Data Science London meetup at Strata Conference
This is a summary report of what happened there. O’Reilly invited Data Science London to host its community meetup at the end of Day 2, Strata Conference. The meetup took place at the grandiose Buckingham Room in the Hilton Metropole. Funded by contributions from community members, lots of delicious sandwiches and beers were provided to all meetup attendees for free. Everyone seemed to be enjoying the conversations and meeting new, interesting people. The room was packed with data scientists and data geeks, and –on the last count- we tallied up more that 275 people. Check out the photos.
The meetup was a special session dedicated to Recommender Systems. We had 4 speakers. Dr. Neal Lathia at Cambridge University Computer Lab, kicked off the talks and presented some of the highlights of 6th ACM Conference on Recommender Systems, Dublin. Neal did a great job, given the short time slot, providing an overview of key aspects of 6th ACM RecSys and also –importantly- summarizing 5 open problems in recommender systems. You can read a more detailed post on these five issues here but here is a summary what Neal presented:
- Why do we need recommender systems? We now implement recommender systems to foster engagement and community, and the web has become an ecosystem of personalisation
- Problem 1: Predictions. The research community has become very aware of the fact that there is more to recommendation than predicting ratings. How can you make recommendations novel, diverse and serendipitous? How do you deal with conflicting objectives?
- Problem 2: Algorithms. We need to find a balance between the effort required on users to rate things in order to improve recommendations vs. improving algorithms that can deal with few ratings and make better rankings
- Problem 3: Users and Ratings. The traditional mode of thinking about recommender systems has been “users” and “items,” who are linked by “ratings.” This paradigm is slowly being shown to be incomplete.
- Problem 4: Items. The idea of having tangible “things” that you recommend is also slowly shifting.
- Problem 5: Measurement. Understanding how to measure progress in recommender systems, and also ensuring that algorithm-people, usability-people, and academic researchers work closely are two main issues not solved yet
Second in line, was Tamas Jambor, a PhD student at University College London. The title of Tamas’ talk was Beyond Accuracy: Goal-Drive Recommender System Design. In his talk Tamas explained the differences between goal-driven and metric-driven recommender systems, and also provided a step-b-step, structured approach to goal-driven recommender system design. You can read the slides from his presentation here.