CrowdE: Filtering Tweets for Direct Customer Engagements

Abstract

Many consumer brands have customer relationship agents that directly engage opinionated consumers on social streams, such as Twitter. To help agents find opinionated consumers, social stream monitoring tools provide keyword-based filters, which are often too coarse-grained to be effective. In this work, we introduce CrowdE, a Twitter-based filtering system that helps agents find opinionated customers through brand-specific intelligent filters. To minimize per-brand effort in creating these brand-specific filters, the system used a common crowd-enabled process that creates the filters through machine learning over crowd-labeled tweets. We validated the quality of the crowd labels and the performance of the filter algorithms built from the labels. A user evaluation further showed that CrowdE's intelligent filters improved task performance and were generally preferred by users in comparison to keyword-based filters in current social stream monitoring tools.