基于上下文的餐饮推荐算法

互联网的发展,为餐饮用户的上下文信息获取提供了基础。在用户选择适合其餐饮模式的前提下,加入动态上下文因素以满足用户的需求。为改善传统的协同过滤方法应用于餐饮O2O推荐存在的稀疏矩阵、冷启动等问题,设计了基于上下文后过滤的协同过滤推荐方法。先通过转化基于项目属性效用的评分矩阵,计算出用户对项目评分的偏好相似度。根据用户的评分偏好和静态上下文信息构建相似组,结合上下文信息加权的贝叶斯模型,采用基于KL散度的加权方法进行动态偏好分析,解决上下文信息缺乏时难以构建概率模型以及推荐系统的用户冷启动问题。实验结果显示,随着邻居数目增加时,基于上下文的推荐算法与传统的协同过滤算法相比,能维持较高的准确率和召回率,验证了推荐算法的有效性。

Algorithm on Catering Recommendation Based on Context

The development of the Internet provides the basis for the acquisition of context information for users of catering.Under the premise that users choose catering modes which are appropriate,dynamic context factors are added to satisfy users’demands.To improve the property of traditional collaborative filtering algorithm which involves defects such as sparse matrix and cold-start,the catering O2O recommendation algorithm is designed based on post-context filtering.First,by converting the rating matrix based on the utility of the items’attribute,the users’preference similarity to the items rating are calculated.Based on the users’rating preference and static context,a similar group is constructed which is combined with the weighted context of Bayesian model.Then the dynamic preference analysis is performed utilizing the weighted method based on KL divergence.The research successfully figured out problems of users’cold-start and modeling probability when context is scarce.The experiment shows that as the number of neighbors increases,the context-based recommendation algorithm maintains a higher accuracy and recall rate than the traditional collaborative filtering algorithm,verifying the effectiveness of the recommendation algorithm.