Academos
How to use Text Mining to understand your organization
Using Text Mining techniques to automatically segment thousands of mentors into homogeneous categories and improve recommendations for youth.
Client information
The business problem
Founded in 1999, Academos is a non-profit organization that connects young people aged 14 to 30 with the reality of the working world through a virtual mentorship app that allows them to chat freely with thousands of professionals passionate about their careers.
To find a mentor, young people use a search bar and enter keywords. The algorithm then suggests profiles containing those keywords. If a young person searches for 'medicine', doctor profiles will likely interest them, but perhaps pharmacists, psychologists, and other health professionals would interest them even more!
With thousands of mentors registered on the app, manually segmenting them was impossible. A way to teach the computer which mentors are similar was needed.
Key challenges
The solution
Text Mining techniques were used to group similar profiles by leveraging the hundreds of words contained in the descriptions written by mentors upon registration: work description, typical day, job highlights, and least favorite aspects of their job.
Our hypothesis: mentors using similar words in these descriptions likely hold similar positions. After several hours of data cleaning, transformations, and algorithm refinement, we achieved a convincing result.
Two simpler approaches (grouping by degrees and by job title keywords) were first evaluated and ruled out, confirming the need for a more sophisticated Text Mining approach.
Text Mining analysis
Knowledge extraction from mentor-written texts to identify similarities between profiles.
Automatic segmentation
Algorithmic grouping of thousands of mentors into homogeneous categories based on the vocabulary used in their descriptions.
Fine-grained job distinction
The algorithm can distinguish nuances, such as physical construction trades vs. those related to design and planning.
Improved recommendations
Results enable recommending profiles that are similar in interests but different in occupations.
The results
The algorithm successfully segments mentors into uniform groups. Some groups are fairly obvious while others are more surprising but logical: texture artists, film producers, and 3D animators grouped under 'Visual Arts'; scrum masters, solution architects, and developers under 'Software'; linguists, technical writers, and translators under 'Languages'; physiotherapists, speech therapists, and zootherapy workers under 'Helping Relationships'.
Thousands of mentors were grouped into fewer than a hundred homogeneous categories. Academos can now better understand the themes covered by its mentors, focus recruitment efforts on underrepresented categories, and improve mentor suggestions for youth.
Technologies used
Ready to achieve similar results?
Let's discuss how we can transform your business.
Talk to our teamRelated case studies
How to double your revenue through technology
Centralized dashboard and customer traffic forecasting
