Ranking factors in devising practical POI search model

POI (Place of Interest) search is a fundamental component in various LBS (Location Based Service) applications. In this paper, we discuss and evaluate several critical factors in devising a practical POI search model based on users' search log data and our experience gained from developing and operating K.ai, one emerging and interesting LBS product serving Chinese users. Our discussion starts with single variable, and proceeds by adding one additional ranking factor into the model each step. Regarding each factor added, we explain why it is important, based on explicit users' feedback and implicit information extracted from the search log, and how we deal with it. We also mention our understanding and some early trials on personalization and friendly UI design in addition to the scoring algorithm. This kind of practical discussion is quite rare, hence, it is believed a valuable reference for both practitioner and researcher of this field.