By Chundyanto W.
It is hard to imagine the existence of e-commerce websites or any other popular entertainment and social media sites that do not utilize recommender systems nowadays. Recommender system is a technology to filter information in a website or a system in order to predict the rating or preference of a product for users. It has made a massive change in the way people interact with websites. In e-commerce websites, for example, users are easily guided to products they like according to their preference and taste based on their past shopping information. Another example can be taken from social networking sites, where users are suggested to connect with other users they may know based on their mutual interest, friends, or occupation. This improvement has certainly created a more attractive online experience for users by giving an ease of access through recommendations from the system.
Aside from providing an ease of access to guide users in interacting with a website, there are some other important benefits that are produced by the recommender systems. Recommender systems deliver relevant content to users because users will only be directed to items that match the users’ preference. This will lead to another benefit, which is assisting users in personalizing their profiles within the website. These efforts are meant to make users feel comfortable with the website, so that they will turn from just “regular shoppers, into loyal customers.” (Certona) Thus, not only it will increase customer satisfaction, but it will also help online businesses generate more profits from retaining customers.
Recommender systems can be classified into three categories. They are “Collaboration Based Recommender System,” “Content-Based Recommender System,” and “Hybrid Recommender System.” (Jones) First, collaborative filtering is constructed from users’ behaviors with similar traits. For example, users that buy book A are likely to buy book B, therefore other users that buy book A will be recommended with book B. Next, content-based filtering predicts users’ preference based on “the basis of a user’s behavior.” (Jones) Users that buy books about analytics, comment, and subscribe to analytics newsletter, will be suggested products or news that relate to analytics. Lastly, the hybrid approach is a combination of collaboration filtering and content-based filtering. This combination of approaches enhances the performance and usability of recommender systems in websites.
Some tools that functionalize recommender systems include LensKit, Scout Portal Toolkit (SPT), Recommender.org, and Duine Toolkit. To learn about how the tool works, we can take LensKit, a free open-source software to build recommender systems, as an example. LensKit allows us to create recommender algorithms through two main utilities, which are the “ItemScorer” and “RatingPredictor”. Based on the user’s profile, trait, and behavior, any numbers can be put into ItemScorer, then evaluated by RatingPredictor as an algorithm to determine whether the score matches or is similar to a related item. ItemScorer’s number can be based on purchase probabilities, which will then be scored by RatingPredictor to output the predicted ratings to determine the user’s preference and relation with a certain item. (LensKit) In addition, an interface called “Top-N recommendation” can organize recommended items that are scored in ItemScorer for users based on their user ID. (Lenskit) This way, an algorithm can be implemented into a website or a system to utilize recommender systems.
Big names such as Amazon, Ebay, Facebook, YouTube, and Netflix adopt recommender systems to be successful in their performance. To increase performance efficiency, most of these websites utilize the hybrid approach, or more than one recommender systems approach in their system. For example, Amazon uses the collaborative filtering to determine the likelihood of a customer with a certain trait to buy a product, as well as content-based filtering to narrow the product’s criteria to match the user’s preference. Netflix, the video rental and streaming service giant, flawlessly utilize the hybrid approach as well. It uses the collaborative filtering to see if users that watch a certain genre tend to watch the other genres or related movies. Collaborative filtering uses this information to make “individualized predictions.” (Bell, Koren, and Volinsky) Then it uses content-based filtering to narrow down the list of movies that may interest the user.
Tremendous functionality comes with a price, which applies to recommender systems. Although recommender systems work really well in terms of assisting users, there are challenges and limitations for the systems. First of all, a sufficient amount of data is needed to help predict users’ preference. It is hard for the system to predict a new user’s preference if the user only provides a limited amount of information and profile, especially if the user does not show a consistent pattern of behavior in the system. Changing user’s preferences can potentially be a problem as well. If the system only retains users’ past information, it will not recommend users based on their new preference. This will lead to a decrease in users’ satisfaction. (Muthukumar) Lastly, recommender systems are complex and can be difficult to implement.
It is clear that recommender systems open up a new challenge for online sites and businesses to compete for a competitive advantage in retaining their customers. Users always seek online experiences and transactions with websites that match their preference, which is mainly manipulated by recommendations from the system. Hence, websites with excellent recommender systems are likely to be successful.
Keywords: recommender systems, competitive advantage, benefits, limitations