Sunday, May 30, 2010

Introducing the "Tastegraph" - discussing recommendations with LikeCube's Emmanuel Marchal

After a post on Tnooz about non-destinational search I spoke with Emmanuel Marchal a Director of recommendation based B2B search provider LikeCube (Crunchbase profile here) about the future of travel search and the power of recommendations.

I have not doubt that the future of online travel is recommendations. Finding a way to give people a targeted recommendation based not only on their demographic or individual wants but on their wants at a particular moment in time. I call this my EveryYou principle (EveryYou definition here, more articles here). Developing a way to do micro targeting at scale.

Marchal's LikeCube is a B2B software as a services solution that purports to provide travel review and retail sites with targeted recommendations and search. His opening sales pitch for LikeCube is similar to that of any search site - give me your data and I will give you the best results for a search string. The LikeCube twist on traditional search is that answers are based not on links and page ranks but on the commonality between the likes and preferences of the content contributors. Marchal pitched this to me as targeting results based on the users "Tastegraph" rather "Sociograph". Instead of linking recommendations and results to the thoughts and feedback of those in the same social network as a user (sociograph), LikeCube targets recommendations and search results from those with the same likes and preferences of the user (tastegraph).

He gave two great examples of how standard approaches to reviews on a content or retail sites do not result in a useful recommendation. The first example was the reviews around TripAdvisor's Dirtiest Hotel of 2010 - the Heritage Marina Hotel in San Francisco. It won the "dirtiest" award in part because 174 of the 330 reviews ranked the hotel as "Terrible". Clearly - according to this rating - no one would ever want to stay there. But 48 of the reviews rated the hotel as "Very Good" or "Excellent". That means 15% of the reviewers of the dirtiest hotel of 2010 think the Heritage Marina is a fantastic place to stay and are keen to stay there again. People with different tastes and needs can see the same hotel in a very different light. Review sites are trying to manage these contradictions by helping users to refine a recommendation by reviewer demographics. The plan being to allow a consumer to see reviews only from people of the same demographics to maximise the chance of getting a recommendation form people with the same interests. This leads to Marchal's second example. He showed me two reviews for the same hotel on Both reviews were by people self classified as "young couples". One rated the hotel a four out of ten the other a nine point five. Additional classification and demographic criteria is not the answer to making reviews more relevant. As Marchal says (and I agree), these two examples highlight the need for a next generation of recommendation search.

That's the theory of LikeCube. I like the theory and enjoyed the conversion with Marchal. The challenge is seeing LikeCube in action. The first implementation of LikeCube is Qype ( a European Yelp clone).

Unfortunately as a B2B solution that requires access to a client's data source, it is impossible to review LikeCube in action outside an implementation. With only one implementation to view it is very hard to make an accurate assessment of LikeCube in action. In LikeCube's defence they only launched last November (at World Travel Mart).

LikeCube's challenge is not winning over people to the idea of targeted tastegraph based search results, it is building scale and proving that their technology can do what is promised. Winning more implementations and proving the case behind the technology (and as a result winning more implementations). Is the standard challenge for a B2B software start-up - made harder by the consumer facing nature of a LikeCube implementation. If they can do it then I predict this to be a very valuable space to be in.

1 comment:

Phoebe said...

I appreciate the insight that it's better to base social recommendations on people with shared tastes, rather than those who happen to be in the same social circles.

Actually, at Jinni we reached the same conclusion for a different product, basing social recommendations for movies and TV shows on taste "neighbors."