Pull-Factor Regular Expressions in Job Advertisements

Introduction
Every job advertisement contains features intended to make the position more attractive to candidates. Examples include a company car, pension scheme, or flexible working hours. These are called pull factors.

Research has shown that job seekers in different sectors value different pull factors. Textmetrics developed a tool for recruiters that supports the writing of inclusive and diverse job advertisements. One of its features is a checklist that indicates whether important pull factors are present in the text.

Task
In the Textmetrics application, pull factors were previously recognized using lists of words or phrases. Our task was to create regular expressions so that a much wider variety of pull-factor mentions could be identified. Regular expressions are more flexible and can capture different phrasings of the same concept. For example, a company car may appear in a job ad as:

• We offer the possibility of using a lease car
• You get a company car
• A car is available

At the same time, it was important to minimize false positives. Simply matching every sentence containing the word “car” would create too many errors. Instead, expressions had to be designed carefully so that patterns such as “a car from us” or “a car from our company” could be included without becoming too broad.

Method
Our method consisted of two main steps:

• Creating extensive regular expressions using the original pull-factor lists and 500 job advertisements annotated by colleagues.
• Balancing correct detections and false positives by iteratively refining the expressions.

Evaluation
We evaluated the regular expressions by running them over 10,000 job advertisements. For each pull factor, we selected a representative subset for manual review using word embeddings. This allowed us to refine the expressions and strike a balance between keeping correct detections and removing false positives as much as possible. For end-users of the Textmetrics app, it is frustrating if a genuine pull factor is present but does not receive a check mark, but equally undesirable if a check mark appears when the factor is not actually mentioned.

Outcome
After completing the project, Textmetrics had a much more comprehensive pull-factor detection system. It went well beyond a simple list of keywords and was more flexible in recognizing varied phrasing, while maintaining a practical balance between recall and precision.