These days my search results page has started to look a little different. I see links that friends have shared earlier. I see a lot of the sites and links I’ve been to earlier, sometimes in a different context. Briefly, there were even Twitter feed results from my friends – though those have thankfully gone now.
Search results’ personalization has been the holy grail for a while now. After all, you’re shown – one imagines – what you want!
The heuristics that govern ranking are built around behaviour. To start with, linking to pages and sites indicated authority for them. And we all know that the euphemistically named SEO, link exchanges and other such mechanisms often manage to game this assumption of natural behaviour. And now the half-baked model of what a click on a link implies, or something my social group shares, makes the possibility of a SEO-self-goal very possible!
Why do I say this? Let me explain.
For starters – and we all have heard this numerous times – the first page of results is very very important, the second less so for users, and you’ve more or less lost us by page three or four. So those first few results are not to be trifled with!
I often go to a search engine to discover things I’m not sure I know about, much less know for sure that I want. Sometimes one search yields answers to what I might be looking for, sometimes not. Sometimes I’m researching stuff for a couple of days, and click on links which seem like they might have answers end with wild goose chases.
Can’t trust all my own clicks-in-the-past now, can I? Why let Google do it?
In the worst case scenario – when I’m looking for “other” stuff since the links I initially found did not really work for me and/or my understanding or depth of a topic I’m trying to get deeper into using Google has expanded – promoting stuff I already found – especially recently – is in fact counter-productive.
Things are on an even more uncertain ground when its shares from friends that’s part of the SERP. For one, I might’ve already seen content from friends. For subjects which have opinions, points of view and insights rather than plain facts, I would like using a search engine for a wider world view than one that merely echoes that which I’m comfortable with or familiar with.
Whatever you want it to be!!?
Of course there’s a case for including results I’ve shown an interest in sometime in the past. And for product and other factual information, my friends’ shares are immediately useful.
But their integration with “pure” results – which one assumes are based on relevance, and by trianing – on authority – in the current form is naive at best. Here’s some directions this could start maturing in over time:
- For starters, do not train users to use results as “bookmarks” – “You visited this page 20 times” is surely a reason to NOT include it in my search results, and move it into another bucket which I could look through if needed. I mean I probably own that page, or am a member, or some such – and don’t really need a search index to find it.
- Use earlier clicks to sift through or define affinity for concepts for me, and only slightly. The specific link/site may not be directly key, unless it “naturally” shows up because of a higher affinity to some concepts.
- Try and get a sense of the kind of data a page has, and the kind of query a user has fired. Does it seek information, or is the user “fishing” or “researching” ? This is, of course, non-trivial and will take time to mature. So keep the whole thing about including/boosting friends’ shares down for now – and perhaps solve it through a smart interface instead? I might get curious if some folks I trust seem to have shared content with matches and might ask for it if presented with the option.
Beyond keywords, search engines try to model our behaviour and intent at the point we create data, and fighting the battles where this behaviour and intent are gamed is a challenging enough one to solve. Adding social data sharing – where the behaviour and sharing get way more complex given the huge diversity in the nature of audiences and interactions involved – makes this modelling a completely new and very very dynamic problem to solve. Simple interfaces that communicate the nature and source of the data, and provide tools to mine it independent of the “web of content” are likely to lead to more useful set of results.