OpenTV: Personalized Reviews
Plenty of people have a negative reaction to recommendations. The suggestions might be perfectly good, but they don’t much like the idea that a machine knows them. We hear that in customer feedback, and we can see it in the data — a rail explicitly labelled as recommendations gets less engagement than exactly the same weighting applied quietly inside an ordinary list or a set of search results.
So instead of hiding the AI harder, we went the other way. Personalized Reviews (patent pending) puts the model that does the recommending into the audience’s hands, in a form that reads as human rather than statistical.
The starting point was TV critics. Go back to ancient times when TV listings lived in newspapers (well maybe not that long ago) and the person writing them usually had a voice. Some were straight and factual, some funny, some cheerfully absurd about a Tuesday night schedule. Whatever else they were, they were interesting. Sometimes you’d even read the entry for a programme you had no intention of watching.
AI is usually not interesting. Left on its defaults it’s wordy, evenly-toned, and reaches for the same handful of constructions over and over. Configure it properly, though, and you can get genuine style out of it. So we built five critics, each with their own personality — from Mickey, who is a enthusiastic cynic, through to Robin, who is warm, gentle and safe to put in front of a child.
Here are those two on the same show:
| Critic | The Traitors |
|---|---|
| Mickey | Claudia Winkleman, her fringe and her fingerless gloves preside over a Highland house where ordinary people lie to each other beautifully. Somehow both cosy and cutthroat — the best thing on television, and you know it. |
| Robin | A clever game show where grown-ups in a big castle try to work out who is lying. Clever, grown-up-feeling, and a bit slow-paced for very young viewers. Good for ages 10+. |
The viewer picks who they want to hear from. That critic writes a short review of the title in their own voice, with your viewing history feeding into it — so the personalization is still there, it just arrives as an opinion from someone you chose rather than a number a system assigned you.
It’s all opt in. You ask for the review, you pick the critic. I’d argue that’s more interesting than quietly generating a default review for everyone and hoping they read it.
Come and find me on the NAGRAVISION stand (1.C81) at IBC, have a look and hear about our next step - bringing back the voice announcer!