نبذة مختصرة : Effective utilization of training data is a critical component for the success of any artificial intelligence algorithm, including natural language processing (NLP) tasks. One particular task of interest is related to detecting or ranking humor in texts, as exemplified by the Humicroedit data set used for the SemEval 2020 task of assessing humor in micro-edited news headlines. Rather than focusing on text classification or prediction, in this study, we focus on gaining a deeper understanding and utilization of the data through the use of information visualization techniques facilitated by the established NLP methods such as sentiment analysis and topic modeling. We describe the design of an interactive visualization tool prototype that relies on multiple coordinated views to allow the user explore and analyze the relationships between the annotated humor scores, sentiments, and topics. Evaluation of the proposed approach involves a case study with the Humicroedit data set as well as domain expert reviews with four participants. The experts deemed the prototype useful for its purpose and saw potential in exploring similar data sets with it, as well as further potential applications in their line of work. Our study thus contributes to the body of work on visual text analytics for supporting computational humor analysis as well as annotated text data analysis in general. ; This work was partially supported through (1) the ELLIIT environment for strategic research in Sweden and (2) the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation.
Processing Request
No Comments.