BERTERS: Multimodal representation learning for expert recommendation system with transformers and graph embeddings

•We propose BERTERS, a transfer learning-based and multimodal approach for ERS.•BERTERS presents each expert candidate by a single vector representation.•The representation includes authority, text similarity and reputation scores.•BERTERS uses transformers and graph embeddings to convert modalities into vectors.•BERTERS can be extended into both the academic and question answering communities.