NoTube’s Beancounter software is a user-profiling service that implicitly determines a person’s most up-to-date interests based on an aggregation of their existing activities on the Social Web. This profile can then be used as input for personalised TV recommendations.
During the last few weeks, Lora and I have been conducting some user research to test some of the ideas relating to the prototype. This has involved running an online survey (completed by 130 people) followed by a workshop with seven participants to explore some of the survey findings in more depth. Thank you to everyone who gave up their time to take part in both these activities, and to VU student Christoph Eilers for all his help.
NoTube Social Web and TV Survey 2012: Demographics (click to enlarge)
As part of the convergence of the Social Web and TV, one of the key issues we wanted to explore is how people feel about trading their personal data in return for more personalised TV services, including those incorporating friends’ data. With the current interest across the Web in data mining of activity and Social Web data as input to recommendations and other personalised services, this is a topic of growing relevance and importance.
It is interesting to note that the findings of our research are consistent with previous observations highlighting the paradox that, whilst we’re keen to see what other people are doing, we’re less keen to reveal the same kind of data about ourselves to others.
What other people do is interesting…
The Beancounter only collects data from sources you’ve given it permission to access (your Twitter, LastFM, and Facebook accounts for example) and it keeps your user profile private by default, unless you explicitly decide to make it public.
However, our idea is that by sharing your profile, you’d make it available for comparison – for example, it could be matched with people who have similar or contrasting tastes or behaviours to you .
- Nearly half of the people in our survey (48%) said they want to know which of their friends have similar TV viewing habits.
- Unprompted, several of our workshop participants said that, using the Beancounter, they’d like to be able to connect with new people with the same TV interests/habits, and to see which of their friends watch the same things as them.
A shared profile could also be used to influence other people’s TV recommendations, and likewise your own recommendations could be influenced by the profiles of other people whose tastes you like. Attempting to increase content discovery through social recommendations seems to be a shared objective for many Web services, including Facebook and Google. And during the recent user testing of NoTube’s N-Screen prototype, participants responded very positively to the concept of swapping TV programme suggestions with specific individuals. Again, this notion seems to be supported by our findings:
- 62% of the people in the survey said they are interested to know what their friends have been watching.
- When the survey participants were asked about the various different methods they use to find new programmes to watch, 77% of the time they get suggestions from friends, followed by 57% by reading or hearing about a programme mentioned in the news, 38% by browsing the electronic TV guide, 36% by using online recommendations, and 28.5% of the time by reading TV listings magazines. The results of the 2011 Ericsson ConsumerLab study on global consumer trends for TV and Video similarly found that advice from friends and family are the most used and trusted recommendation method for TV content.
…but privacy is still an issue and we want to feel in control of our own data
Despite the desire to see other people’s data, our results also suggest that in general people still feel relatively cautious about sharing their own activity and preference data on the Web:
- 74% of our sample disagreed with the statement that disclosing their data online is ‘not a big issue’ for them.
- Whilst they are willing to share data about most of the things they do online (their activity data), as shown in the graph below, their preference is to share it only with those closest to them – i.e. their friends and family, not with everyone – or even with everyone in their social networks, as is the case with the current trend for frictionless sharing.
- There is a strong reported preference for a high level of control over personal data: 94% agreed or strongly agreed with the statement “I want to be able to delete specific activities and preferences”, and 73% with the statement “I want to keep certain programmes I watch private”. Our workshop participants also agreed with these sentiments.
NoTube Social Web and TV Survey: Attitudes to sharing different types of personal information on the Web
These results validate many of the assumptions behind the design of the Beancounter user interface, which keeps all your data private by default, allows you to delete specific activities, and allows you to delete or hide specific interests. We felt these features were important to help protect users from inadvertently making public potentially sensitive or incorrect information that might emerge when previously disconnected pieces of personal data are combined and analysed for patterns.
The results are also consistent with initial findings of some recent BBC user testing for the European Future Internet research project. Participants in the study also said they wanted control over their data, even though the perceived privacy risks associated with the data (the BBC programmes they had watched on TV) were relatively low and the perceived benefits, in terms of examples of more personalised future TV services, were relatively high.
So, we’d like to have granular control…
Reasons for wanting such flexible control would seem to be attributable to a variety of factors, including:
- A general awareness, and concern about, the possible risks associated with disclosure of personal information online.
NoTube Social Web and TV survey: Attitudes towards risks associated with loss of privacy online
- A general sense of wanting to be in control.
- The feeling that the information shared won’t be of interest or relevant to other people.
I may be happy to share selected bits, i.e. a TV show that I think is great, without sharing that I watch the 10 o’clock news!
I share music tracks, books and movies *explicitly* if I think they are novel or unknown and deserve promotion. I don’t really like implicit sharing services such as Spotify. I carefully manage the photos, tastes and opinions that I (publicly) share. Websites I visit are private, unless I decide to share them.
It’s more interesting to other people to know what you *really* like – i.e. aggregated data, like you’d watched something ten times.
Although I’ve said I’d be happy to share most things, I would like to be able to choose certain things. Also, at the moment I purposely don’t share/post these things in social media feeds as the way they’re shared is often very irritating for other people.
- The need to correct things about ourselves, even if the data is private.
I want to be able to delete things I don’t want the system to know about me.
My sister already messes up my Spotify listening profile, I don’t want to share it and get useless suggestions.
- Anxiety about what other people might think or imply about us based on particular activities – e.g. watching specific TV channels with a political bias, gossip programmes, or programmes which might be seen as a bit babyish. Workshop participants also mentioned wanting to keep quiet about watching a lot of TV in the middle of the night, or revealing quite how much TV they watch.
My mother always complains that I waste my time watching too much TV of the wrong kind.
Personal information, or any information for that matter, tends to change over time. Therefore, there is a risk that disclosed information is no longer valid, and you are not able to change or delete that information.
- And perhaps, to a lesser extent, the desire to adjust the data to get more accurate recommendations. However, our workshop participants felt that the recommendations should “just work” so it shouldn’t be necessary to tweak!
However, as I’ve mentioned before, this requirement for granular control poses some design challenges for the user experience. It seems that there’s a careful balance to be struck between supporting the needs a core group of users by providing the controls they say they want, whilst not burdening less concerned users with too many choices.
We explored this idea some more during a card-sorting exercise in which we asked our workshop participants to group which aspects of data about their TV watching behaviours (programmes, channels, genre, time of day, devices, etc.) they’d be willing to share, and with whom (anyone, nobody, friends, family, people in your social networks, etc). It emerged that there was little consensus in the group about this, with each participant having their own individual preferences. Nobody said they’d be willing to share anything with anyone, or that they wouldn’t be willing to share anything at all. Unsurprisingly, ‘who I’m watching with’ and ‘time of day’ seemed to be considered relatively more private than genres (comedy, sport, factual, etc.) but otherwise there were no significant patterns.
However, we seriously wonder whether, in reality, people would actually use these controls as much as they say they would, and this will be a question for future research into the usage of the live Beancounter service.
Part 2 of this blog post will look at how our workshop participants responded to the Beancounter screens that represent a user’s profile. These are comprised of the activities collected by the Beancounter, the interests derived from these, and personal data analytics presented as graphs and charts.