While preference modelling is becoming one of the pillars of machine learning, the problem of preference explanation remains challenging and underexplored. In this paper, we propose PREF-SHAP, a Shapley value-based model explanation framework for pairwise comparison data. We derive the appropriate value functions for preference models and further extend the framework to model and explain context specific information, such as the surface type in a tennis game. To demonstrate the utility of PREF-SHAP, we apply our method to a variety of synthetic and real-world datasets and show that richer and more insightful explanations can be obtained over the baseline.

Values are normalized for comparability of FID progression, as FID scores are not upper bounded and as such, absolute values for different networks and pretraining methods differ. To analyze the impact of the network architecture, pretraining method and training data, respectively the learned feature representations, on the construction of train-test splits and the entailed difficulties, we repeat our class swapping and removal procedure introduced in Section 3 in the main paper using different self-supervised models. Subsequently, we select train-test splits from the same iteration steps. Figure 1 compares the progression of distribution shifts based on FID scores normalized to the [0,1] interval for valid comparison. We observe that across all pretrained models, the general FID progressions and sampled train-test splits exhibit very similar learning problem difficulties, indicating that our sampling procedure is robust to the choice of readily available, state-of-the art self-supervised pretrained models.

T/P refers to "tokens" per patient, C/T to codes per "token," and # Codes to the number of unique codes present in the dataset. Missing values reflect quantities not reported in the source publication.

A.1 Proof of Theorem 1 In this proof, we adopt a simplified version of our message-passing function that ignores the skipconnection: The HGNN trained in the experimental results shown in Figure 2 also does not use skip-connections and hence represents a theoretically-exact KTN component. In the real experiments, we use (1) skip-connections, exploiting their usual benefits (12), and (2) the trainable version of KTN. Without loss of generality, we prove the result for the case where R = {(s,t): s,t T }, meaning the type of an edge is identified with the (ordered) types of the neighbor nodes. In other words, there is only one edge modality possible, such as a social networks with multiple node types (e.g. "friendship" and "message"), the result is extended trivially (through with more algebraically-dense forms of ats and qts). The output of Aggregate is a concatenation of edge-type-specific aggregations (see Equation 3).