Discovering Topical Aspects in Microblogs

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 860–871, Dublin, Ireland, August 23-29 2014. Discovering Topical Aspects in Microblogs Abhimanyu Das Microsoft Research abhidas@microsoft.com Anitha Kannan Microsoft Research ankannan@microsoft.com Abstract We address the problem of discovering topical phrases or “aspects” from microblogging sites like Twitter, that correspond to key talking points or buzz around a particular topic or entity of interest. Inferring such topical aspects enables various applications such as trend detection and opinion mining for business analytics. However, mining high-volume microblog streams for aspects poses unique challenges due to the inherent noise, redundancy and ambiguity in users’ social posts. We address these challenges by using a probabilistic model that incorporates various global and local indicators such as “uniqueness”, “diversity” and “burstiness” of phrases, to infer relevant aspects. Our model is learned using an EM algorithm that uses automatically generated noisy labels, without requiring manual effort or domain knowledge. We present results on three months of Twitter data across different types of entities to validate our approach. 1 Introduction Microblogging sites such as Twitter and Weibo are evolving into the social platforms of choice for users to express and discuss, in real-time, their thoughts and ideas on a plethora of subject matters. It is thus important to use these microblog streams to identify the “buzz” or “talking points” regarding any topic or entity of interest, including organizations, products and social issues. This has several applications: For businesses, identifying what its customers are mostly talking about allows them to better engage with their customer base (Burton and Soboleva, 2011; Patino et al., 2012), fine-tune brand awareness and mar- keting campaigns (Popescu and Jain, 2011), and provide real-time feedback about customer preferences and complaints. Similarly, policy makers and think tanks would benefit from understanding the buzz around various socio-cultural or environmental issues, that could enable them to make well-informed choices and decisions. Infering such key talking points in social media also enables higher layer social- analytics applications such as trend detection, event tracking, and fine-grained opinion and sentiment analysis. The goal of this paper is to automatically infer such entity-specific buzz in social media, which we represent using key phrases identified from microblog posts about the entity. Following past litera- ture (Kobayashi et al., 2007; Mukherjee and Liu, 2012), we call these topical phrases as aspects. Thus, given a stream of microblog posts about an entity of interest, we devise an algorithm that automatically discovers a ranked list of the top aspects that succinctly represent the buzz or key talking points among users about the entity. As an illustrating example, Figure 1 shows the top 10 aspects discovered for each month by our aspect discovery algorithm for the Microsoft Surface tablet using 6 month of Twitter data. For each month we depict the key events and news stories (below the timeline) related to the Surface, along with the set of discovered aspects (above the timeline). As seen from the figure, several of the top aspects do …

Content truncated to 500 words