Abstract: To reduce the loss of accuracy, with numerous attributes of items available, a group recommendation method based on enhanced group preference is put forward. This approach takes items’ ratings and its attributes into consideration. From the aspect of items’ ratings, the method generates item candidate set for group recommendation leveraging collaborative filtering. And as for the items’ attributes, the preference of group user to each attribute is defined. Taking the group preferences as the weights of different attributes of items, the method clusters the items based on the weighted similarity of items to produce items preferred by group. Then the produced items are added into item candidate set to enhance group preference. An experiment is conducted with the dataset from Da Zhong Dian Ping with the proposed method and three other methods. The result shows that the proposed method has better performance.

Key words: group recommendation, meta-path similarity, group preference, hybrid recommendation

摘要： 为减少评分数据稀疏性造成的群推荐精度损失，借助用户生成的项目属性特征，提出一种增强群体偏好的混合群推荐方法。一方面，针对用户-项目评分信息，采用协同过滤手段产生群推荐项目候选子集。另一方面，利用群体生成的项目属性分布特征，挖掘群体对项目属性的偏好，并以项目属性权重的方式融入到项目相似性计算中。通过聚类，产生反映群体偏好的项目集，将群体喜好的集合扩充到用于推荐的项目候选集中，实现群推荐项目候选集中群体偏好的增强。最后，从项目候选集中生成群推荐结果。将该方法应用大众点评网上餐厅的推荐，验证了项目属性特征对群推荐结果的积极影响。实验结果表明该方法在准确率和召回率上较经典群推荐方法都有大幅度提高。

关键词: 群推荐, 元路径相似性, 群体偏好, 混合推荐