基于LDA的社会化标签系统推荐技术

LDA based tag recommendation in social tagging system

标签推荐中采用将三维模型拆分成多个二元关系的方法,导致用户信息的描述模糊、语义丢失、标签的个性化信息减弱问题,提出一种基于LDA模型的个性化标签推荐模型(LTR)。使用LDA模型的吉布斯采样算法对参数进行估计,利用模型输出的概率关系进行排序,选取最高的N个预测结果作为最终的个性化推荐。以CiteULike数据集为研究对象,实验结果表明,该模型考虑了具有丰富语义信息的摘要文本,发挥了涵盖用户意识的个性化标签作用来增强推荐的准确性,有效为用户推荐个性化标签,提高了推荐效果。

To solve the problem of the fuzzy description of user information,semantic loss and the weakening of personalized information of tag,which were caused by splitting 3D model into multiple two element methods,apersonalized tag recommendation model based on subject model was proposed.The Gibbs sampling algorithm of LDA model was used to estimate the parameters.The probability of model output was used to sort.The highest N prediction results were selected as the final personalized recommendation.Taking the CiteULike data set as the research object,experimental results show that the model considers the rich semantic information,and increases the accuracy of the recommendation.