In an attempt to balance precision and recall in the search page, leading\ndigital shops have been effectively nudging users into select category facets\nas early as in the type-ahead suggestions. In this work, we present\nSessionPath, a novel neural network model that improves facet suggestions on\ntwo counts: first, the model is able to leverage session embeddings to provide\nscalable personalization; second, SessionPath predicts facets by explicitly\nproducing a probability distribution at each node in the taxonomy path. We\nbenchmark SessionPath on two partnering shops against count-based and neural\nmodels, and show how business requirements and model behavior can be combined\nin a principled way.\n