Content-to-content recommendation is Amazon’s “Similar items you may be interested in” section. Doesn’t sound very funky, does it? Well, we can do a lot more with such technology. It’s not just about Amazon getting you to buy more products. It could be, for example, recommending movies, podcasts, art, books or any creative content. Show users the best you have to offer!
Rumo uses, among other things, content-to-content recommendations in its system. To find the similarity between two items or between a query and an item, the system uses metadata. Metadata is a complex term for “information about a given item”.
The problem is that it is difficult to find the best compromise between too much and too little metadata. It’s the work of someone who knows his or her content and is able to assign the best keywords. And the expert needed to accomplish such a crucial task is none other than the client’s editorial team. They are best placed to assess which keywords are most relevant. In this way, we ensure the quality of the description rather than standard AI-generated metadata, allowing us to provide the best possible user experience. It is then up to the algorithm to match the user’s query with the metadata in the catalog and provide tailor-made recommendations.
What are the strengths of content-to-content recommendation systems?
1. Transparency: by using a content-to-content system, Rumo does not need to collect personal information to make relevant recommendations. Its analysis capacity combined with the human work of indexing content works perfectly!
2. User independence: the system needs nothing more than the query and the metadata of each indexed content. Whether your platform has few or many users does not influence the quality of your recommendations.
3. Avoiding cold start: content-to-content recommendations can suggest new items. It’s not necessary for several users to review the content before it can be recommended.
4. Optimized for niche content: by using metadata, content-to-content recommendations can recommend niche content that would not be recommended by a system based on item’s popularity.
Drawbacks ? Nope, never heard of.
A popular drawback of recommendation systems in general, is that it traps users in a “filter bubble“. If the user gets recommended the same type of content over and over, he ends up trapped in a bubble with content he’s already familiar with. The user is no longer exposed to information that could challenge him or her, broaden his or her perspectives and worldview.
And this is where content to content recommendations comes in handy! It helps users discover new content, as the algorithm strives to create new links and connections between content in its catalogue.