Research & Development

Posted by Libby Miller on

IRFS are in our fifth week working from home and we’ve been thinking about the work we are doing and its relation to the virus. We’re very keen to do things that are helpful where we can, and we’re also using our lack of physical embodiment to build better relationships around the BBC.

The Internet and Society team have decided to wrap up their work on decision-making, and are figuring out how best to do so.

We have some really interesting results on techniques for individual decision making, and Alicia and Kristine are turning that that into a #newnews style interactive explainer.

Our user research on group decision making with groups of housemates yielded lots of new information on leadership, influence, consensus and culture within shared houses, and we’re working out what best to do with those results.

We’re super-happy that the results of our research with young people on how they form and use digital memories is now public as a website. There’s a BBC-only version with quotes that we can show colleagues on request, but the public site shows all the findings and conclusions. There’s more information in this blog post by Holly and David.

Chris meanwhile, is continuing his sterling work with W3C, starting up a collaboration with DASH-IF on the proposed DataCue API, as well as joining the new Privacy Community Group, which aims to improve user privacy on the web.

Like a lot of people, we’ve been thinking about what we can do to help in the current situation. Libby, Jess and Alicia have been working with a group from around R&D and further afield on a synchronised player for BBC content, and Henry and Libby have been working on a buzzer for remote gameshows - but we’ve realised we need to look further into the future after these quick, small, useful things. How should we direct our research in the next 6 / 12 / 18 months in the new circumstances?

Screenshot of a zoom meeting without participants

Meanwhile, Team Anansi have been continuing to gather insights from their work on synthetic voices. They have sent questionnaires to those people who contributed a voice about the personality and quality of their own synthetic voices, and received feedback so far from twenty-two people.

For practical reasons, Looking For Nigel is on hold (it’s a physical game in space). But they’ll return to it later. For similar reasons, Not A Robot 0.2 can’t be completed as is, but the team are adapting it for online play, using audio assets from the experience and existing R&D device orchestration technology in collaboration with the Audio Team.

The team are also collaborating with BBC News Labs on segmented media discovery, where they have gathered requirements from tech, journalists and UX. They are working with a journalist to refine the data requirements and hierarchy, and are currently developing wireframes. This project is a fully remote collaboration between teams who never worked together before. It started just as R&D went into lockdown.

Since the last weeknotes, the Data Team have held their hackweek on content similarity, which involved understanding problems with current metadata, figuring out what other metadata could be created and then creating some examples of how it might be used, such as creating thematic journeys, discovering clusters, and for diversifying content.

Since then they have been focused on three main areas: speech-to-text, cross-media recommendations, and sentiment analysis. They are making improvements to their speech-to-text system, including better segmentation and accuracy, building more up to date language models, and understanding how the use of words change over time. Ben, Alex, Chris and new recruit Polina (welcome Polina!) are working on cross-media recommendations, and Chris and Ben are helping iPlayer to test and compare recommendation algorithms. Denise and Ollie are continuing to work on sentiment analysis.

 

This post is part of the Internet Research and Future Services section