Linkedin Learning is website providing video courses of professional skills of various business areas. After logging in, the recommender system will generate a few lists of suggested course videos in the home page according to your learning history and demographic information.
The following link is a snapshot of the home page:
The recommended lists can be divided into 4 sections:
Trending and popular course: The logic behind the screen is simple. It can be done by sorting the number of recent views and the sorted list can be updated frequently online by using bubble sort.
Top skills based on the company or industry that the user is currently in: I believe the lists are generated using a Collaborative Topic Model. The model is run offline to identify the subjects of the course videos. Based on the user’s industry area, the system then provides videos of related subjects.
Courses based on the user’s learning history: The list is provided with reasons such as “Because you have xxx skill on your LinkedIn profile” and “Because you watched ‘x-y-z’”. It seems that the system is using Search-Based Methods to generate the list, using the skills in the user’s profile and previous watched videos as search items.
Courses that are watched by similar learners: The list is provided with the reason “Learn xxx, a top skill for similar learners”. Apparently, this is a Cluster Model. The model identifies similar learners and recommends you with courses that they have learned.
From the company’s perspective:
1) There are two groups of users, one is the learners and the other one is the industry experts who create the course materials.
2) The key goal is to maintain current subscribers’ interest in the website by recommending courses that the learners may like to study.
3) To accomplish this goal, I suggest the website to provide periodic update on the technologies, skills and news about the subjects that a learner has studied.
From the instructors’ perspective:
1) Their users are the course learners.
2) The key goal is to attract more new learners. However, the recommender system is not so friendly to instructors whose courses are not as popular as the others.
3) In order to present less popular courses to new learners, an idea is to reserve a few slots in the recommendation list and loop through the courses dynamically.
From the learners’ perspective:
1) The users are themselves
2) The key goal is to empower themselves with new professional skills
3) Though the recommender system suggests new course videos to the learners, some videos may cover a large portion of identical materials. An algorithm similar to Amazon’s Item-to-Item Collaborative Filtering can be developed to run offline to compute the similarity of videos of the same subject. The difference is that we pick the most non-similar items within the same subject in this case. The recommender system would be able to select and present new courses to the learners that would benefit them the most.