The work focuses on predicting human decisions in a setting that exhibits context effects and behaviors strongly dependent on market segments such as a complex case of airline itinerary booking. Pairwise Choice Markov Chains may come handy in this specific scenario. However, when examples of alternatives are scarce, overfitting is a common phenomenon. Additionally, the class is inappropriate when new alternatives are present in the test set. PMCN-Net proposes an amortized inference approach for PCMC based on a NN that uses the alternatives’ and individuals’ features to determine the transition rates.