A content recommendation engine is the process and platform that decides what content to recommend to individual users. Often times, this means showing pieces of content like blog articles or products to visitors on a website based on their user profile.
Companies that want to provide a personalized experience for their online customers frequently employ some sort of content recommendation engine. This technology can create dynamic webpage content for visitors as well as recommend products for shoppers. The top websites today -- including Amazon, Facebook, Google, and Netflix -- use content recommendation to better serve their users.
A content recommendation engine provides personalized content and suggestions to website visitors to optimize their experience
There are four steps involved in content recommendation: data collection, data storage, data analysis and data filtering
Collaborative filtering makes recommendations based on how similar a user is to other users
Content-based filtering makes recommendations based on a user's likes and dislikes
The hybrid recommendation model uses a mix of both collaborative and content-based filtering to make the most accurate recommendations
A content recommendation engine is a software solution that creates personalized user experiences by analyzing user and product data. The engine looks at a user's past online behavior, their likes and dislikes, and other key information, and uses that data to supply personalized content or make buying or viewing recommendations specific to that user.
It is through the use of content recommendation engines that Amazon can recommend products to you when you're shopping online, or Netflix can suggest new shows and movies you might like to watch. If you're brand new to the site, you might not find many recommendations -- or the recommendations you do have may not be useful. But as you continue to use the site and the site gets to know what you like and dislike, you'll find more and more accurate suggestions.
Some content recommendation engines feed dynamic content to the webpages you visit. Based on your past viewing behavior -- and the viewing behavior of other users like you -- the content recommendation engine can automatically generate personalized content as you scroll down the page.
Users like content recommendation because it provides them with a more personalized experience. Companies like recommendation engines because a more personalized user experience leads to increased viewership and purchases. It's a win-win for all involved.
Content recommendation is typically a four-step process. It's a matter of collecting data, storing that data, analyzing the data and then filtering the data to yield recommendations.
All content recommendation engines need data on which to base their recommendations. These metrics can be about the user (demographic information, buying/viewing habits, etc.) or about the products (keywords, description, etc.). Some data is explicit (gathered from customer input); some is implicit (garnered from customer behavior, such as order history).
The dataset collected must be stored in some sort of database, such as an SQL database so it can execute the recommendation algorithm.
The content recommendation system then analyzes the stored data and looks for relationships between data points. This can take place in real time or via a non-dynamic batched analysis.
The final step in the content recommendation process filters the data to obtain the relevant information necessary to make an accurate recommendation to the user. This is typically done via some sort of algorithm -- collaborative, content-based, or a hybrid of the two approaches.
There are three primary types of filtering used for content recommendation. Some models use collaborative filtering, some use content-based filtering, and some use a hybrid of those two methods.
Collaborative filtering gathers and analyzes a variety of data to predict what users will like, based on how similar a user is to other users. A collaborative filtering engine uses information about users' activities, behaviors and preferences, such as whether they like certain foods, movies or clothing. Predictions are made using various machine learning techniques.
The advantage of collaborative filtering is that it doesn't actually analyze or understand the underlying content. It simply picks content based on what is known about the user. That's also a disadvantage, as the recommendations often bear only surface-level similarities to what the user actually likes.
For example, if user A likes the same TV shows as user B, and user A also likes polo shirts, a collaborative filtering engine might surmise that user B would also like polo shirts and recommend polo shirt-related content to that person. If recommendations are based on enough data points, they can be surprisingly accurate. Recommendations based on fewer data points, however, may result in only superficial recommendations.
Amazon.com uses collaborative filtering for its recommendation engine. Amazon employs sophisticated algorithms to recommend similar products based on what customers have recently purchased to sustain retention. The site then displays those recommendations in the "Items you may like" section on each product page. The result? According to McKinsey & Company research, 35% of Amazon's sales come from product recommendations.
Content-based filtering takes a different approach. This type of engine leverages artificial intelligence to recommend items similar to ones the user has previously viewed or purchased in an effort to enhance the customer experience. The thinking is that if a person likes item A, and item B is similar to item A, then the person will also like item B. For example, if a user has watched or purchased one or more Marvel movies, the content-based filtering engine might recommend a Marvel TV show to that user, because they are obviously similar in content.
The effectiveness of content-based filtering is limited to recommending similar types of content or items to similar users. For example, knowing a user's movie preferences would be of little help in determining what types of foods that person would like.
For a good example of content-based filtering, look no further than Facebook. When Facebook recommends potential friends for you, it does so based on your personal content -- where you live, where you've worked, where you went to school. It's almost pure content-based filtering.
The hybrid recommendation model blends the collaborative and content-based filtering models. It looks at both customer usage data and content descriptions, and as such produces more accurate recommendations than either individual method.
Streaming video giant Netflix represents a real-world example of employing the hybrid recommendation model. To provide recommendations to its viewers, Netflix looks at the shows similar viewers have watched, as well as the content of shows you have watched. The resulting recommendations are more personalized than could be achieved otherwise -- and account for 75% of what viewers watch on the service.
Optimizely recommendations is a content recommendation engine you can add to your own website. It makes it easy to include recommendations on any webpage, and then measure the impact with Optimizely's Stats Engine. Choose the algorithm you want to provide the most personalized experience for your users.