InMacs: Interactive modeling and comparison of sentiments from sequence data

Machine learning (ML) has changed various problem domains by offering insightful solutions. For example, urban planners (usually non-experts in ML) model sequence data such as text using AutoML systems (e.g., OrangeML, Google Cloud ML, etc.). Specifically, these users mine unstructured text data using Twitter API to compare peoples’ sentiment/opinion on urban spaces. However, the current AutoML tools restrict the active participation of end-users in model construction/adjustment. To resolve this problem, we designed an effective technique that combines an interactive visual interface with an AutoML model solver incorporating users’ domain knowl- edge as feedback that adjusts the underlying models’ behavior. In this paper, we present InMacs, an innovative visual analytics (VA) system that allows urban planners to interactively construct senti- mentclassifiers andvisualizethe outputofthesemodelstocompare peoples’ sentiment across multiple geolocations. Through a case study we discuss our on-going work with urban planners that in- cludes design, build, and validation of our prototype. Furthermore, we discuss the effectiveness and the generalizability of our inter- active technique on other domains by presenting a case study that compares business reviews from the publicly available Yelp dataset.