Prof. Vasudeva Varma and his team came second at the CL-Aff Shared Task for their poster on Ingredients for Happiness: Modeling constructs via semi-supervised content driven inductive transfer at the AAAI-19 Workshop on Affective Content Analysis (AFFCON 2019) – Modeling Affect-in-Action at Hilton Hawaiian Village, Honolulu, Hawaii, USA from 27 January to 1 February. The authors of this paper are Bakhtiyar Syed, Vijayasaradhi Indurthi, Kulin Shah, Manish Gupta and Vasudeva Varma.
In the quest to understand user expression, the team proposed a task focusing on one facet of human affect – happiness. They contributed a new labeled corpus of happy moments and posed two novel challenges to spur the development of supervised and semi-supervised approaches to model human affect.
The purpose of the CL-Aff Shared Task is to challenge the current understanding of emotion analysis through a task that models the experiential, contextual and agentic attributes of happy moments. It has long been known that human affect is context-driven, and that labeled datasets should account for these factors in generating predictive models of affect. The Shared Task is organized in collaboration with researchers at Megagon Labs and builds upon the HappyDB dataset, comprising human accounts of ‘happy moments’. The Shared Task comprises two sub-tasks for analyzing happiness and wellbeing in written language, on a corpus of 100,000 descriptions of happy moments.
The team was given an account of a happy moment, marked with individual’s demographics, recollection time and relevant labels.
Task 1 consisted of deciphering the ingredients for happiness. In this semi-supervised learning task the team had to predict agency and social labels for happy moments in the test set, based on a small labeled and large unlabeled training data.
More details of the conference at https://sites.google.com/view/affcon2019/cl-aff-shared-task