Attention Based Element-wise DRMM (ABEL-DRMM)

Attention Based Element-wise DRMM is an extension of deep relevance matching model (DRMM). It’s a new model for document relevance ranking. Like DRMM, ABEL-DRMM is document-length agnostic. Unlike DRMM, it is end-to-end trainable.

ABEL-DRMM takes context-aware query terms (q-term) and document (d-term) encodings and generates mixed dimension document-aware q-term encodings used in DRMM.