Earlier this evening I was scrolling through one of my social media feeds and I came across one of those posts where someone is asking, “What did I do to get an ad for [insert completely absurd product] in my feed?”
Designing a website that understands the likes and dislikes of a person is not a simple task. And the reality is, that website has to have some information about you to make accurate recommendations. We at Enterprise Knowledge build websites and intranets that focus on making your information easier to find – and sometimes that means we have to get beyond the passive (waiting for someone to enter a search term or navigate to the right place) to the active (recommending the right thing). So how do we increase the relevancy of our recommendations?
Option 1: Recommendations by Taxonomy
A taxonomy is simply a hierarchical list of terms which can be used to describe the content on your website or intranet. EK’s website, for example, has a topic taxonomy with seven topics including Change Management & Communications, Content Strategy, and Technology Solutions. This blog – because it’s about a technical solution – is tagged to the topic Technology Solutions.
As you read this blog, the EK website learns that you are currently reading at least one blog post in the topic area, Technology Solutions. We can therefore assume, that you might be interested in similar blog posts and recommend others which have been tagged to the Technology Solutions topic.
Recommendations by taxonomy are relatively simple to build into your website or intranet so this can be a cost-effective entry point to recommend content. While the relevancy of the recommendations is not as technically rigorous as some of the other options, relevancy can be improved when care is taken to design the most effective taxonomy for you organization – and to tag content accurately.
Option 2: Recommendations by Ontology
An ontology is a defined model that organizes structured and unstructured information through entities, their properties, and the way they relate to one another. Building upon an existing taxonomy, relationships can be created between the taxons, or concepts. As you read this blog, the EK website uses an/its ontology to know a lot more about the content including who wrote the blog (Rebecca Wyatt). Our ontological model knows additional information about Rebecca Wyatt through her biography on the EK website. Using this additional information, the system could recommend content regarding educational technology and the development of systems which increase knowledge transfer and learner outcomes. The content doesn’t have to be explicitly tagged with metadata – our ontology allows the system to infer these connections and make additional recommendations.
Recommendations by ontology are incrementally more complex than recommendations by taxonomy, and this increased complexity can improve the relevance of recommendations. However, the devil is in the details and these additional benefits are maximized if the ontology is effectively designed.
Option 3: User Preference-based Recommendations
In both of the previous options, content recommendations are made solely based upon what the system knows about the content a website user is currently viewing. You’re viewing this blog, we know this information about this blog, and you might like other information which is similar to this blog. This is a great approach in a highly restricted environment where it’s difficult to store information about the website’s user (such as a Federal system where PII is a concern).
In a less restricted environment, however, we can improve the relevance of content recommendations by building user profiles – allowing users to indicate topic-level interests – or by using cookies and analytics to store information about what types of content a user has viewed on your site in the past. In these examples, the fundamental approach is the same, but the user profiles and cookies give us more information with which to curate the right recommendations than one, single article that is currently being viewed.
Option 4: Collaborative Content Recommendations
Options 1, 2, and 3 are all examples of content-based recommendations – meaning they rely on similarities between the content on the site to make recommendations. If you like this piece of content, and other content is similar to it, you might like the similar content also. Even in option 3 – which takes into account historical records of user actions and user preferences – we’re still finding similarities between the content and one, individual user.
Collaborative content recommendations are a little different in that they rely on how other users have behaved in relationship to similar pieces of content. To create collaborative content recommendations, the system will create a user profile about an individual user (similar to option 3), but will identify users who are similar to each other – not just users who are similar to content. For example, on the EK website, we could create two groups of users:
- Technical Site Visitors and Developers: We could categorize visitors to the EK website who we know are interested in more technically detailed knowledge management information into this group. This could be site visitors who indicated they are developers or information architects – or they could be visitors who seek out technically detailed “how to” blogs posted by EK’s Technology Solutions team.
- Business-Oriented Site Visitors: We could categorize visitors who are interested in the business applications of knowledge management into this group. We might place visitors who indicate that they are non-technical executives or project managers into this group – or we could categorize people who routinely read EK’s blog posts about Change Management and KM Strategy in this manner.
After we identify which site visitors are most similar to you, we are then able to make recommendations to you based upon similar visitor activities. If a visitor to the EK website who is very similar to you liked this blog post, you might like it too.
Collaborative content recommendations can be very powerful, but they do depend on a large data set for their accuracy. Because of this requirement, this is a solution that is best leveraged for websites with a lot of visitors. This is how Yelp and YouTube recommend content to their visitors.
Summary
If you’re building a website or intranet that is recommending content to your site’s visitors, you’ll want to invest in the most relevant content recommendations. There are a lot of factors to consider when deciding whether recommendations by taxonomy, ontology, individual user preference, or collaborative content filters are the right solution. Need some help making this decision and building the right tool? Enterprise Knowledge can help.