October 16, 2024 at 12:00 am
Comments posted to this topic are about the item The Modern Algorithm of Chance
October 16, 2024 at 8:13 am
I worked on a product recommendation algorithm. For the most part if was a personalised sort order for results from a customer search. We recognised the danger of a self-fulfilling prophecy so 10% of the products from a search would be selected at random from the results beyond the 2nd page (people rarely go beyond page 2 of the results).
When we started the project it looked like a big data problem. 5 million customers and 250,000+ products and generating a scoring matrix between each customer and each characteristic for each product. However, we found that the attributes we had on 5 million customers produced 43 strongly cohesive clusters. When we looked at the product characteristics there were a lot that didn't exist in the real world. This collapsed the data down from trillions of records and terrabytes of data down to millions or records and low gigabytes of data.
The next logical product algorithm benefited from this collapse of data too. It became far easier to comprehend what the algorithm was doing and whether it was doing it as intended and to introduce a random element into it to avoid self fulfilling prophecies.
One area I think personalisation hasn't touched is the workplace. The debate on return to the office vs working from home is passionate. Some companies are adamant that people should return to the office as it was pre-COVID where the office was pretty much a one size fits all.
What is clear is that some people are on a spectrum with "suited to home working" being on one end and "suited to office only" being on the other. Could personalisation algorithms be used to affect the design of the workplace to make it more palatable? Can we match the work environment to the needs of personality traits? The goal being to design an office around what allows different groups of people to be productive?
Viewing 2 posts - 1 through 1 (of 1 total)
You must be logged in to reply to this topic. Login to reply