TRENDS IN EDUCATION
MACHINE LEARNING AND DATA ANALYSIS
MACHINE LEARNING AND DATA ANALYSIS
Coursera Partners Conference 2019 main outcomes
About the Coursera Global Skills Index and the practical use of machine learning in education and human potential development
The Coursera Partners Conference is one of the most anticipated international events of the year in the educational world. Coursera, which sets the trends in online education, this time, also surprised all the participants by presenting something that everyone has been talking about and dreaming about for years, but has never seen before – the Global Skills Index, a model of skills and methods for determining levels of competence, based on data collected by the platform about a person's completed trainings.
It seems that the dream of the ELDF team will soon come true. The dream is to give people the opportunity to design their development trajectory and according to it to get qualitative and necessary, in the moment and in the future, knowledge to achieve their goals! We strive to guide people on the "path" of knowledge and skills for our students, but the development of the skill model will allow the formation of long–term programs for the development of the individual.
Taras Pustovoy
CEO & Founder at ELDF
“
The Coursera conference that was held last week in London was a pleasant surprise.
After seven years of data collection and three years of analysis, Coursera has presented the Global Skills Index. And there are two important consequences of this news.
1) Coursera, after a significant expansion of its Data Science team in the last 2 years, has finally started processing its data. As their first result, they gave out something that everyone was talking about and waiting for – their model of skills + a method to recalculate the possession level of these skills by all users on their platform (and there are 40 million of them now).
The basic principles of the model are very simple: it's based on an already created taxonomy (in this case, the Wikipedia rubricator), and some of the areas were smoothed with the help of experts. In this case, they've got a five-level model.
And then it gets more interesting.
After seven years of data collection and three years of analysis, Coursera has presented the Global Skills Index. And there are two important consequences of this news.
1) Coursera, after a significant expansion of its Data Science team in the last 2 years, has finally started processing its data. As their first result, they gave out something that everyone was talking about and waiting for – their model of skills + a method to recalculate the possession level of these skills by all users on their platform (and there are 40 million of them now).
The basic principles of the model are very simple: it's based on an already created taxonomy (in this case, the Wikipedia rubricator), and some of the areas were smoothed with the help of experts. In this case, they've got a five-level model.
And then it gets more interesting.
On the one hand, it's necessary to mark content and tasks according to the elements of this taxonomy. They did it in two steps: first, they analyzed assignment texts, automatically translating them into English if the content was not in English. And then they used crowdsourcing: asking teachers about what they were teaching and students about what they (in their opinion) were learning for each content element.
On the other hand, by taking descriptions/titles of professions and career goals from texts from job sites, they've marked up skills by career goal. And now, when the student chooses their goal and development directions, they learn the skills they need and get personalized content recommendations. What's more, these recommendations begin to follow them everywhere (forming the search output, to course catalog and homepage carousel, and so on)
On the other hand, by taking descriptions/titles of professions and career goals from texts from job sites, they've marked up skills by career goal. And now, when the student chooses their goal and development directions, they learn the skills they need and get personalized content recommendations. What's more, these recommendations begin to follow them everywhere (forming the search output, to course catalog and homepage carousel, and so on)
And at the final stage, the task was to calculate the level of skill development for each student at each point in time.
To do this, they searched for a model with several criteria: it must consider that skills develop over time, be explainable, and be reasonable in terms of computational resources for regular values recalculation. After a series of trials and errors they've come up with the Elo rating system.
If I explain it very simply (specialists forgive me), it's a competition where students fight against tasks. If a student solves a task that has a mathematical expectation of the skill difficulty level higher than the mathematical expectation of his personal skill level, then his level of development of that skill increases. And so for all tasks and for all users on the platform. The picture shows that in case of this model the results are quite stable regarding the number of solution attempts. The number of attempts to complete tasks for popular skills is around tens of millions.
It's important to understand that the rating value itself is relative, and one and the same skill development level may have different absolute values in different recalculations. For those who are especially curious, here is a photo with the formula
To do this, they searched for a model with several criteria: it must consider that skills develop over time, be explainable, and be reasonable in terms of computational resources for regular values recalculation. After a series of trials and errors they've come up with the Elo rating system.
If I explain it very simply (specialists forgive me), it's a competition where students fight against tasks. If a student solves a task that has a mathematical expectation of the skill difficulty level higher than the mathematical expectation of his personal skill level, then his level of development of that skill increases. And so for all tasks and for all users on the platform. The picture shows that in case of this model the results are quite stable regarding the number of solution attempts. The number of attempts to complete tasks for popular skills is around tens of millions.
It's important to understand that the rating value itself is relative, and one and the same skill development level may have different absolute values in different recalculations. For those who are especially curious, here is a photo with the formula
2) And this calculated and constantly updated data provides room for a variety of interesting things.
At the macro level, for example, we can compare countries in terms of the development of skill levels, show the dependence of the level of prosperity (GDP) in the country on the skill development, talk about which skills are critical in which areas, which skills are most popular now and which are becoming a trend.
Equally interesting things we can do at the micro level. From finding students with the best skills in specific areas to finding hidden talents and offering them areas where they could be most successful.
At the macro level, for example, we can compare countries in terms of the development of skill levels, show the dependence of the level of prosperity (GDP) in the country on the skill development, talk about which skills are critical in which areas, which skills are most popular now and which are becoming a trend.
Equally interesting things we can do at the micro level. From finding students with the best skills in specific areas to finding hidden talents and offering them areas where they could be most successful.
