering of product features.  Accounting for the use of renement tools is necessary since the ranking and search actions shape the consideration sets for consumers' choices together.  Next, we formulate a data set with Monte Carlo Simulation and estimate the parameters of the model.  The results show that our method could replicate a consumer's choice precisely than the model using compensatory rules and the model without consumer learning.