More on Social Search

So, yesterday I started making the case for social search as an excellent way to find information locked away within an enterprise, and the ability to get good information from social search pays for investments in social networks.

I didn’t cover much detail though. So, that’s the focus of today’s post.

Social search in my mind doesn’t replace traditional indexed search. Most companies have a search engine (or several) already, and any social network will come with its own engine to index its own data. The starting point for social search is aggregating results.

So, when you search for keywords, you get results from within the network, e.g. user profiles, groups, ideas, questions on Connect and Mix. Your keyword search should also be run against other search engines, e.g. Facebook allows you to run your search against Facebook or against the whole ‘tubes using Microsoft Live.

You can’t see these results together though. Social search in my head provides more integrated results, maybe not all jumbled together, but definitely surfaced as a single result set, with cues on the source for each result. Why you ask? Because people won’t spend much effort checking the other sources, e.g. how many of you have ever used Facebook’s Microsoft Live search? I’d be curious to see how much it’s actually used.

You need to have filters, natch, to allow people to refine their results, but I’m a proponent of a single results stream.

Now, that you’ve got a dump of results, you’ll need to add social metadata to each result. People should be able to comment and rate/like each result, maybe bookmark and tag it too. Yup, every single one. This sounds like huge overhead, but keep in mind that very few people will get past that first page of results. Building up useful social metadata will take time; look at Amazon as an example.

People should also be able to see other people’s metadata, their comments, who rated/liked the result, who bookmarked it or tagged it. This needs to be obvious for a couple reasons. First, they need to know they can do stuff to the results, and second, you want to surface and emphasize the social aspect.

The ability to take action on search results is a bit of mind-bender for some people, since the results themselves aren’t seen as objects. Also, the results should highlight social metadata from people in the user’s network, since in theory, that relationship has strength.

The goal is to get people to notice the results are social and contribute.

Beyond the UI, social search needs to tweak the algorithm to be really good. There are varying degrees of relevancy weighting you could try, but at a minimum, any results that have been socially marked up by members of a person’s network should be higher in relevancy.

This helps alleviate one problem that enterprise search has. Enterprise search decisions are frequently made at the department level (vs. corporate) to fill a team or group need. Plus, information in many enterprises is highly siloed and distributed throughout the company, and even if groups agree to open up their data, the big iron problem remains.

So, most enterprise search software can federate with other instances from other vendors, allowing for a more homogeneous results page. However, relevancy calculations differ wildly between search engines, making it very hard to normalize results with any accuracy.

Social search eases relevancy problems by placing top priority on what people say about results. You search for “benefits”. Someone in your network liked one of the results. That result is relevant.

This isn’t foolproof, since a comment doesn’t always mean a result is relevant. It could mean the opposite. You still need to have control over the weighting of social metadata.

Social search applied to information retrieval is a great example of the power of weak ties. Weak ties help you discover information from people you don’t know really well. In a work environment, lots of your connections will be weak ties.

Tools like Twitter/OraTweet and network status help you discover this information.

Frank’s comment on yesterday’s post gives an example. Paul has a similar story. He needed to find the corporate NDA, so he used his Connect status to ask if anyone knew where it was. Someone replied within 15 minutes.

Maybe, he could have found it faster by searching, but that would have required more effort. He’d need to get the right keyword combination and check the document to make sure it’s current. Instead, he asked an expert, saving overall effort and leveraging the combined knowledge of his network.

Classic case of improved efficiency. Now, that’s ROI anyone can support.

Social search is great at work. I’m not sure it’s great on the ‘tubes though. Google SearchWiki is an interesting example; first off, I don’t think it’s obvious enough, and the changes to results only affect you, making me wonder why I wouldn’t bookmark them.

Comments are public, however, and not very useful or polite. Surprise.

The enterprise has a chance to do social search much better for a number of reasons. Bex mentions one, no spam or viruses or phishing or misreprentation. Another is trust, which provides useful metadata. Because of the transparency, you’re likely to get very good comments. This becomes another way for colleagues to highlight their knowledge.

Trust also allows for better aggregation. Andrej pointed out that REST is a great way to query a distributed data model. This becomes an easy and standardized way to query structured data, which is the holy grail of enterprise search.

Anyway, the comments on the first post were interesting. Keep them coming.