Abstract In this study, we focus on understanding and mining user’s preferences and intentions via user-based aggregation in the context of a social network. Understanding preference and intention in microblog texts is more difficult and challenging than understanding such characteristics in the context of standard text. The main reason is that search history and click history are difficult to obtain due to data privacy in social networks. Meanwhile, the text is sparse, and the number of background topics in social networks is enormous. To overcome the above challenges, we explore an indirect method of user’s preference and intention understanding by leveraging a user-based aggregation topic model (UATM). Our UATM aims to mine the distributions of user’s preferences and intentions by utilizing user’s preference and intention distributions and followees’ preference and intention distributions. Furthermore, to alleviate the sparsity problem, we discriminatively model common words and topic words and incorporate a user factor into our model. We combine the recurrent neural network (RNN) and inverse document frequency (IDF) as the weight prior to learn word relationships. Moreover, to further weaken the sparsity of context, we leverage word pairs to model topics for all documents. We also propose a collapsed Gibbs sampling algorithm to infer preference and intention in our UATM. To verify the effectiveness of the proposed method, we collect a Sina Weibo dataset consisting of microblog users and their pushed content to conduct various experiments. Both qualitative and quantitative evaluations demonstrate that our proposed UATM model outperforms several state-of-the-art methods.

中文翻译：

---

一种基于用户的聚合主题模型，用于理解社交网络中用户的偏好和意图

摘要 在这项研究中，我们专注于在社交网络的背景下通过基于用户的聚合来理解和挖掘用户的偏好和意图。理解微博文本中的偏好和意图比理解标准文本上下文中的这些特征更加困难和具有挑战性。主要原因是由于社交网络中的数据隐私，搜索历史和点击历史难以获得。同时，文本稀疏，社交网络中的背景话题数量巨大。为了克服上述挑战，我们通过利用基于用户的聚合主题模型（UATM）探索了一种间接的用户偏好和意图理解方法。我们的 UATM 旨在通过利用用户的偏好和意图分布以及追随者的偏好和意图分布来挖掘用户的偏好和意图分布。此外，为了缓解稀疏性问题，我们对常用词和主题词进行了有区别的建模，并将用户因素纳入我们的模型中。我们结合循环神经网络 (RNN) 和逆文档频率 (IDF) 作为学习单词关系之前的权重。此外，为了进一步削弱上下文的稀疏性，我们利用词对来为所有文档建模主题。我们还提出了一种折叠吉布斯采样算法来推断我们 UATM 中的偏好和意图。为了验证所提出方法的有效性，我们收集了一个由微博用户及其推送内容组成的新浪微博数据集，以进行各种实验。