Discrete Listwise Content-Aware Recommendation

To perform online inference efficiently, hashing techniques, devoted to encoding model parameters as binary codes, play a key role in reducing the computational cost of content-aware recommendation (CAR), particularly on devices with limited computation resource. However, current methods for CAR fail align their learning objectives (e.g., squared loss) ranking-based metrics Normalized Discounted Cumulative Gain, NDCG), resulting suboptimal accuracy. In this paper, we propose novel method based Factorization Machine (FM), called Discrete Listwise FM (DLFM), fast and accurate recommendation. Concretely, our DLFM is optimize NDCG Hamming space preserving listwise user-item relationships. We devise an efficient algorithm resolve challenging problem, which can directly learn relaxed continuous solution space, without additional quantization. Particularly, theoretical analysis shows that optimal optimization problem approximately same original discrete problem. Through extensive experiments two real-world datasets, show consistently outperforms state-of-the-art hashing-based techniques.