ProtoSiTex: Learning Semi-Interpretable Prototypes for Multi-label Text Classification

In three linesProtoSiTex is a semi-interpretable framework for fine-grained multi-label text classification. It combines unsupervised prototype discovery with supervised classification using hierarchical loss. Experiments on a new hotel reviews benchmark and two public benchmarks demonstrate SOTA performance with faithful, human-aligned explanations.

Summary generated by Claude — human-verified