Skip to main content
Skip to footer
Contact Us
We are taking you to another website now.
May 8, 2020

Recommendation is ubiquitous today in any e-commerce, banking or e-learning portal. Each and every time we browse YouTube, Netflix or Amazon Prime, we get recommendations of movies, web series and so on. At the same time, a friend will get a different set of recommendations for the same browse navigations. Similar phenomenon happens on e-commerce portals such as Amazon or Flipkart where they recommend products.

How these systems are so intelligent that they make an assumption about your personal taste and preferences?

The answer is nothing but a machine learning enabled recommendation system. Same technique can be used in the Human Resource Management Systems (HRMS) to recommend learning courses for the employees of any organization.

When it comes to build a simple but robust course recommendation solution, Association Rule Mining (ARM) is considered to be the most suitable approach. ARM is a technique to find frequent patterns and correlations in a given dataset. Essentially it tries to find the rules that relates how or why such items occurred together. These frequent item sets determined with Apriori algorithm are used to mine association rules and generate recommendations. Only the user’s learning history data is sufficient in this approach.

Popular and Conventional Recommendation in HR Applications

Following recommendation techniques are used in the enterprise applications:

  • Demographic filtering: In this approach the most frequently enrolled courses are recommended which are often very generic.
  • Content-based filtering: This algorithm suggests courses based on similarity in content. Data such as be course type, subtype, category or tag-words can be used to determine similarity. This approach is static and requires frequent updates.
  • Association rules: A simple but popular machine learning approach often used for market basket analysis. This technique is popularly known to have recommend obvious items like jam for bread to totally unrelated items like beers and diapers. On the other hand in the more complex and conventional methods, user's interactions along with the learning history are needed to build a robust recommender using matrix factorization technique. Interactions can be captured in two ways.
  • Explicit interactions: Direct preference data from the users likes or dislikes, user feedback, ratings, etc.
  • Implicit interactions: Indirect preference data such as frequency of visits made to a course page, previous unsuccessful attempts made to enroll a course and so on.

Course Recommendation using Association Rule Mining

Right now, most of the HR management systems that many organizations use have little or no scope to capture users' feedback. Also, implicit interactions are usually very sparse for learning systems and hence are not of much use to train a machine learning model. In addition to this, the recommendation systems built using collaborative filtering using user reactions and ratings, involve costly computational power and advance technical expertise to build and sustain.

By using ARM first we can find course rules matrix for close user groups, i.e. for each Job Grade and each Organization Unit. From each matrix, using user-based collaborative filtering we can recommend personalized courses for those users who come under the same group of Job Grade and Organization Unit, based on user’s learning history.

For Example, suppose in Job Grade – “C2” and Organization Unit – “Product Engineering” it is found that HTML, CSS, JavaScript, jQuery and Angular JS have a strong correlation in learning history. By these redefined rules, if a user has done HTML, CSS and JavaScript then the recommendation would be Angular JS and jQuery for users in the same Job Grade and Organization Unit.

For scenarios where there is no historical data available, for example for freshers or employees interested in learning new technology or skill-sets, this model can find most enrolled courses for each Organization Unit along with Job Grade and recommend accordingly.

Above solution might also work for those employees who have been promoted to next level where they do not have a learning history. Here we can use purely demographic filtering to find top trending courses for each job level in the organization. In future, if some new Job Grade or Organization Unit is created then in such cases also this technique can be applied.

This approach not only provides tuned learning recommendations for specific job grade within a business unit but also helps in faster gasping of competencies and knowledge on role change.

Advantages of ARM-based Course Recommendation

  • Association Rule mining technique will work without explicit user reactions about courses
  • Low cost solution can be built using open source model building libraries and statistical techniques
  • Scalable for even large volume of user and courses 


Nowadays we are so busy with our daily work that we rarely get time to upgrade our skills. Therefore, when it comes to career growth and skill enhancement then course recommendation platform will be of great help.

Pratik Kayal is a developer and team member of the Innovation and Product Engineering - Analytics (IPE - Analytics) group within TCS Platform Solutions. He possesses strong skills in Python, Data Science, Machine Learning and Deep Learning. He is responsible for multiple product engineering and innovations, working as a business intelligence and machine learning engineer. He has been contributing to the development of embedded analytics and also in implementation and hosting of Web Services for predictive analytics use cases in the home grown HRMS and Procurement platforms of TCS Platform Solutions. He holds a bachelor’s degree in Electrical Engineering from Academy of Technology under Maulana Abul Kalam Azad University of Technology (Formerly known as West Bengal University of Technology).


Thank you for downloading

Your opinion counts! Let us know what you think by choosing one option below.