A new model helps employers find the best website to post a job
Whether you are looking for a new job or looking for a new employee, you are unlikely to go about it without using the Internet. These days, there are hundreds of different websites for posting job opportunities, making it difficult for recruiters to know where to advertise. Now, a model presented in the journal Big Data Researchcan help companies decide which online job board to use.
The idea came from a recruitment technology company called Multiposting based in France and a social media recruiting company called Work4 based in the US. They believed that if they could help recruiters work out which websites to post new jobs to, by predicting where applicants would search, they could make the process much more efficient.
So to help build the algorithms that go into the model, they approached Professor Sidahmed Benabderrahmane, from the University of Paris 8, France. “They were needing an expert in time series prediction algorithms,” said Benabderrahmane. These algorithms take data plotted over time and predict what will happen in the future.
The new system is more advanced than any previously used because it takes into account the behaviour of the individual job applicants. The algorithm targets the kind of people the job opportunity is hoping to attract. Based on the history of their clicks, it uses a time series algorithm to predict where they will click in the future.
Alongside the time series algorithm, the model uses a semantic classification method – which analyses websites using textual analysis and vocabulary dissemination. “The combination of these two procedures helped to enhance the results of the recommendation,” said Benabderrahmane. The model also predicts how many applicants each advert will receive.
Now, the two start-ups have begun using the system as part of a service called Sonar. Benabderrahmane hopes to continue to improve the system as its used. “We propose to consider the information that is represented as time series to improve the efficiency of the recommender system.”
Read Article Free Online until 4th July 2018
Sidahmed Benabderrahmane, Nedra Mellouli, Myriam Lamolle, Patrick Paroubek
Big Data Research, Volume 7, March 2017, Pages 16–30