Everyday decisions, from which products to buy, movies to watch and restaurants to try, are more and more being put in the hands of a new source: recommendation systems.
Recommendation systems are changing the very ways we make up our minds, guiding us through a new digital reality whose evolution is bringing us closer to exactly what it is that we want — even if we ourselves don’t know it yet.
Recommendation systems (or RS for short) are intelligent information filtering engines that narrow the decision-making process to just a few proposals, and they’ve become an integral part of the user experience within some of our favorite platforms.
Just think about how different the experience would be if Netflix didn’t offer future titles to watch, or Amazon didn’t offer alternative products to purchase? What about suggested connections on Facebook and LinkedIn, or news recommendations on Yahoo? None of these would be possible without RS, and that’s just the tip of the iceberg.
Breaking Down The Recommendation Process
Recommendation systems are fueled by both our explicit and implicit interactions. Explicit interactions are mined through your expressed preferences via ratings or reviews and data that has been gathered on your profile (i.e., gender, age, etc). Factors such as location, weather, the device used for access and the time and date are examples of implicit interactions.
Derived passively, the information here is often gathered without user consciousness. Amazon’s recommendation system, for instance, can be broken down into a few simple elements: what you’ve purchased or viewed in the past (explicit), what shoppers with similar profiles have viewed or bought (implicit) and the date and time of viewing (implicit).
The recommendation systems market predominantly operates on a proprietary basis, with companies such as those listed above developing their own models built around one of three designs, or system “types” (highlighted below).
Collaborative Filtering
This model is built on the user’s previous ratings and actions, as well as the ratings and actions given by other users in the system, with the data then leveraged to predict preference. A successful implementation, however, often requires a sizable volume of data, creating the so-called “Cold Start Problem.” Also, such systems often push popular items higher in results than unpopular counterparts. Companies that employ this model include Amazon, Facebook, Twitter, LinkedIn, Spotify, Google News and Last.fm.
Content-Based Filtering (CBF)
In addressing the problems above, CBF methods focus on the “item” itself rather than previous user behavior, breaking down items into sets of attributes and characteristics to then recommend to the user similar items. The main advantage here is the avoidance of a cold start, as there isn’t a reliance on feedback provided by other users. However, in their current state, the results have a tendency to be over-specialized, and limited to the items that have been characterized. Websites like IMDB, Rotten Tomatoes and Pandora are popular examples.
Hybrid Methods
To overcome the shortcomings of the above approaches, there are various ways of connecting two or more recommendation systems, with each having its own advantages and disadvantages. While this route has the potential of outperforming individual recommendation systems, they’re usually more complex, not to mention computationally expensive. Netflix is one of the most popular companies that has implemented a hybrid model, investing upwards of $150 million each year to improve functionality.
How Recommendation Systems Are Evolving
Recommendation systems are quickly becoming the new way users are exposed to the whole digital world through a prism of their own experiences, habits and interests. They are quickly evolving to be less obvious, more persuasive and far more useful, for both business and users.
Cloud computing has played a massive role in the technology’s exponential growth in the last few years, allowing systems to run thousands of computing nodes in parallel and creating naturally more sophisticated solutions in the process. This has helped RS models meet one of the biggest user requirements: accurate results in real-time.
Still, in our current landscape, RS models predominantly follow a generic algorithm, bucketing users into consumer sets rather than a fully tailored approach. The future space for RS looks to be headed toward a totally customized environment for each user, which will be unlike anything we’ve ever experienced. Real-time factors such as mood, time of day, location, sleep cycle and energy output will be weighted. Depending on the information the individual is willing to provide, our social media history and offline purchases can also be added to the equation.
Like the evolution of most technologies, consumer trust plays a critical role.
The most immediate industries to benefit from this evolution are the one’s already investing heavily here: retail and media. The more brands and entertainment platforms know about consumers, the more likely they’ll be able to foster long-term engagement and build stronger affinity. By understanding a consumer’s history, taste, mood and interests, retailers and media platforms will be equipped to provide a hyper-tailored offering, a notion that has fueled the big players’ race through investment here.
Other industries that are taking notice include banking and financial services, looking more and more to gain a holistic view of customers to predict their next moves. This includes going beyond their financial account patterns and mining sources such as call center data, social media patterns, websites, emails and customer feedback to then personalize their offerings. Even early childhood education is poised to shift, with start-up technologies emerging that recommend the pieces of reading material most likely to help specific students improve, taking the burden off the teacher.
So what’s holding us back from going further here? Like the evolution of most technologies, consumer trust plays a critical role, and with sensitive data hacks making headlines every month, it makes you wonder how many of us will be willing to keep giving our personal information to these companies. To emerge as the winners in this market, the onus is now on businesses to prove security in their systems.
Only time will tell the breadth of impact recommendation systems will have, but all indications so far point to this technology driving a watershed of change across industry.