Saturday, 9 August 2008

SUCCESSFULLY USING THE SOCIAL GRAPH TO DRIVE RECOMMENDATIONS


“We give nothing so generously as our advice.” Rochefoucauld

Businesses want us to promote their products. We are social animals, so we share good and bad news when we hear it. There is a herd instinct, but recreating it, or igniting it is difficult, but very rewarding if you can. Many of us don't give our advice online on Amazon with user reviews and ratings, or in video uploads on YouTube, but nonetheless we give our advice. Many advise offline after buying the product, or when their advice is sought. This word-of-mouth is the most powerful marketing force ever created.

The social graph is essentially the interlinking of the address book of people on social networks, other social media media platforms, and increasingly everywhere else with the advent of Google Friend Connect. These innovations offer consumers many advantages, but brands also have advantages such as the opportunity to seed viral communications, and word-of-mouth recommendations. But what form will these word-of-mouth recommendations take?

I've identified three types of recommendations:

Active Algorithmic Recommendations: This is when users actively share their opinions in ratings and votes, knowing that they will be aggregated and used to offer recommendations. Amazon's star rating system is an example of this.

Passive Algorithmic Recommendations: This is when users don't actively enter data, but their behavioural activity is used to offer recommendations. This data can be aggregated - such as with Amazon's collaborative filtering technology that aggregates purchase behaviour. People who bought this also bought this. Or the data can be used to present to your social graph recommendations that come directly from you, such as in controversial Facebook Beacon,which uses purchase and visit data.

Active Non-Algorithmic Recommendations: This is when a user produces a one-to-many recommendation such as with a review on Amazon, or perhaps more interestingly on something like lemonade.com, where they users set up lemonade stands of the products they recommend - see a previous posting on this.

The success of the social graph and web 2.0 in general is dependent on finding a business model that does generate recommendations. I'll be exploring this over the next few weeks in more detail.

2 comments:

Paul said...

A few other types of recommendations to consider:

content-based - recommendations based upon content analysis. If you like the band 'Weezer' a content-based recommender would recommend bands that sound like Weezer, irregardless of the social graph. Content-based recommenders are espeically helpful for generating recommendations for new or unpopular content.

rule-based - where a domain expert establishes a set of rule that are used to guide the recommendation. These types of recommendations are often used for big ticket items where social recommenders can't help. For instance, a social recommender generally won't help you buy a camera - the heuristic 'People who bought X also bought Y' just doesn't work for cameras.

hybrid - A combination of collaborative and content-based techniques - combines the best properties of both social and content based recommenders

Tony Effik said...

Thanks for these additional thoughts Paul.

I'd have agreed with your opinion on rule-based systems if I hadn't just bought my first SLR camera after looking at reviews on Amazon, and discovered a variation to the usual people who bought x also bought y. They listed cameras that helped by showing 'What Do Customers Ultimately Buy After Viewing This Item?"

Not sure if this aligns with what you are saying or not, but it hopefully shows how social search is more flexible than its orginal implementations.