- Table of contents
- Welcome to Solid Media
Solidflix is a demonstration of the use of Solid in a social sharing and personalised recommendations app. It is inspired by Media Kraken.
You can use the latest deployed application here: SolidFlix, or you can clone the repository and run it locally.
To play with the application you will need a WebID or a Solid pod provided by solidcommunity.net or any other pod provider such as Inrupt, Solid Web, solidweb.me or using your self-hosted pod url
If you wish to deploy Solid-media locally, then please follow the instructions below to build the project.
Solidflix is developed using Typescript and snowpack. If you are unfamiliar with them, then you will need to have your npm set up. Please follow this tutorial.
-
Make a directory.
-
Clone on GitHub here:
git clone git@github.com:OxfordHCC/solid-media.git
-
cd solid-mediainto the directory. -
Set up your
snowpack
npm install --save-dev snowpack
Run
npm run start
The application will be opened on your local web browser http://localhost:8080.
A public version can be accessed https://oxfordhcc.github.io/solid-media/.
To get started with sharing your movies, you will need a WebID to log into the system, so that you can store movies on your Solid pod and share them with your friends. We recommend you register with solidcommunity.net. Alternatively, you could use other supported pod providers such as Inrupt, Solid Web, solidweb.me or using your self-hosted pod url.
Once you have set up a Solid/WebID then you can log in and start to add movies to your pod.
After logging in, you'll be able to add movies, add friends, view your personalised movie recommendations, view your movies (watch history, liked and wishlisted movies), your friends movies (watch history, liked and wishlisted movies), like and wishlist movies.
You can add movies by uploading your netflix viewing history csv file, or by manually adding movies with the search feature. If you input a keyword of your favorite movie, you should see a list of search results. You can choose the tick button if you have already watched the movie, or the plus button to add the movie to your wishlist. And the movie should appear in your list.
Once you have added movies in Solidflix, you can express your level of interest of a movie by the thumb up or thumb down button, and the movies will show up as your favorites.
One particular thing that Solid-Media supports is sharing movies amongst friends.
Click on the "Add Friends" button, and enter your friends WebID (eg: https://FRIEND.solidcommunity.net/profile/card#me) and click on "Add". Now, the newly added friend will get added to your profile (foaf:knows) and they will get added as a member to the /friends group with which your movies data READ_ACCESS is shared. For adding friends in real-time, a PATCH request is sent using UpdateManager from RDFlib
Once you finish adding friends, you will be able to see your friends' movies on the top of the Solidflix application.
Solidflix provides personalised movie recommendations which come from a content-based movie recommendation model fitted with TMDB 5000 movies dataset. Based on randomly samples movies from users' watched, liked or wishlisted movies, the recommendation model returns personalised similar movie recommendations with the highest cosine similarity score. To retreive recommendations, a POST request with a list of movies is sent to our hosted REST API: https://api.pod.ewada.ox.ac.uk/solidflix-recommender/. The code for the recommendation model can be found here.
The diagram below shows the architecture of Solidflix, where the client is the users' browser and we have our server for the recommendations. Once a user signs in and gives authorisation consent to Solidflix, the app will be able to store and retreive data from the users' pod. We are using the TMDB API to fetch movies metadata while storing movies on the pod, which takes in only the movie id's. The recommendation model only takes in movie id's, such that no personal data is ever sent back to the server.
The authentication is handled by Login.tsx and uses @inrupt/solid-client-authn-browser.
The basic structure can be supported using these lines.
async function login() {
if (!session.info.isLoggedIn && !new URL(window.location.href).searchParams.get("code")) {
await session.login({
oidcIssuer: provider,
clientName: "Playlist example app",
redirectUrl: window.location.href
});
}
}
Here we used getSolidDataset from @inrupt/solid-client to retrieve movies data from a pod.
const movieList = (await Promise.all(people.map(async x => {
try {
const parts = x.id.split('/');
const pod = parts.slice(0, parts.length - 2).join('/');
# retrieve the movies from a personal pod, in which `movie` data are stored on the root directory of the pod.
const moviesDataset = await getSolidDataset(`${pod}/movies`, {fetch: session.fetch});
const movies = getContainedResourceUrlAll(moviesDataset);
return movies.map(m => ({...x, url: m}));
} catch {
return [];
}
}
The metadata about a movie, including its title, a short description and an icon, is retrieved from tbmd:
const urls = getStringNoLocaleAll(movieThing, 'https://schema.org/sameAs');
const [tmdbUrl] = urls.filter(x => x.startsWith('https://www.themoviedb.org/'));
const {title, released, icon} = await loadData(tmdbUrl);
Here we used rdflib.js to set a local data store and associated data fetcher to perform the PATCH operation. We are using UpdateManager to “patch” the data as the data to update it in real-time.
// Set up a local data store and associated data fetcher
const FOAF = $rdf.Namespace('http://xmlns.com/foaf/0.1/');
// create a local RDF store
const store = $rdf.graph();
// create a fetcher to read/write
const fetcher = new $rdf.Fetcher(store, {fetch: session.fetch});
// create update manager to "patch" the data as the data is updated in real time
const updater = new $rdf.UpdateManager(store);
// RDF Statement (subject, predicate, object, why)
let ins = [];
ins.push($rdf.st($rdf.sym(webID), FOAF('knows'), $rdf.sym(newFriendWebID), $rdf.sym(webID).doc()));
updater.update([], ins, (uri, ok, message) => {
if (!ok) alert(message_f);
});
The diagram below shows that the sharing of resources on solid pods currently relies on the authentication to be set up at both the profile level and the pod level, where the group is setup as a vcard. Once a user adds a friend, they get added to users' RDF and to the vcard group as a member. READ_ACCESS for the users' movies data is shared with the /friends group.
If a user deletes any movie data or unlikes a movie, the movies shown to their friends get updated accordingly.
- Privacy Preserving ML -based Collaborative Filtering Recommendations
- Cross-app or platform integration beyond just Netflix (eg: Trakt, LetterBoxd, or other streaming services)
- Optimising data storage and fetching (Storing hundreds of movies takes a couple of minutes on NSS and CSS. While loading 1000 movies takes 3 minutes using NSS, and 20 seconds using CSS)
- Enabling a friend to control whether they would like to their movies with another friend, instead of having their movies shared by default when they become friends.