融合社会关系的用户标签主题模型

摘要

[Purpose/Significance]Recognizing the latent topics of tags effectively is the key to the problem-solving of the semantic fuzziness of the social tagging systems.It is of important significance in modeling the user's interests, providing accurate tag recommendations for users, and improving the user's experience in social tagging systems.[Method/Process]To recognize the tag's topics from the perspective of the user's annotating behaviors, a user tag topic model fusing the user's social relations is proposed.Firstly, the user's social relations are modeled from the social tagging system.Secondly, to acquire the user's authority scores, the random walk method is utilized to analyze the user's social relation links, and then the user's authority scores are weighted on the binary relations between users and tags.Based on this, the tag LDA model based on user weighted is constructed, and the tag's latent topics are acquired by iterative learning.[Result/Conclusion]The experimental results show that mainly due to the effective fusing of the user's social relations, the proposed model has better topic expression performance than the un-weighted user tag topic model.%[目的/意义]标签潜在主题的有效识别是解决社会化标注语义模糊问题的关键,对于建模用户兴趣、为用户提供精准的标签推荐、改善社会化标注系统的用户使用体验具有重要的意义.[过程/方法]为了从标注行为的角度识别用户所标注标签的主题,提出一种融合社会关系的用户标签主题模型.首先,从标注系统中建模用户的社会关系;其次,利用随机游走的方法对用户的社会关系进行链接分析并获取用户的权威度分数,将其加权到"用户-标签"的二元标注关系上;此基础上,构建基于用户加权的标签LDA模型,通过迭代学习出标签的潜在主题.[结果/结论]实验结果表明,由于有效融合了用户的社会关系这一重要信息,提出的模型与未加权的用户标签主题模型相比,具有更好的主题表达性能.