What is a recommender system?

A recommender system uses artificial intelligence (AI), machine learning, and big data to suggest items a user might be interested in. These systems are commonly used by e-commerce, media streaming, and social media sites to suggest or recommend other products and services to people.

Recommender systems operate using sophisticated machine learning algorithms that can harness data to understand – and predict – people’s preferences based on their:

• search history
• demographic details, such as location, gender, or age.
• purchase history or other online interactions and metrics, such as clicks, likes, dislikes, and connections.

Types of recommender systems

Collaborative filtering

Collaborative filtering systems utilise past user behaviour to predict what the user might like in the future. These filtering systems are more commonly used because they are the easiest to implement. 

They work by finding similar users who have comparable interests and then recommend other items the other users have also liked, based on information such as user ratings, and so on. Essentially, the algorithm identifies User A’s past preferences, finds other users who have shown similar user preferences, and then makes product recommendations to User A based on the preferences shown by User(s) B.

Content-based filtering

Content-based filtering systems are more complicated than collaborative filter systems, and are more difficult to implement because they require the system to understand the user’s interests – but they are also typically more accurate than collaborative filtering.

They work by using a person’s interests to predict what they might like. These interests are determined using additional information about the user, such as their demographics, as well as additional information about their past choices. For example, if suggesting new songs on a music platform, a content-based filtering system would consider the user’s preferred genres, artists, song length, and so on.

Content-based systems are also less likely to experience what’s known as the cold start problem, which occurs when a system does not have enough data or information to draw from. This data sparsity means the system cannot reasonably make suggestions for users. Instead, content-based filtering means even new users or new items will have enough associated data to enable content recommendations.

Hybrid filtering

Hybrid recommender filtering methods are a combination of collaborative and content-based filtering, so users will receive recommendations based on their past preferences and historical data, their interests and the interests of users with similar tastes, and other user-specific information.

What are the benefits of recommender systems?

Recommender systems are a helpful tool for people who want to find related content, whether it’s similar products and services, new music and films, or other accounts and articles. The level of personalisation offered by recommendation systems also supports content optimisation, creating a better user experience for people.

This kind of tailored content also helps businesses increase their online user engagement – and their sales.

Understanding recommendation algorithms

Recommendation engines are typically written in codes like Python and C++. They rely on various machine learning models, algorithms, data mining techniques, and other technologies in order to function. These include:

Matrix factorisation

Matrix factorisation algorithms are typically used for collaborative filtering. They break down the data that sits within what’s known as the user-item interaction matrix to identify relationships between users and items.

Deep learning

Deep learning is a subset of machine learning, and its algorithms often support sophisticated recommendation models. It enables recommenders to delve into multiple layers of datasets to extract useful information, connections, and relationships in order to make more accurate, helpful recommendations for users.

K-nearest neighbours 

According to IBM, the K-nearest neighbours algorithm uses proximity to make predictions about the grouping of an individual data point. It’s typically used as a classification algorithm that works off the assumption that similar points can be found near one another – which is ideal for recommender systems – but it can also be used for regression problems.

Neural networks

Neural network algorithms mimic the connections in the human brain to recognise relationships between data points.

There are also autoencoders, which are a type of neural network that can work with unlabelled data, and transformers, which are a neural network model that can process sequential input data and learn context. They are often used for language translation applications.

Natural language processing (NLP)

Natural language processing enables computers to understand human language in its written and spoken forms.

Are recommender systems supervised or unsupervised?

Recommender systems can use both supervised and unsupervised learning. 

What is the difference between supervised and unsupervised learning?

Supervised learning uses labelled datasets that effectively train algorithms to classify data or predict outcomes. The training dataset will include typical inputs alongside correct outputs, which teach the model how to operate as new inputs are received.

Unsupervised learning, on the other hand, can run without the need for labelled training datasets provided by data scientists. Instead, its algorithms analyse and group unlabelled datasets to discover patterns and are typically used for one of three purposes:

1. Clustering – groups unlabeled data based on similarities or differences.
2. Association – finds connections between variables in datasets.
3. Dimensionality reduction – reduces the number of large-scale dataset inputs down to a manageable size.

There is also reinforcement learning, which includes algorithms that learn through trial-and-error feedback on its actions.

Examples of recommender systems

Amazon

Amazon offers a compelling case study for recommender systems. Through its complex algorithms, it can recommend products to people based on their past purchases and search history, as well as the purchases of similar consumers, among many other variables.

Netflix

Streaming platform Netflix uses recommender systems to offer TV and movie recommendations based on collaborative and content-based filtering.

Spotify

Music platform Spotify makes user recommendations based on people’s listening history, preferred artists and genres, and other user data.

MovieLens

MovieLens recommends films to people based on their ratings of other movies.

YouTube

YouTube uses recommender systems to offer users new content suggestions based on their past watch history and interests.

Learn more about recommender systems

Explore recommender systems and machine learning in greater depth while gaining specialist skills in a fast-growing area of computer science with the flexible MSc Computer Science with Artificial Intelligence at Abertay University. This flexible master’s degree is taught part-time and 100% online, and is suited to mid-level professionals who want to move into computer science, or take the next step in their existing computer science career.

You will develop specialist knowledge in artificial intelligence, including how to use it to improve processes and how businesses and systems can be improved through AI techniques. You’ll also gain a firm foundation in the field of computer science and data science, with core knowledge in networking, databases, and web development.