Another look at implicit feedback and beyond
Since we simply can’t get enough of the social web world and the trappings of ‘real-time social search engines’ – it seems that having another look at potential signals Google could be using (today or in the future) would be fun to do. This year alone we've seen increased activity with FriendConnect and Open social not to mention Google Reader even, which recently got more social... it's in the air if you take a deep, slow breath, you can almost taste it.
But how does a search engine become social? This isn't the media darling of the year, 'real time search', this is more about personalizations based on yours, and your networked friends activities. Let me show you what I mean....
It is always important to note is that major engines such as Google tend to be constantly evolving. As goes the technology, so does the evolution. This journey into the social sphere that we’re in, will likely be no different. There will be a process of hit and miss layers on top of the existing infrastructure.
Delivering quality results is always a process of understanding user intent, ranking relevant documents and dealing with the ever present spectre of spam. Think of it as the trinity of search…
Can Google be social?
And so first off; what is a social search engine? Well, everyone's fav Wiki says; “A social search engine is a type of search engine that determines the relevance of search results by considering the interactions or contributions of users.”
For Google’s part, Marissa Mayer has said, “We believe social search is any search aided by a social interaction or a social connection” – via Venture Beat
Or every SEO's pal, Matt Cutts, "(...) improving search by unlocking the power of people" via his Blog
What would a social Google look like and how would they get there? It's an interesting question; so we’re gonna take a stab at answering it (from the depths of a search geeks imagination) in the hope others will jump in as well…
Now then, there are a variety of methods and signals that would like come into play including;
Core systems - Google is already crawling merrily around the web doing discovery, indexation and ranking of a wide variety of media, documents including social sites. They are also employing methods such as QDF (query deserves freshness) to better hon in on more immediate temporal query spaces, (not everything is part of the social sphere ye know). Much of what they’ve already been doing can be adapted.
One thing the social world is certainly good for is discovery. Much like people are finding new content through social sites, so is Google. This is one area they are certainly adapting to within the core system.
Implicit user feedback (and grouping) – the next tools that would certainly come into play are would be those in the implicit user feedback group. Implicit feedback is when users are giving (implied) information, generally through their actions. Some examples being;
- Personalization (Search history); this one has been around for awhile and we can say that looking at a users habits may well provide some great ‘social’ signals. If one spends 60% of their time on Twitter (or searching related topics) 20% on Facebook and 5% on say, Sphinn; patterns emerge. You could weight documents returned in a search accordingly along a social level (User Social Score).
- Common user types – another area more popular with things such as personalized PageRank and Social Network Profiling are groupings. Essentially it can be very resource inhibitive to collect and correlate data on a user-by-use level, thus grouping is more efficient. They can use collective groups where they are implied (social interactions, site types, other related users etc..) to refine search results and recommendations.
- Query analysis (click analysis) – much like search history, your actual click data could be used to enable further groupings (as mentioned above). We already know that personalization is expanding past simple search history and that query analysis is popular with elements such as QDF. By understanding your click data, they can begin to better profile you which aligns well in adapting to the social web.
Explicit user feedback – this is when you are actually telling a search engine things that you like or dislike. Unlike the implied, one would (hopefully) get much clearer signals. Of course the problem historically with this is users tend not to like dealing with this extra layer. Some examples include;
- Search Wiki (voting) – we all know this and it’s adoption has been less than stellar from what I can tell. Hopefully over time users may start to use it more and they can easily get some strong signals worth implementing into personalization schemes (and ultimately social signals).
- Favourites (browser, reader, alerts etc..) – another way that Google has looked for signals are in the bookmarks via either toolbar or more recently with Chrome. Furthermore there is a logical extension to Google Reader and other applications (think Andriod and Google OS). They would be able to get more social signals from this element for use in many ways including the categorization mentioned earlier.
- Sentiment analysis – another interesting area that was released with the recent Searchology updates is ‘reviews’. This type of sentiment analysis can be yet another layer in any ‘social’ search engine. These are explicit feedback signals which can also be used (and are) to establish a more social implementation.
- Page Interactions – which can include printing of a page, emailing to a friend, scrolling and so forth. While not directly as useful in social signals, Google can certainly use these signals for establishing common user types and even social network profiling (see below) signals.
Social network profiling – last but certainly not least, are the ever fascinating set of patents on network user profiling (so-called FriendRank). This system works by understanding social user web graphs (where the users are nodes) and relating potential recommendations (of content, ads and other users etc..).
Not only is the system a clear sign of understanding social web users, but it also explains a lot when it comes to initiatives such as OpenSocial and even Google Connect. Many of the signals we’ve already touched on would play into this type of a system greatly. There is definitely a framework to better implementing social signals into the core search offering (or vertical search even).
Scratching the surface…
These are just some of the different methods that are already available to Google which could go a long way to creating a more social search engine. Let’s remember that Marissa defined it (loosly) as integrating a, “(…) social interaction or a social connection”.
As we can see, there are plenty of signals which fall nicely into that…
Problems with social search
Now obviously there are challenges beyond implementation and resource management. Some problems can include;
- Ease of use; the larger masses tend to like things that are simple and easy to use. To me most people don’t want an active roll in search. This means past the ‘gee wizz’ phase, the bulk of the results would be the few editing for the many. The power users would begin to give an editorial slant to the process that is skewed
- Spam-ability; once again nothing stops web spammers from infesting them and at a certain point people are ‘turned off’ poor results. They want the search engine to do the quality control.
- Freshness – going hand in hand with spam, the more real time a search engine becomes, the more open to spamming it is. Also, simply ranking based on trusted user types, removes some of the democracy of the web. There are plenty of problems with real-time search.
- Social Gangs; obviously we also have the problem of voting gangs. Social search engines, should they ever gain popularity, would be open to all kinds of shenanigans that would like need to be addressed editorially as well as algorithmically.
There are a great many obstacles that would need to be dealt with and the least of which is simply processing all the data and mining meaningful (but efficient) signals which would produce greater value at minimal cost. We might tho don some tin foil and assert that the recent infrastructure upgrades might play into such plans…
Indexing and regurgitating the world’s information is no easy task….it will be interesting at least to watch.
Only time will tell
Ultimately Google is certainly well positioned to be the social search engine of the future. More and more they’d evolve into more than a simple query dependent interface. Already there are more and more elements of being a recommendation engine as well as simply a search engine. There is every reason to believe they have the capacity to start using some of the massive stores of data available to them.
We’ve been hearing about social search for a few years now and it is highly unlikely that it’s gone un-noticed by those in the know at the ‘Plex. To those fond of striking out to say that Google need be worried about the spectre of social search, I’d say to them that they’re not only tuned in, but making moves towards meeting that need.
By the sheer mass of data they have available it’s foolhardy to think that Google isn’t still the likely leader once more… but only time will tell. Next time out I've been making a list of ALL Google potential avenues for social search and shall be listing them for you to brainstorm yer own mad theories... Until then...
Grab the feed (new and improved even.. remarkababble..)
(What do you think? Do they inherently get a leg up with data and cash arsenals? Or can a new player capture the ‘social search’ market?)