基于图神经网络和时间注意力的会话序列推荐

Graph neural networks with time attention mechanism for session-based recommendations

为解决基于循环神经网络及其改进的方法在处理会话序列数据时只考虑序列行为,无法从有限的点击中获得准确的会话向量表示的问题,提出一种基于图神经网络和时间注意力的会话序列推荐算法。结合门控图神经网络和项目浏览时间信息,有效建模会话中所有点击项目之间的复杂转换,更充分利用用户浏览信息,使会话向量表示的计算更准确、区分度更高。实验结果表明,该方法能够提高推荐结果的准确性,更为有效地预测用户的下一次点击。

To solve the problem that the recurrent neural network and its improved methods only consider the sequence behavior when processing the sequence data,and cannot obtain the accurate session vector representation from the limited click,the graph neural networks with time attention mechanism for session-based recommendations algorithm was proposed.Gated graph neural network and the item browsing time information were used to model the complex conversion between all the click items in the session effectively and make more full use of user browsing information.The calculation of the session vector representation was then more accurate and discriminative.Experimental results show that the proposed method can improve the accuracy of the reco-mmendation results and predict the user’s next click more effectively.