Find-Your-Match

This paper proposes a deep learning-based approach that simulates the user-feedback loop in entity matching and relevance modeling without the need for costly and time-consuming data collection. Our method formulates the relevancy modeling problem as a text/entity matching task and leverages NLP techniques to learn matching pat- terns from logged interactions. We evaluate our approach on datasets from three popular domains and show its ef- fectiveness, achieving close human performance. Further- more, our approach demonstrates robust performance on complex negative samples, indicating its potential to benefit any domain that relies on user feedback.