麻豆约拍

Welcome to the regular update from the Internet Research & Future Services team in 麻豆约拍 R&D.

For the past couple of weeks in the Internet & Society team, Oscar, BenH, David and Libby have been showing our bird-identifying machine learning explainer to some students in years 10 and 12 to see if it makes sense to them. Everyone seemed to enjoy using it, and we got some great pointers for where to go next. Thanks very much to Oscar for arranging it.

For our work joint with the Data team on a 麻豆约拍 Introducing prototype, Alicia and the team have been refining its appearance and functionality, while David's been working with BenH to figure out how we can best test it with the 麻豆约拍's mixes curators. The interesting part is - are the automated tags (things like genre, bpm) that the system generates actually useful for curators in navigating the similarity between tracks? What about the matches with 麻豆约拍 radio stations and mixes that Kristine created? Stay tuned, we're finding out...

Also part of our explaining ML work is Tristan, David and Miranda's - this seemed to go well (as far as anyone can tell on Zoom) - we got lots of comments, ideas and pictures.

Alicia did an excellent talk within the 麻豆约拍 about how and why she made her imposter syndrome chatbot, and got a lot of positive feedback. She's writing it up as a blog post and collating the feedback from her survey. You can still .

Chris is busy collating 麻豆约拍 feedback on the , as well as running his and helping with others.

Libby and David have been trialling short "Futures" experiments that can be run quickly with small groups online. We think we're onto something!

Libby's been doing some pruning of the bird machine learning model in her 10% time using . Pruning removes low-weighted in the model and is supposed to make it smaller and faster. She eventually managed to stop the "pruned" model simply recommending the same 5 birds in the same order, and even slightly improved the success rate. But for some reason the model is now 4 times bigger 馃槵.

The Data Team are just embarking on a new release so there鈥檚 lots of exciting work in the offing. In the Recommendations team, Ben is looking at how we can best help the iPlayer team make sense of their page impression data and deliver evidence-based decision making in areas like assessing sorted list recommenders, while Alex is carrying on with his classification work, looking at ways to help iPlayer curators who have to combine the best of automated approaches with their own editorial understanding of what customers want. Polina is carrying on with her excellent investigation into Public Service Recommendations and what this means for our research, and Ben and Kristine are coming to the end of the build phase of their Introducing tool and are looking to test and write up results.

In speech-to-text land, Misa, Matt and Ollie are spreading themselves across a similarly wide range of projects for the next fortnight, turning their audio set work into and independent tool ready for open sourcing, drafting a paper on X-vectors (and deciding what conference to present it at), delving into some new Russian and Farsi Kaldi models they鈥檝e got their hands on, and looking at a very interesting dataset of regional British accents.

In NLP there鈥檚 a similar panoply of projects with Tamsin鈥檚 work on non-news tagging moving into the realm of drama research, Chris exploring the Creative Works API and tying it to some 麻豆约拍 Passports data, Alex moonlighting on some extreme multi-label classification research and Ollie doing a similar thing with his side project on extracting training data from Newslab鈥檚 Slicer system.

Andy is carrying on with his DSRP workload of dataset curation, external engagements (including helping to kick off a Phd project with Ulster University), internal engagement and project management. And finally Matt is leading our infrastructure work by arranging to get a new set of hard drives installed as well as investigating R&D鈥檚 new public cloud.

The Interaction and Prototyping team are still heads-down user testing the multiplayer adventure. By all accounts it鈥檚 going great! They鈥檙e getting really positive feedback, to the point where one of the groups in the test loved playing the game so much they didn鈥檛 even want the incentive (but they will get it!). A group really wanted to play it again (and we agreed) in another session even if they were recruited only for one. There鈥檚 more testing to come, but we鈥檒l be publishing more about the prototype and its findings soon enough.