Lately, there has been a lot of discussion about the future of our information consumption. Are we going to be using search in a different way or using niche search engines? Are we going to get recommendations from from our social network? Will the results be personalized based on our behaviors, or even just a list of topics that we like? If you ask three different people you will probably get three different answers. The context can also change the answer. If someone is mobile, search may not be as relevant as recommendations. If you are looking at niche topics, then search engines are likely a better answer. When focusing on news, personalization like my6sense is probably the best option.
AJ Kohn has an excellent example of what Facebook could do. His thinking is centered around recommendation based on content context, not behavior:
What if Facebook added a simple More Like This link to certain news feed items? Clicking on the More Like This link would return a news feed with related content. In this instance, it would return Open Graph pages related to Samsung and HDTVs.
The benefit of context within your social network stream could also aid this process. So, if you follow a bunch of software developers, you likely see a lot of development related content which is already curated by your connections. Adding a “More Like This” link provides a recommendation layer on top of curated links. Facebook could get there eventually, but it is not a common idea for many sites. This type of recommendation is prevalent in ecommerce and has shown that it can grow revenue without a lot of marketing.
What the media industry really needs is some way to filter all of that information in useful ways, and recommend things you might not have read yet…social networks like Twitter and Facebook have proven to be the best way of getting recommended content.
In this little quote, they actually mention three different ideas. First there is the filtering of the news, typically into categories or even using tagging. Second, they want this information for stories that they may not have read yet. This is difficult as it means that you need to retain history of what you have seen, and possibly liked, in order to create the context of the recommended stories. The third part is actually curation from Twitter and Facebook, even though they call it recommended content. I am keeping the term recommendation for the algorithm that will “recommend” content based on whatever context is provided.
You will notice that I have mentioned “context” several times. I will mention it several more times until people really start to focus on the different contexts that people need to work with. As an example of different contexts, there is the group of people sharing links on Twitter and there is the group of blogs that I read in Google Reader. The links that I see in Twitter do not carry the same recommendation weight that the blog posts I see in Google Reader. Because I have curated my blogs, they have a higher likelihood of being something that I will like. I have curated the list of people I follow on Twitter, but not all of my interests match all of their interests. Obviously, recommendations based on a direct context will be much better than those based on a secondary context.
This leads to tools that are based on this secondary context but include personal behavior, like my6sense. The idea here is that you can improve recommendations due to the links that are clicked, even though they come from the secondary context. The secondary context becomes a limitedly curated feed and your behavior allows the personalization algorithm to create recommendations. The main drawback to any behavior driven approach is the lack of behavior data. In some cases, creating enough behavior data to provide relevant recommendations could take a long time. It is also not a good method when dealing with casual users of a system.
So, is there a holy grail? Absolutely not, at least not in a general sense. Because we do not have one place to look for all of our information, there cannot be one general solution. I use Google Reader on my PC. Other people use Twitter to get their news. Some people use Facebook. Others may do the same things, but while on a mobile device. Creating one system to capture all of these signals in the various contexts is tremendously difficult. An application could create plugins for each system in order to understand behavior better, but where does the integration stop? Do you stop with Google Reader, Facebook and Twitter? What about people that view news on sites like Yahoo or AOL? One thing that could work is someone creating the recommendation engine and providing a solid and free API. This would allow them to create an ecosystem of applications, all benefiting from the core technology.
How does search fit into all of this? Search has been the way that many people found information. Now, this information discovery is moving into the curated feeds of Facebook and Twitter. Search is still required when you need to look outside of your real-time stream. In addition to based search technology, adding recommendation and personalization could create a powerful way to discover high-quality information. Google has been trying to add a social layer to search, but it has received lackluster reviews at best. Facebook search is not nearly as usable as it should be and does not have the same quality of results as search engines. The combination of the two concepts could be one of the next killer applications.