On Extractive Summarization for Profile-centric Neural Expert Search in Academia

Identifying academic experts is crucial for the progress of science, enabling researchers to connect, form networks, and collaborate on the most pressing research problems. A key challenge for ranking experts in response to a query is how to infer their expertise from the publications they coauthored. Profile-centric approaches represent candidate experts by concatenating all their publications into a text-based profile. Despite offering a complete picture of each candidate's scientific output, such lengthy profiles make it inefficient to leverage state-of-the-art neural architectures for inferring expertise. To overcome this limitation, we investigate the suitability of extractive summarization as a mechanism to reduce candidate profiles for semantic encoding using Transformers. Our thorough experiments with a representative academic search test collection demonstrate the benefits of encoding summarized profiles for an improved expertise inference.