Recommender systems
Imagine going to a shop to purchase a mobile. Immediately after entering the shop a salesperson will greet us and start querying about our requirements. We answer the questions and depending on our answers the salesperson will recommend a mobile. We purchase the mobile and happily leave the shop. Now imagine this scenario for an online shop. The recommendations, in this case, are given by the computer with the help of Recommender systems. Recommender systems help online shoppers navigate through products, recommending items based on search history and demonstrated interests. In short, they act as a virtual salesperson. This article aims to briefly explain the concept of Recommender systems.
Recommender systems were first introduced in the year 1979. A computer-based librarian named ‘Grundy’ was one of the first demonstrations of the Recommender system. Grundy would simply suggest suitable books to users depending on their interests. The first commercial Recommender system named ‘Tapestry’ was introduced in the year 1990 by Xerox Palo Alto Research Center. It was designed to filter emails based on topics a user indicated as relevant. Shortly after Tapestry, the recommender system “GroupLens” was invented at MIT. It assisted users in finding relevant items in the large amount of content posted in newsgroups.
One of the most prominent attempts to improve recommender systems was: the Netflix Prize. In 2006, Netflix, a provider of on-demand Internet streaming media, offered a prize of USD 1m to a person or group that could improve the existing Netflix recommender system by 10 percent. 20,000 teams from more than 150 countries registered for the competition. In the end, 2,000 teams submitted over 13,000 modifications to the existing system to predict users’ preferences. The competition ran for three years and the participating teams tested their modifications to the Netflix system using the big dataset the company had provided. The dataset included information about users, movie descriptions, and ratings.
Let us see how the Recommender systems work. Recommender systems give suggestions based on 2 types of data which can be background data and input data. The information which is already available in a system is known as background data. This data is not dependent on the user of the system. Input data is the data that is entered into the system by the user. This input data is transformed into background data continuously. Hence, we can say that the background data is regularly updated. The recommender system calculates the recommendations from these two types of data.
The two most common types of recommender systems are collaborative and content-based Recommender systems.
1) Collaborative Recommender system
In this system, new items are recommended to users based on the interest and preferences of other similar users. The similarity between users can be tracked by studying their browsing patterns, search options, purchase history, and ratings.
2) Content-based Recommender system
In this system, a user is recommended similar products based on the products the user has purchased or liked in the past.
Recommended by LinkedIn
Recommender systems are a powerful sales tool in this online world today. A humane touch can be given to a website by Recommender systems. A user is more likely to purchase products from a website after receiving personalized recommendations. A one-time purchase can be converted into recurring purchases by smartly implementing Recommender systems. Since many of us are avid online shoppers, we can bear testimony to this fact.
I would also like to point out some of the challenges for Recommender systems. The major challenge is tremendous data generated as a result of the social media usage of a person. Social media produces very different types of data like posts, replies, shares, and so on. The data can also be in different formats like text, links, or pictures. It is very difficult to transform this information into a format that can be interpreted by the system. There is also the question of data privacy. Since Recommender systems thrive on data, how they collect the data is a matter of concern.
In conclusion, Recommender systems are becoming increasingly important in this internet age. We do not even realize how most of the purchase decisions taken by us online are driven by these Recommender systems. With time Recommender systems will improve vastly in their suggestions and help us make better buying decisions.
Sources:
