From Tweets to Wellness: Wellness Event Detection from Twitter Streams

Social media platforms have become the most popular means for users to share what is happening around them. The abundance and growing usage of social media has resulted in a large repository of users’ social posts, which provides a stethoscope for inferring individuals’ lifestyle and wellness. As users’ social accounts implicitly reflect their habits, preferences, and feelings, it is feasible for us to monitor and understand the wellness of users by harvesting social media data towards a healthier lifestyle. As a first step towards accomplishing this goal, we propose to automatically extract wellness events from users’ published social contents. Existing approaches for event extraction are not applicable to personal wellness events due to its domain nature characterized by plenty of noise and variety in data, insufficient samples, and inter-relation among events. To tackle these problems, we propose an optimization learning framework that utilizes the content information of microblogging messages as well as the relations between event categories. By imposing a sparse constraint on the learning model, we also tackle the problems arising from noise and variation in microblogging texts. Experimental results on a real-world dataset from Twitter have demonstrated the superior performance of our framework. Recent years have witnessed the revolutionary changes brought by the development of social media services through which individuals extensively share information, express ideas, and construct social communities. These changes can advance many disciplines and industries, and health is no exception (Nie et al. 2015; Lee et al. 2014). In such a context, many users are keen to share their wellness information on social platforms such as Twitter and Facebook (Hawn 2009; Yang et al. 2014; Dos Reis and Culotta 2015; Paul et al. 2015). Take diabetes as an example; diabetic patients not only share about events happening around them but also frequently post about their current health conditions, medication, and the outcomes of medications. For instance, they frequently post the latest values of their blood glucose, diet, and exercises using “#diabetes” and “#BGnow” hashtags on Twitter. This provides new opportunities to understand individuals’ wellness that can be used to assist them in managing their health in a scope that previously was impossible. As a first step Copyright © 2016, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. towards accomplishing this end, we propose to automatically extract wellness events from users’ published social contents. Extraction of personal wellness events (PWEs) will provide significant insights about individual’s wellness and community lifestyle behaviours. At the individual level, it can summarize the past wellness events of individuals which significantly facilitate lifestyle management through coarse and fine-grained browsing. PWE summary can be useful for downstream applications such as user health profiling, personalized lifestyle assistant, and targeted online advertising. Take diet as an example; if one diabetic person consumes a lot of carbohydrates, the system can offer diet suggestion. At the community level, accumulating the wellness information of a large set of individuals makes it feasible to analyze and understand the lifestyle patterns and wellness of social groups in a scale that was impossible with traditional methods in terms of both time and cost. Despite its value and significance, extracting PWEs from social media services has not been fully investigated due to the following challenges. First, the language used in social media is highly varied, informal, and full of slang words. Second, PWEs are relatively rare in social media posts as users tend to post their personal significant events together with lots of trivialities and other public events (Li et al. 2014). As a result, wellness events are buried among other contents produced by the users and their social connections. Identifying wellness events from a huge volume of other non-wellness events poses a big challenge. As a result, even a large annotated dataset might contain just a few examples of PWE categories. Third, the structure of wellness events exhibits a hierarchical taxonomy as shown in Table 1. Indeed, events under the same category are closely related. For instance, clinical tests are much more related to treatment, than running. These events may share some features such as entities, attributes and relations, which makes the problem arduous. How to mathematically model such relations and integrate them into a learning framework remains a challenge. In health sciences, it has been intensively studied and well-established that physical activities, diet planning and taking prescribed medications are the key therapeutic treatments of many diseases (Pastors et al. 2002; Hu 2011). Further, unhealthy lifestyle behaviours such as unhealthy dietary habits, sedentary lifestyle, and the harmful consumption of alcohol are mainly related to the risk factors of nonProceedings of the Thirtieth AAAI Conference on Artificial Intelligence (AAAI-16)