Analyzing Foreign-Language Social-Media Reaction to Televised Speeches: Lessons Learned

摘要

This paper updates our approach to analyzing social media response to speech events, such as the President's State of the Union address, presenting some new lessons learned in deploying this system. We previously described how we determine which specific lines in the speech resonate with different cohorts of the audience on social media (and how that alignment takes place, etc). In this paper, we describe an update to our approach that allows us to incorporate foreign languages into this analysis. That is, given a speech in English, such as the State of the Union, we describe how we can incorporate Tweets in foreign languages and still perform the clustering. From our development and deployment, we learned two important lessons to share: first, how to bootstrap the data collecting when supporting foreign language analysis, and second, how to order the clustering and translation operations to support a wide variety of languages with minimal cost (in terms of man hours).