摘要： 针对传统模型不能真实反映标签的生成过程以及无法加入用户角色等问题，提出一种新的用户-内容联合标注模型。该模型中标签的生成同时受用户兴趣和资源主题的影响，反映用户真实的标注过程，能够学习到包括用户的兴趣分布概率、资源的主题分布概率、词语的主题分布概率以及标签的主题分布概率等参数。实验结果表明，与CI-LD、ACorrLDA等模型相比，该推荐模型的正确率提高了10%。
关键词:
社会标注,
标签,
隐含主题模型,
狄利克雷分配模型,
用户-内容联合标注模型,
Gibbs 抽样
Abstract: Aiming at the problem that tag generation process is not reasonable, or does not add user role and so on in this kind of model, a new generative model called user-content joint labeling model is presented. In the model, social tags are generated by user profile and resource topic together, which reflect the real label process of users. What’s more is that this model can learn a lot of useful parameters: the topical distribution of each resource, the interest distribution of each user, the word distribution of each topic and the tag distribution of each topic. Experimental results show that the proposed model’s accuracy raises by 10% compared with other two models proposed in previous research.
Key words:
social annotation,
tag,
latent topic model,
Dirichlet allocation model,
user-content joint annotation model,
Gibbs sampling
中图分类号: