Improving Customer Engagement with Recommender Systems

Since the past few years, online streaming businesses have been successfully using machine learning based recommender systems to increase and improve customer engagement, loyalty and sales, and retail companies are quickly following suit.

A recommender or recommendation system is essentially an information filtering system which aims to predict the rank or preference a user is likely to give to an item.

These systems aim to predict users’ interests and recommend items that may be most suitable for them, based on their previous interaction, search and buying history. For example, sites like Netflix and Spotify which will recommend movie titles and song/playlists based on the user’s preferences, or Amazon, which will recommend items that are like what you are searching for.

Recommender systems are used in many areas, including search queries, movies, products, music, books, and news selection.

Types of Recommender Systems

The Machine Learning algorithms in recommender systems are classified according to the kind of data that the system will use.

Content-based

Content-based recommender systems use characteristic information (information based on keywords, categories and user profiles and preferences). The hypothesis is that if a user was interested in or bought a specific product in the past, they will again be interested to buy or select the same product in the future.

In this type of recommender system, keywords are used to describe the items and then a user profile is built to indicate the type of item this user likes. The system further groups similar products based on their features. The recommender system utilizes the user’s profile and product features to make recommendations.

Content-based systems rely on machine learning techniques like Bayesian Classifiers, cluster analysis, decision trees, and artificial neural networks to calculate the probabilities of users liking a product. IMDB and Pandora Radio are examples of sites that use content-based recommendation systems.

Collaborative filtering

These systems utilize user interactions to filter for items of interest. Collaborative filtering methods collect and analyze information on users’ behaviors, activities or preferences. It then predicts what users will like, based on their similarity to other users. This approach does not require machine analyzable content and is better equipped to accurately recommend items without needing to ‘understand’ the item.

Machine Learning algorithms like the k-nearest neighbor (KNN) approach and the Pearson Correlation are commonly used in collaborative filtering systems. Facebook, LinkedIn, and other social networks use collaborative filtering to recommend new friends, groups, and other social connections. They do this by examining the network of connections between users and their friends.

Hybrid systems

A third category is the Hybrid system that combines both types of information. This is built to avoid problems that are faced when working with a single approach. The Hybrid approach can be implemented with multiple techniques — collaborative, content-based, knowledge-based and demographics. Knowledge-based recommender system suggests choices based on inferencing what a user needs and prefers. Demographic recommenders suggest products based on the demographic profile of the user.

Hybrid systems can be designed to use multiple techniques — it could use two different types of content-based filtering systems or it could unify both the content-based and collaborative filtering systems into one model. Another method that is often used is to make content-based and collaborative-based predictions separately and then combine them. Or you could also add content-based capabilities to a collaborative-based approach.

Hybrid recommender systems are used in Netflix. The website uses collaborative filtering to compare the searching and watching habits of similar users and content-based filtering to identify and recommend movies that have similar characteristics to films that a user has previously rated highly.

Benefits

Apart from widescale and successful use cases in media and e-commerce businesses, recommender systems are generally a good fit with businesses that have and use extensive data and have AI teams that can implement and benefit from these systems.

Enhanced customer experience and increased sales are a result of good recommender systems. Recommendations make it easier for users to access the content they prefer by speeding up the search process. The data about the data is used to send them personalized offers or further recommendations that are more likely to result in a purchase or selection.

This results in better user experience and the customer is more likely to stay loyal to the service or brand and purchase more. With comprehensive knowledge of each customer, the business gains a significant competitive advantage over the competition.

Originally published at www.triconinfotech.com on April 11, 2019.