Recommending Interesting Writing using a Controllable, Explanation-Aware Visual Interface

Rohan Bansal, Jaan Altosaar

Abstract:

We build a visual interface for recommending articles to editors at The Browser, a curation service for interesting writing.
From a large list of candidates, editors decide which articles
are selected and shared with subscribers. To aid the editors
in this decision-making task, we build a visual interface for
a recommendation model, RANKFROMSETS (RFS) [2], that
classifies articles based on their words. Control of the recommendation model is built into the visual interface. For example,
an editor can use a topic slider to receive a new list of recommendations according to topical words in articles. These topic
sliders might be used to increase or decrease the ranking of
articles with words related to crime, business, or technology.
The visual interface is also designed to be explanation-aware:
words that contribute positively or negatively to an article’s
ranking are displayed. For the backend of the visual interface,
RFS is trained on historical data. In an offline empirical study,
we find that RFS outperforms BERT [4], a competitive classification model, in terms of recall. Further, we measure RFS to
be 10 times faster to train and to return predictions 2000 times
faster than BERT. This speed is a beneficial property for the
visual interface, and we demonstrate that RFS can be deployed
on the free tier of AWS Lambda using a short python script
and numpy dependency. For reproducibility, transparency, and
trust of the visual interface, we open source and release a
public demonstration,1 data collection, training and deployment
scripts, and model parameters