TrendQuery: a system for interactive exploration of trends

Abstract

The surfacing of trends from data collections such as user-generated content streams and news articles is a popular and important data analysis activity, used in applications such as business intelligence, quantitative stock trading and, social media exploration. Unlike traditional content analysis, trend analysis includes an additional vital time dimension: a trend can be defined as a temporal pattern over a group of semantically related items. The unsupervised discovery of trends is often not sufficient, either due to inadequacies in the trend analysis algorithm, or because the data collection itself does not possess all of the information to identify the trend. Thus, it is necessary for an expert human-in-the-loop to be involved in the process of trend analysis. To this end, we introduce TrendQuery, a system designed towards iterative and interactive surfacing of trends. Our system provides a set of trends to the expert, and enumerates iterative operations to curate the result. This process continues until the expert is satisfied with the surfaced trends. Since the space of possible tweaks to the result can be extremely large, the system continually provides feedback and guidance to the expert to prioritize possible operations. Our system allows interactive curation of trends providing better insights than a purely unsupervised approach.