Scroll down

AI algorithms applied to the learning field

Pascal Lim |

Artificial intelligence has taken an important position in all industries, enabling tasks to be performed on a larger scale, faster and more accurately. This gives the players concerned the possibility to perform tasks with higher added value and centered on the human being. The applications of AI in learning are multiple and continue to grow. In what follows, we look at several algorithms used in this field. 

Clustering to better understand learners and their needs 

Clustering is considered the entry point for many AI algorithms in learning. It consists in separating elements into homogeneous groups according to certain common characteristics. In the past, these characteristics were predefined, for example from demographic or geographic data, but thanks to machine learning algorithms, this task becomes fully automated. One of the best-known clustering algorithms is K-Means1, which is based on unsupervised learning, i.e. it defines clusters without the need for labeled data. 

This type of algorithm is often used in pedagogical contexts: it allows, for example, to group learners together in order to bring out groups of people with difficulties or, on the other hand, others who are more advanced in a skill to be acquired. This type of information will prove useful for the trainer who, depending on the results, will be able to offer more personalized support to each learner. Clustering can also be used to predict future behaviors such as dropping out, allowing the trainer to be “one step further” in his particular follow-up of the learners. 

Recommendation systems to guide learners

A recommendation system is an engine that produces individualized recommendations or has the effect of guiding the user in a personalized way to interesting or useful content among many possible options. These tools are already an essential part of our daily lives: for instance, when we shop online, a system recommends products that might be suitable for us based on those already saved in our shopping cart. Also, when we want to view content on Netflix, the platform offers us a personalized list of movies or series to watch according to our preferences. 

There are also recommendation systems for learning purposes, however, the goal of these systems is not to encourage users to buy a product but to help them find appropriate resources and learning activities to better achieve their skill development goals. The interest of implementing a recommendation system on a learning platform (such as LMS, LXP, etc.) is to guide learners in their learning path while avoiding overloading them with unnecessary information. These recommendations are adapted to their needs and characteristics such as their level of mastery, prior knowledge, cognitive abilities, skills; to help them progress.

The most popular recommendation algorithm is collaborative filtering, to name but one. This algorithm recommends to users learning content that people with a similar profile, but a more advanced level of proficiency have consulted in the past. It predicts the pedagogical impact of a content on a learner based on the interactions of other people with similar learning profiles. There are two types of collaborative filtering: the user-user, which recommends to learners’ content that other learners with similar profiles have consulted, and the item-item, which recommends to learners’ pedagogical content that is similar to what they have consulted. 

Automatic natural language processing for pedagogical content management

Natural language processing (TALN) is one of the fastest growing fields. This discipline of AI allows us to process language-related data that we usually use in text format with machine translation tools, or in voice format with voice assistants (Siri, Alexa or Google Home, to name just a few). The use of this data by the computer requires a projection of the words into a semantic representation: this step is called word embedding

The learning sector is full of resources for the use of TALN with various learning contents (courses, questions, books, videos, etc.). There are many application cases that can benefit the various players in the learning field, from the pedagogical content designer through the learner to the trainer. Some of the possible applications include: 

  • Classification of similar content under the same theme using supervised learning algorithms  
  • Automatic generation of questions from several text corpora  
  • The creation of summaries and content summaries  
  • Help with correction by highlighting, for example, the passages related to a particular keyword 

In conclusion, AI embodies a real opportunity in the learning field: the acquisition of skills by learners is reinforced by the personalization of learning2, a method now possible thanks to the modeling of learners by AI (in contexts where the adaptation of pedagogy by humans is impossible). In addition, large-scale data processing allows training actors (learners, teachers, trainers, managers, etc.) to be freed from restrictive or time-consuming tasks to focus on actions with higher added value. 

1 Source: Principles and operation of the “K-means” algorithm 

2 Source: B. S. Bloom, “The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring”, Educational Researcher, pp. 4-16, Jun. – Jul. 1984