一种基于社会化媒体和社会网络结构的混合推荐模型

近年来,社交网络的迅速发展为在线用户之间的沟通和交流带来极大便利,为良好的信息推荐服务提供了丰富的资源,与此同时也为个性化推荐带来了更为复杂的技术挑战.本文通过自然语言处理技术获取用户在社会化媒体(新浪微博)中的个性化兴趣标签信息,应用到自行设计开发的社会化阅读应用牛赞网中.进一步地,利用用户在牛赞网中的阅读行为和社交信息,结合用户的社会化媒体兴趣,提出了一种混合推荐模型.实验基于牛赞网中的实际数据集,并与基于用户的经典协同推荐模型进行了对比,结果表明,提出的模型在推荐性能的几个指标(AUC、准确率、召回率、多样性和新颖性)上都有很大的提高.最后,通过对牛赞网中几个典型用户进行进一步的案例分析后得出,混合推荐模型的最优参数需要根据不同社会化行为的用户进行调节.

Hybrid Recommendation Model Based on Social Media and Social Networks

Recently,the rapid growth of social networks has provided rich contents and yet a huge challenge for recommender systems.To better uncover its underlying role in information filtering,a hybrid algorithm was proposed based on the integrated effect of social media and social networks.The social interests(so-called tags)from social media were extracted by natural language process technology;the social influence was measured based on social network analysis;and a tunable parameter to integrate those two effects was adopted to provide recommendation results.Numerical results on a real-world dataset,Newzan,show that the presented model outperforms the classical user-based collaborative filtering algorithm in various metrics,including AUC(area under the curve),precision,recall,diversity and novelty.Furthermore,the case studies on three typical users from Newzan were performed.Statistical analyses show that the optimal recommendation parameter should vary from user to user according to the degree of their social involvement.The results can provide an in-depth understanding for unifying social media to develop social applications.