With the rising popularity of social media, information overload has become a serious problem. Users are easily overwhelmed by huge amount of information disseminated daily through their social friends, and thus might feel difficult to find useful information. To help users discover interesting content from the overwhelming information streams, a better user modeling strategy is needed. This thesis focuses on several key problems of user modeling on Twitter, one of the most successful social media platforms. The main contributions of our work include: 1. We propose an interest-related latent topic model to represent user interest over latent topics. Traditional user modeling frameworks on social media simply build a bag-of-words profile for each user based on his posts, and try to extract important information from this profile, such as key words, entities, categories or latent topics, to represent user interest. However, since a lot of tweets do not necessarily indicate user interest on social media, previous works fail to capture the real motivation of tweets and thus easily suffer from the large amount of interest-unrelated posts. We propose a modified author-topic model by introducing a latent variable to indicate whether a tweet is related to its author’s interest. By ruling out those tweets which do not represent user interest, our model can reach a better modeling of user interest on social media. 2. We present a novel mixture framework to model user posting behavior on social media. While user generated content is the basic element of social media websites, relatively few studies have systematically analyzed the motivation to create and share content, especially from the perspective of a common user. Inspired by those early works about user behavior on social media, we assume that user posting behavior is mainly influenced by three factors: breaking news, posts from social friends and user interest. By borrowing the idea from research area of text mining, we use a mixture model to represent user posting behavior, and present the inference of our model based on collapsed Gibbs sampling. Experiments show that our model outperforms the state-of-the-art user modeling frameworks. 3. Analyzing individual retweet behavior on social media. Retweet is not only the key mechanism for information diffusion on social media, but can also be viewed as an important signal of user interest and needs. While previous works about analyzing retweet have mainly...
修改评论