Using HyperLogLog to Prevent Data Retention in Social Media Streaming Data Analytics

Abstract

Social media data is widely used to gain insights about social incidents, whether on a local or global scale. Within the process of analyzing and evaluating the data, it is common practice to download and store it locally. Considerations about privacy protection of social media users are often neglected thereby. However, protecting privacy when dealing with personal data is demanded by laws and ethics. In this paper we introduce a method to store social media data using the cardinality estimator HyperLogLog. Based on an exemplary disaster management scenario, we show that social media data can be analyzed by counting occurrences of posts, without becoming in possession of the actual raw data. For social media data analyses like these, that are based on counting occurrences, cardinality estimation suffices the task. Thus, the risk of abuse, loss or public exposure of the data can be mitigated and privacy of social media users can be preserved. The ability to do unions and intersections on multiple data sets further encourages the use of this technology. We provide a proof-of-concept implementation for our introduced method, using data provided by the Twitter API.

On the Privacy of the Count-Min Sketch: Extracting the Top-K Elements

“…HyperLogLog (HLL) has been proposed as a base sketch to provide privacy too (e.g., [20]) that can be applied to, for instance, social media data processing in disaster management [21]. In [22], the authors have concluded that cardinality estimators, which are used for aggregation (e.g., HLL sketches that are merged by data analysis tools) and consequently require accuracy, do not preserve privacy.…”