Supporting Job Mediator and Job Seeker through an Actionable Dashboard
(this work was realised in 2017)
This paper is the result of a collaboration with a regional government job mediation service. The mediation service created a preliminary dashboard for job mediators, which would show details on the output of a recommendation system: showing what jobs would be appropriate for the job seeker, what actions would be help progress in the job seeking, but also what parameters would create a hurdle to reach a certain job goal. I was tasked with exploring how job mediators, but also job seekers, would experience the dashboard, the results of the recommendation system, and how this translation of AI output to the actual user could be improved.
Top: the dashboard initialised with the recommended information. The mediator analyses this information and attempts to find a story that is actionable, removes errors that the recommendation system has made, and eliminates demotivating factors (e.g. age is removed from the dashboard as the job seeker has no control over that. In other cases, an unrealistic goal can be left on the dashboard to make sure the job seeker does not get their hopes up incorrectly).
The result is shown above. Through extensive user studies (a customer journey, 23 hours of observations, and questionnaires) and previous research, the dashboard was redesigned as a collaborative visualisation: the mediator would control the message by being the user interacting with the dashboard, but the dashboard would also be used to pass the message visually to the user.
The mediator would be given control to the output of the message: the story parts are created by the recommendation system, but the mediator decides how to turn this into a story suitable and customised for the person sitting in front of them.
For a more thorough explanation of the dashboard, the user research, and the results, head over to our article published in the Proceedings of the 24th IUI conference on Intelligent User Interfaces, or get in touch!
Small bonus: we compared forest plot (used in the original dashboard I was tasked to redesign), a simple circle visualisation, and bar charts in order to understand what would be best suitable to indicate positive and negative impact on job success predictions. While this sounds trivial, the dashboard needs to be used by experts (job mediators), but also be understandable for a wide range of job seekers (e.g. age and education. Here's an overview of the representations and the user perception.
Three ways of presenting the parameter data, with age as example. Forest plot (F) indicates the job seeker’s age has a positive impact. Circles (C) segments the data across the age value, indicating positive and negative impact through colours: the job seeker’s age is within the positive 26-30 range. Barcharts (B) add the size of impact to each segment through height: the 26-30 age range has more positive impact than the 18-25 range.
Boxplots of the five-Likert scale questionnaires (1-Completely disagree, 5-CompletelyAgree) comparing the different parameter visualisations regarding clarity, clarity with additional explanation, clarity towards the job seeker, clarity for the job seeker with additional explanation, added value, positive and negative motivational impact. Top: responses of mediators with job seekers with higher education background. Middle: responses of mediators with job seekers with lower education background. Bottom: the combined results. The dark line indicates the median, “+” the mean, and circles the outliers.
