What Makes Videos Accessible to Blind and Visually Impaired Users?
To appear at CHI '21 [Manuscript]
Videos on sites like YouTube almost universally lack audio descriptions. Our formative studies with BVI users revealed that participants used a time-consuming trial-and-error approach. BVI users also reported video accessibility heuristics they considered to describe accessible and inaccessible videos. We instantiate seven identified heuristics as automated video accessibility metrics to indicate video accessibility.
Making Memes Accessible
We present two methods for making memes accessible semi-automatically through (1) the generation of rich alternative text descriptions and (2) the creation of audio macro memes. Meme authors create alternative text templates or audio meme templates, and insert placeholders instead of the meme text. When a meme with the same image is encountered again, it is automatically recognized from a database of meme templates.
Identifying Terms and Conditions Important to Consumers using Crowdsourcing
Currently under review for CSCW '21 [Manuscript]
We present a workflow consisting of pairwise comparisons, agreement validation, and Bradley-Terry rank modeling, to effectively establish rankings of T&C statements from non-expert crowd workers on this open definition, and further analyzed consumers’ preferences. We foresee using our workflow and model to efficiently highlight important T&Cs on websites at a large scale, improving consumers’ awareness.
Analyzing Biases at Multiple Stages of an Algorithmic Decision Support System
We provide a flow model and study how different forms of biases can arise and propagate throughout the ORES pipeline. Specifically, we (1) identify potential sources of biases across four phases of the pipeline, (2) show evidence of biases through quantitative and qualitative analysis, and (3) propose interventions to mitigate these biases and test their feasibility within the real-world context.
Method, Device and Computer Product for Predicting Disk Failure
We propose a machine learning solution for disk failure prediction on ASUP database. Our method can directly deal with ASUP database, with an automatic process of noise reduction, feature selection and engineering. The machine learning model can learn the processed data and output a strong disk failure predictor. Our prototype system can reach a failure detection rate of 90%, with a false alarm rate lower than 0.5%.