19/01/2017 | Written by: Think Blog redactie (0cB)
Categorized: Generic | Watson
Share this post:
Watson tackles youth unemployment.
How long and why do young people stay unemployed after entering the labor market? And what actions are the most effective to help them find a job quicker? An IBM Impact Grant helps the Flemish Employment Agency (VDAB) answer these important questions. In just ten weeks, IBM consultants delivered a predictive model prototype using Watson cognitive technologies for natural language processing and self-learning. The tool has the potential to offer vital insights into youth unemployment and support VDAB career counselors in personally assisting each individual young job seeker.
Every year, 70,000 young jobseekers enter the Flemish labor market. Finding a job is not always easy though. VDAB is the public agency that matches demand and supply by guiding jobseekers and by providing training that leads to jobs. The agency supports 233,349 unemployed people, of which 21% are under the age of 25. Youth unemployment is not only an issue in Flanders, it is a European problem. The European Union has launched a program to tackle this. Paul Danneels, Chief Information Officer of VDAB, explains: “One of the targets of the program is to make new graduates the right offer within three months. This can be a matching vacancy, retraining or counseling. The longer young people stay unemployed, the greater the chance they will have a problem in finding the right job later in their career. This goes for graduates at all levels.”
In order to achieve the targets of the European program, VDAB needed better insights into Flemish youth unemployment and the effectiveness of its services. That is why the employment agency applied for the IBM Analytics Assessment and Insights Impact Grant. However, this was not the only reason. Danneels: “At VDAB, we value information very highly, both from an IT and a business perspective. We want to use it to improve our existing services and offer new, innovative services. The grant offered us the opportunity to analyze an enormous amount of historic data on youth unemployment, both structured and unstructured information, and to help us learn from all this information to optimize our services.”
“Our recommendations have an impressive level of certainty.”
Power of big data
During a 10-week assignment, a team of analytics consultants from IBM combined different data sets over a 4-year period of all job seekers registered with VDAB between the ages of 18 and 25. The objective was to create a predictive model prototype that offers a prognosis for the amount of time young people will remain unemployed. First of all, such a model enables the VDAB to deploy its counselors and tools as efficiently as possible so that the largest possible group of job seekers can find jobs quicker. Moreover, the agency can also better determine what factors will more quickly lead to employment for each individual job seeker.
Danneels: “Based on the historic data, we were able to get very important insights into all the services we offered in the past, how we handled things and into the final results. After that, the model was continuously refined by additional data sets. It allowed us to split young job seekers into three groups and predict who would get a job within 30 days, within 180 days or in more than 180 days. A remarkable finding was that thirty percent of the jobseekers would get a job within 30 days. That means you can focus on the rest.” The prediction has a certainty of eighty percent. “This is very high, due to the fact that the total population was used for the analysis, and not just a sample. It’s a clear example of the power of big data,” says Danneels.
The model also allows VDAB to predict the particular services that are best suited for the career counselors to personally assist each jobseeker. They can make fictitious changes in a number of variables, such as education level, language skills, desired jobs, work experience (e.g. internships), or personality traits. In many cases, a change in one of these variables leads to a significantly better chance of finding a job. The model that IBM developed is also self-learning. Based on examples, the system learns to discover links between previously unknown factors. Then it can make a number of suggestions to job seekers. Because the system makes these suggestions based on success stories from the past, career counselors are able to predict what actions will be successful for each job seeker. Danneels: “The advantage of working this way is not only that our career counselors can give personalized recommendations that have an impressive level of certainty. It is equally important that they all base their decisions on the same information, their approach does not depend on individual expertise and experience.”
“We didn’t expect to get such clear answers to our questions.”
Thanks to the Impact Grant, VDAB now has a powerful prototype model for predicting youth unemployment. As a next step, the tool will be refined and tested in actual practice. “Our career counselors will use the model to work with their clients over a period of three months. After that it will be refined and applied to all job seekers of all ages. In a later stage, we may also offer the tool directly to our clients,” states Danneels. His expectations have been exceeded. “We had to draw and prepare a huge amount of data from difference sources over a 10-year time period. The fact that we could also analyze unstructured information such as counselor notes made it possible to delve much deeper into our data than before. We didn’t expect to get such results and clear answers to our questions that quickly.”