About Cinevarium
Cinevarium is a place to rate films and series the way you actually talk about them: not with one number, but criterion by criterion ("Screenplay 8.5", "Cinematography 9"). Everyone's scores add up to community averages per criterion, and you can look for what to watch next by the criteria that matter to you. It is a non-commercial hobby project by Ali FINDIK, free to use, in English and Turkish, on the web and as a mobile app.
Data sources and credits
Film and series information, posters and images come from The Movie Database (TMDB). This website uses TMDB and the TMDB APIs but is not endorsed, certified, or otherwise approved by TMDB.
Wikipedia. The "AI" scores are derived from the critical-reception sections of Wikipedia articles, in English and in other languages, which are available under the Creative Commons Attribution-ShareAlike 4.0 licence. The texts themselves are not shown on the service.
How the "AI" scores are made
For titles with published critical-reception summaries on Wikipedia, a language model we run on our own computer, outside the service, reads them and expresses them as criterion scores on the same 1–100 scale users use, calibrated against the aggregate scores quoted in them. The same model first translates the summaries written in other languages into English; these translations serve only this purpose and are not published. The category prizes of well-known film and television awards, as recorded on Wikidata, raise the score of the criterion they reward slightly: by 3 points for a win and 1 for a nomination, and only where the summaries already gave that criterion a score. For titles without such a summary, the reviews of users who allow it may be used the same way. The model does not run in the website or the app, nothing is sent to an AI service and no TMDB content is used for it. The scores are not the scores of any publication, critic or rating website. They carry an AI badge and count in community averages with a weight of 30% of a user's score while nobody has scored that criterion; the more people score it, the less the AI score matters, and once 10 people have scored it the AI score no longer counts at all. Over time the model is calibrated with the aggregated scores of users (averages per title and criterion) and, where users allow it, with their reviews and the scores they gave the same titles. AI scores are estimates and can be wrong; if one looks wrong to you, tell us.
Score suggestions from your review
When you save a review and ask for suggestions, the same model reads your review and suggests scores for the criteria you wrote about, each with the sentence it is based on. Because the model runs on our computer, suggestions are usually ready within minutes, and later when that computer is off. They are shown to you only and become your scores only if you accept them; you can change each one before accepting.
Contact
Questions, reports and requests: info@alfion.com.tr.
See also the privacy policy and the terms of use.