StarNow is a Wellington-based company started in 2004 by three guys who all worked at Trade Me: Cameron, Nigel (who also started FindSomeone) and Jamie.
They had a good idea: to build a website where people who want to be famous can advertise themselves, and where reality television producers can go to find the next sucker star.
And, they’ve done a great job of executing and have grown the site into an excellent business, which they now call a “global talent casting service” covering actors, models, musicians, dancers, entertainers and photographers.
This is how Matt Cooney described their business in Idealog:
“For New Zealand, this might be the perfect business. Almost every cent is earned offshore. Start-up and infrastructure costs are low, there are almost no transport costs, no expensive offshore offices to maintain, no worries about import duties, foreign exchange hedging or oddball tax regimes. No factories, no shelves to stock, no resource consents. Marketing is largely through word-of-mouth and the business scales beautifully. It’s proof that three guys with some brains, a great idea and a couple of hundred bucks for marketing can literally invent a business in a few weeks.”
Disclaimer: that’s a bit of a vanity link, as Matt also interviewed me for the same article.
I got their latest newsletter and had to smile when I read that Cameron, who is now CEO of StarNow, was asked to be a judge in this year’s Miss England pageant.
It’s a tough job … but somebody has to do it, I suppose. :-)
Rowan,
Is the dating website FindSomeone you’ve quoted a Trade Me owned one? I have some suggestions to make the site improve the relevancy of the search retrieval. As far as I can see that the site uses just SQL search. That’s fine, but the nature of searching for partner matches doesn’t fit well with SQL. It means that it spits everything to the user and the user has to sift thru to find the most relevant target he/she is looking for. This approach is called Case Based Reasoning (CBR). CBR means reasoning by matching how similar of an item (object) to another item (object).
Here is the main concept. Suppose you have a database named people that contains 3 instances (David, Daniel & Patrick) of an object, in this example our database is a human object database, where it has 2 (weight & height) attributes:
name weight (kg) height (cm)
—— ———– ———–
David 108.4 177
Daniel 88.2 183
Patrick 81 175
Suppose that a target instance called Mark, who is not stored in our hypothetical database shown above, has the following attributes: . If you try to do an SQL search for the closest person who has attributes that match Mark’s attributes, you will definitely get a NULL return. Here is the SQL snippets:
SELECT *
from people
where weight = 96
and height = 187
If the query is using CBR, you will retrieve all of the 3 instances but they are rank of how close are they (Daniel, David, Patrick) to target David. Note that with 2 attributes, it is a 2 dimensional graph. If you plot all the points on a graph, where coordinates are:
David = (108.4 ,177)
Daniel = (88.2 ,183)
Patrick = (81 ,175)
Mark = (96 ,187)
So, the coordinate point that is closest to that of Mark is the one that ranks highest, and so forth. The user can specify the cut-off as stated above. It is easy to visualize this example since it only contain 2 attributes, but it could be extended to higher dimension (multi-dimension), however the concept is the same. I can see that the FindSomeone website contains 2 attributes (gender, age). CBR can work in tandem with SQL. SQL should filters the regions first, the apply CBR to find similarities ranking.
CBR will score the percentage of how close ‘A’ is to ‘B’ compared to ‘A’ and ‘C’, for example. Suppose that a target object ‘A’ is queried from a database of 2 objects, ‘B’ and ‘C’. If ‘A’ & ‘B’ are compatible after retrieval (similar) by 77% compared to ‘A’ & ‘C’ which are only compatible by say 68%, then obviously ‘B’ is the most similar to the target (queried) object ‘A’. Suppose that the database contains 300000 objects and we want to find the closest match to the target object ‘A’. Using SQL query will give everything that satisfies the query conditions and it has no concept of similarities amongst the retrieved items in relation to the target. I noted that FindSomeone that the retrieved results could go upto 10 pages long. Using CBR, the user could specify a tolerance of similarity match of say no less than 80%, 90%, etc… The higher the percentage similarity the better the match is. Now, I haven’t found the dating site that is mentioned on this paper here, where it uses CBR technology, but definitely it is applicable in the dating service domain.
The Omnipresence of Case-Based Reasoning in Science and Application
I know that you have involved in Property Search in your previous endeavor prior to your Xero current role, but CBR has been used in Property Leasing & Renting market, where the concept of similarity search is also applicable, such as the following site:
Let on the net
The site above is often cited in CBR research articles as case studies. The site is a simple one, but it could be extended to include more features.
CBR has also been used for property valuation such as the following:
Fuzzy Case-Based Reasoning for Residential Property Valuation
The WEKA project, contains the algorithm for building a CBR system. It has to be extended to be able to do item inferences. There are also varieties of algorithms in there that can do object or face-recognition, again WEKA has to be tweaked a little for that. Perhaps users could be allowed to upload a picture of some persons (male or female) that they want to search for similar face in the database. The uploaded image could be added as a new attribute.