基于排序学习的Top-k软件服务推荐方法

Top-k software service recommendation method based on learning to rank

针对由互联网时代信息过载导致的服务生产和获取不平衡问题,提出一种基于排序学习的Top-k软件服务推荐算法。首先,对用户-服务进行特征提取,通过隐语义模型提取出用户-服务的隐含特征,基于信息熵对用户-服务进行多样性特征建模提取出其多样性特征;然后,将两类特征进行线性组合,按一定的比例融合两种特征;最后,通过排序学习得到推荐列表。实验结果表明:对比三种基准方法(基于用户的协同过滤算法、基于物品的协同过滤算法和一种群体软件开发中的项目推荐方法),该算法在推荐精度上最大可提高16. 9%,且当用户-服务隐特征与多样性特征权重系数为4∶6时,可达到推荐精度为0. 702和推荐多样性为0. 632的平衡效果,从而确保精度又能提供更丰富的服务推荐列表。

Concerning the imbalance between service production and access resulted from the information overload in the lnternet era, a Top- k software service recommendation method based on learning to rank was proposed. Firstly, the features were extracted from user-service relation： the implicit features of user-service were extracted through the latent factor model, and the diversity features were extracted based on information entropy. Then, the two types of features were fused by linear combination according to a certain ratio. Finally, the list of recommendation was obtained through the algorithm of learning to rank. The experimental results show that compared to three baseline methods, the proposed method can improve the accuracy of recommendation by 16.9% to the most, and when the weight coefficient of implicit user-service features and diversity features is 4： 6, the accuracy of recommendation of 0.702 and recommendation diversity of 0. 632 are balanced, which ensures the accuracy and provides more abundant service recommendation list.