Contextual restaurant recommendation utilizing implicit feedback

Selecting a good, appropriate restaurant for an event is a common problem for most people. In addition to the main features of restaurants (e.g. food style, price, and taste), a good recommendation system should also consider diners' context information. Although there are many context-aware restaurant recommenders, most of them only focus on location information. This research aims to incorporate a greater variety of useful contexts into the recommendation process. Instead of explicit user restaurant ratings, our system relies on diners' restaurant booking logs to recommend restaurants. Each booking record contains the dining context: event type, dining time, number of diners, etc. In this paper, we propose using the canonical decomposition Bayesian personalized ranking (CD-BPR) algorithm to model the context information in a restaurant booking record. Experiments were conducted using three years of booking logs from EZTable, the largest online restaurant booking service in Taiwan. Experiment results show that adding context information into BPR significantly outperforms the baseline BPR method.