Abstract:

Billions of users share their ideas through posting photos and texts on Social Network Services（SNSs）. They are interested in various topics, usually have different sentiment tendencies and posting activities. This paper proposes a model to characterize SNS users’ posting activities for predicting users’ interests. Latent Dirichlet Allocation（LDA） is applied to discover typical patterns of users’ postings, such that one user’s posting sentiments are represented as probability distribution on the posting patterns. User behavior features are extracted from the LDA result, and combined with linguist features extracted from users’ likes pages to build an interest prediction model. The experiments results show that using user behavior features that characterized from users’ posting activities can improve the prediction accuracy.

Key words: user topology, interest prediction, sentiment analysis, user profiling, Latent Dirichlet Allocation（LDA） model

摘要：

数十亿用户通过在社交网络服务上发布照片和文本来分享他们的想法。他们对各种主题感兴趣，通常有不同的情感倾向和发布活动。提出了一个模型来表征社交网络用户的发布活动，以预测用户的兴趣。应用LDA来构建用户发帖的典型模式模型，以将用户的发贴行为表示为发帖模式的概率分布。从发布模式结果中提取出用户行为特征，并与从用户点赞的主页中提取的语言特征结合，构建兴趣预测模型。实验结果显示，使用从用户的发布行为中提取出的用户行为特征可以提高预测的准确性。

关键词: 用户拓扑, 兴趣预测, 情感分析, 用户画像, LDA模型