基于门控图游走网络的推荐多样性研究

Research on recommendation diversity via gated graph walk network

近年来,纯粹追求准确性的推荐算法已不再符合用户日益增长的多元化需求.因为该类算法将所有用户同等对待,导致推荐结果趋于单一化.从推荐系统的多样性角度出发,提出由两路图游走网络和门控网络组成的门控图游走网络.图游走网络在原有邻域上扩展一类新邻域,聚合两类邻域的信息,从而生成偏向准确性或多样性的推荐结果.门控网络对两个不同偏好推荐结果进行选择,得到最终推荐结果.不同于其他推荐多样性算法,门控图游走网络的推荐结果准确性-多样性比例可由超参数λ调整,而不是完全由算法决定.3个真实数据集的实验结果验证了门控图游走网络在多样化整体协作推荐方面的有效性.

Recently,purely accuracy-based recommendation algorithms no longer meet the growing diverse needs of users,because these algorithms treat all users equally,leading to the result being homogeneous.To solve this problem,we propose a gated graph walk network(Gated-GWN)consisting of two graph walk networks(GWNs)and a gated network.GWN expands a new neighborhood on the original neighborhood and aggregates the information from these neighborhoods to generate the result with accuracy or diversity.The gated network selects two results with different preferences to obtain the final results.Unlike other recommendation diversity algorithms,the accuracy-diversity ratio of the result is adjusted by the hyperparameter in GWN,rather than being entirely determined by the algorithm.With experiments on real-world datasets,we verify the effectiveness of the Gated-GWN in diversifying overall collaborative recommendations.