Abstract:

In view of the current situation that the current session recommendation algorithm does not fully consider the user’s context information, in order to enhance the personalized recommendation effect of the session-based recommendation algorithm, a context-aware recommendation algorithm fused with user session data is proposed. Context information is mapped to low-dimensional real vector features through embedding, and low-dimensional vector features are incorporated into a session-based recurrent neural network recommendation model through three combinations of Add, Stack, and MLP. A loss function based on BPR is designed to dynamically depict the user preference of session sequence, which enhances personalized recommendation capabilities. Experiments on the Adressa dataset show that compared with the baseline algorithm GRU4REC, the proposed algorithm improves the indicator Recall@20 by 3.2% and MRR@20 by 27%.

Key words: recommendation algorithm, context, session, recurrent neural network

摘要：

针对现有会话推荐算法未充分考虑用户的上下文信息的现状，为增强基于会话的推荐算法的个性化推荐效果，提出一种融合用户会话数据的上下文感知推荐算法。将上下文信息通过embedding映射成低维实数向量特征，通过Add、Stack、MLP三种组合方式将低维向量特征融入到基于会话的循环神经网络推荐模型，设计了基于BPR的损失函数动态刻画会话点击序列的用户偏好，以提升个性化推荐能力。在Adressa数据集上的实验表明，所提算法相比基线算法GRU4REC，在指标Recall@20上提高了3.2%，MRR@20上提高了27%。

关键词: 推荐算法, 上下文, 会话, 循环神经网络