prospective专题

Revisit Recommender System in the Permutation Prospective

总结 提出permutation概念,rnn做rank 细节 user emb和item emb 考虑稀疏特征+稠密特征 permutation-matching offline: 2个模型,分别预测ctr以及next score(是否继续阅览),不共享参数的dnn online: 改进的beam search permutation-rank offline: bi-lstm + ml