id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
cord-026948-jl3lj7yh	Amini, Hessam	Towards Explainability in Using Deep Learning for the Detection of Anorexia in Social Media	2020-05-26		.txt	text/plain	3281	174	54	Results show that the weights assigned by the user-level attention strongly correlate with the amount of information that posts provide in showing if their author is at risk of anorexia or not, and hence can be used to explain the decision of the neural classifier. Previous work in NLP for clinical psychology has typically used this type of attention mechanism to create a representation of social media users: a collection of online posts from each user is fed to the model and the inter-document attention (also referred to as user-level attention) creates a representation of the user through a weighted average of the representations of their online posts, with the most informative posts are assigned higher weights. In this paper, we propose a quantitative approach, specifically focused on the user-level (inter-document) attention mechanism in a binary classification task of detection of a specific mental health issue, anorexia. In this work, we proposed a quantitative approach to validate the explainability of the user-level attention mechanism for the task of the detection of anorexia in social media users based on their online posts.	./cache/cord-026948-jl3lj7yh.txt	./txt/cord-026948-jl3lj7yh.txt