Tapping the Power of Recommendation Systems to Build a Broader Web 2.0 Audience

Social networking websites allow people to share their interests with other like-minded users via groups and forums, music and video sections, and other networking features. With the ever growing number of web 2.0 sites competing for visibility, getting people's attention is no longer enough: it's also necessary to learn how to engage visitors, increase the length of sessions, and promote user retention. In a world of infinite options, recommendation systems play a crucial role in achieving these goals, helping match users to engaging content that would otherwise be untapped and invisible.

OVERVIEW OF RECOMMENDATION SYSTEMS

Recommender systems emerged as an independent research area in the mid-1990s, with the interest in this topic remaining high for the past decades both in industry and academia. Recommendations have since then become an essential part of many online retailers' economic models, such as Amazon and Netflix.

In its most common formulation, the recommendation problem can be described as the problem of estimating ratings for items that have not been seen by an individual user. Once ratings have been estimated (typically with values in a numeric scale), the items with the highest estimated ratings can be recommended.

Recommendation systems are typically classified according to their algorithmic technique and recommendation approach. Algorithmic techniques are referred to as memory-based when they calculate recommendations based directly on the history of user activities, and as model-based (or heuristic-based) when they use previous user activities to first learn a predictive model (normally using some statistical or machine-learning technique) that is then used to make recommendations. Recommendation approaches can be divided into four classes based on knowledge source: content-based approaches recommends items similar to the ones the user preferred in the past; collaborative-filtering approaches find matches based on what other users with similar preferences have liked in the past; demographic approaches suggest items based on a demographic profile of the user; and knowledge-based approaches provides recommendations based on inferences about individual preferences, sometimes using explicit functional knowledge about how certain items meet user needs.

Hybrid recommender systems combine two or more of the aforementioned techniques to avoid shortcomings such as the cold-start problem (the well-known problem of handling new items for which there aren't any ratings yet, or new users for which likes and dislikes are still unknown). For example, a knowledge-based component could compensate for the cold-start problem of providing recommendations to new users for which there isn't enough profile information available for a collaborative approach to have enough traction, while still allowing a collaborative component to identify unexpected matches based on shared preferences with peer users that no knowledge engineer could have predicted.

To learn more about different types of recommendation systems, read this study.

THE BENEFITS OF RECOMMENDATION SYSTEMS FOR WEB 2.0 APPLICATIONS

The increasingly low marginal cost of creation and distribution of web content, and rapid the proliferation of user-generated content, creates new opportunities but also poses new challenges to online businesses. In today's content-rich world, consumers can easily become overwhelmed by the number of alternatives available to them, with negative effects over e-business metrics such as customer acquisition and retention rates.

According to Barry Schwartz, author of The Paradox of Choice: Why More Is Less, the growth of options and opportunity for choices has three, related unfortunate effects: decisions require more effort; mistakes become more likely; the psychological consequences of mistakes become more severe. The undesired effects of too much choice can be minimized by finding ways to steer consumers to the content (videos, news articles, product reviews, etc.) that they actually want, or may find useful or interesting. In a context of more and more websites competing for attention, intelligent, personalized recommendations become a valuable tool to increase customer satisfaction, loyalty and profitability, by facilitating the exploration of choices users might want to consume.

Clearly, web 2.0 applications can benefit from the use of recommender systems by strategically leading users to a broadened content universe that goes beyond content popularity and past individual preferences, with the potential to increase both content consumption and customer satisfaction and loyalty. A mixed recommendation approach, combining collaborative, content-based, demographic, and knowledge-based components, can provide a social networking website with valuable opportunities to exploit an existing content inventory. Consider, for example, a website that allows users to upload videos and create discussions around them. A collaborative-filtering component could take advantage of the fact that many users provide ratings for a video after watching it to recommend items ("people who liked this video also liked..."). A content-based component could use information about user activity, such a the time spent watching videos of a specific category, to select potential good content matches ("since you enjoyed X, you may also like"). A third approach, integrating demographic and knowledge-based components, could explore valuable information provided by registered users in their profile pages, such as age, country of origin, preferred topics, and the importance they attribute to video quality or entertainment value, to recommend items that are a good match to the user's demographics, known tastes and intellectual style.

MAKING RECOMMENDATIONS WORK IN A WEB 2.0 ENVIRONMENT

Here are a few tips for making relevant recommendations to users in the context of web 2.0 applications:

1. Be contextual

Chances are, the movies you like to watch online by yourself after a day at work are not your top choices to watch with friends on a Saturday night. Likewise, in a community-driven website, a homeschooler who likes to exchange ideas with other home educators during the week might be more interested, over the weekend, in connecting with other people who share a passion about Broadway shows. Contextual cues such as the day of the week, the site area being visited by the user, and the type of search query being used during the session, can help determine the relevance of content items for the current user interaction, allowing a contextual recommendation method to enhance the accuracy of the recommendation results.

2. Be unexpected

If a user explicitly tells you s/he likes Alfred Hitchcock movies, recommending more Hichcock-related content to this person will probably be a waste of time. Recommendations should use uncommon sense to figure out things that not only will spark people's curiosity, but also be content that they wouldn't discover on their own. Dan and Chip Heath, in their book Made to Stick, explain that the most basic way to get someone's attention is to break a pattern. Suggesting a forum discussion or video content that is unusual but still relevant will surprise and attract people's attention, encouraging users to stick around and explore more.

As Malcolm Gladwell points out in this TED talk, people's tastes vary, and people don't know what they want. It's not realistic to expect users to be able to describe what they really like, and recommendation systems make it much easier to uncover, for each individual user, potentially interesting content hidden in the tail of the popularity curve. With the help of a good recommendation system, it's possible to emphasize the discovery of new, relevant and interesting content (not necessarily the most popular choices or items similar to the ones users have previously consumed), and make users feel that they are getting value for their time or subscription money.

3. Embrace variability

People are different. The more a website accommodates the variability of tastes and preferences, the bigger the chance that users will find something they like, and remain loyal to the site. Instead of simply promoting "what's new", social networking websites can take advantage of technology's ability to adjust to individual personalities and intellectual styles, presenting users with recommendations they will likely find attractive, and quickly adapting to changes in user's data, content inventory and business rules.

CHOICES, YES--BUT ONLY A MANAGEABLE NUMBER OF THEM

The potential benefits of recommendation systems for social networking websites are huge. Relevant recommendations can enhance the user experience for both first time visitors and loyal members, for which more comprehensive profiles exist and could be explored to surface novel and relevant material. Recommenders help promote potentially appealing but unknown items that would otherwise remain undiscovered inside a massive content inventory, and protect users from becoming frustrated with an excessive number of choices by narrowing down the possibilities to a manageable number of relevant options.