Recommender Systems

As technology and e-commerce advance, the use of recommender systems has been driven by the large availability of user data and the interest within the scientific research community. The topic of recommender systems is addressed in different areas of research, for instance, machine learning, data-driven innovation, and human-computer interaction. Diverse industry sectors such as e-commerce, online music or video streaming, news, search engines, gaming, and even online dating apply similar techniques that maximize the utility of large volumes of data to fulfill a user’s needs in a personalized setting. This form of data mining has broad forms of implementation and allows companies to cater tailored experiences and increase customer satisfaction.

Today, recommender systems are used by many companies in a variety of application sectors. Each domain has its unique recommendation challenges. Most e-commerce sites and applications have a recommendation engines powering segments of their user experience. Many companies have discovered through the years that there is significant business value in incorporating recommendations to personalize their consumer experience.

1. Product Recommender Systems

Amazon is the first large e-commerce company attributed with a recommender system as their primary service. It has initially used a simple item-item collaborative filtering approach. Their current experience includes multiple levels of recommendation: from their homepage to many product pages, it produces a list recommendation of other products bought or viewed. Other retail companies such as eBay have also incorporated item recommendations in their experience.

2. Music Recommender Systems

Music recommendations is another popular application area. Pandora, for instance, centers their business model around creating personalized music stations. They combine collaborative filtering techniques with their curated content and user metrics data. Another company that also provides personalized music recommendations is Spotify.

3. Movie Recommender Systems

Netflix’s approach to recommendations changed a lot since the Netflix Prize in 2006. Its service changed from DVD-rental-by-mail to instant video streaming across many devices. When the Netflix Prize launched, it put a spotlight on the importance and use of recommender systems in real-world applications. Netflix depends primarily on the recommendation system to help people to find something to watch that they will enjoy. It does so by integrating recommendation systems to personalize as much of the user experience as possible.

4. Content Recommender Systems

News is another area where companies use recommendation approaches to personalize news content for a user’s interest. For example, Google News provided recommendations for news articles from the beginning.

5. Networking Recommender Systems

Social networks also provided new recommendation scenarios. Twitter, for example, introduced its Who to Follow algorithm to recommend new social connections. The combination of social with different kinds of signals is a very common approach for Twitter or Facebook.

6. Other Recommender System

Search or browsing often provides an alternative to using recommendations. A situation where a user enters a search query can also be an opportunity for the recommendation. For example, the auto-complete suggestions when the user starts typing can also be personalized and interpreted as a recommendation.

Explicit ratings by consumer are not the only form of data collected from users. Implicit feedback from users is much more reliable and easily available than ratings, and there is no need of extra effort from users. For instance, on a webpage, the URL visited by users or users clicking on a link are considered as a positive feedback. In a music service, a song played by a user is also considered as a positive feedback. Many of these recommender systems focus on helping users choose an action (click, buy, read, listen, watch). The previous actions or information is highly relevant for predicting future actions. That is why many recommender systems focus on more reliable and readily available data as implicit feedback from users. One of the algorithms is used to compute a personalized ranking is Bayesian Personalized Ranking (BPR). Explicit or implicit feedback can also be combined in different manners, for example using SVD++, logistic ordinal regression to provide a mapping.

Examples of Recommender Systems