During Summer 2014, I had the opportunity to make an internship at Antidot, a French company specialized in search engines and Semantic Web technologies.
In order to master the principles of Linked Data, I designed a mashup aggregating Youtube videos related to specific topics extracted from Freebase, and enriched with data from Wikipedia & Wikidata, thanks to Antidot technologies, AIF and AFS. The main idea upon which I developped this project was to propose a platform where it could be possible to browse, with an optimized set of filters, video content previously selected according to their relevance to current news.
I followed 3 major Design steps to refine my inital idea into the most compelling solution :
Aggregation -> Combination -> Vizualisation
Aggregation
My first goal was to conceive a video platform which would be automatically aggregating videos accordingly to news topics. The idea was to query the Twitter Api to identify trending topics, and then confront these topics to the available data and videos on the Linked Open Data Cloud, aka the Web of Data / Semantic Web. Relying on Semantic Web technologies can indeed be very powerful, as considered topics would be truly disambigated (rejection of similarities) and links between such topics and videos would remain strong and reliable, with a wide set of content to browse.
After a phase of exploration of the resources of the Semantic Web, I had to lower my expectations based on the facts that all the available resources were not equal in terms of interconnectedness or consistency. I decided to focus the workflow on a starting point of curated and aligned topics from Freebase. Such topics would be the roots for gathering the related videos and data from Youtube Data API. Extended information related to the topics (multilingual labels, geolocalisation, etc.) was gathered, in parallel, from Wikipedia and Wikidata APIs.
The Linked Open Data Cloud, as in August 2014 (Credits : Richard Cyganiak & Anja Jentzsch / http://lod-cloud.net)
After a phase of exploration of the resources of the Semantic Web, I had to lower my expectations based on the facts that all the available resources were not equal in terms of interconnectedness or consistency. I decided to focus the workflow on a starting point of curated and aligned topics from Freebase. Such topics would be the roots for gathering the related videos and data from Youtube Data API. Extended information related to the topics (multilingual labels, geolocalisation, etc.) was gathered, in parallel, from Wikipedia and Wikidata APIs.
Data workflow for the Videotrend concept (Icons credits : Alfredo Hernandez, Arthur Schmitt, Guilhem / The Noun Project)
Combination
From this promising set of data resources, I made my way into data thanks to the AIF suite, using several languages, standards and technologies to transform and reuse the aggregated data.
Screenshots of the AIF Back Office
Visualization
Thanks to the AFS framework, based on a Boostrap grid layout, I designed an interface where both data and content could be browsed with the most ease.
Because of the multiple video players embedded in the results page, loading time could have been a real issue for the end user. The solution came from hiding the video player while they were loaded, and pushing the video thumbnails first, as the main descriptive element. Clicking on any asset would trigger a JQuery function which would hide the thumbnail and bring the video player to the front.
Screenshot of the interface
(At the time of the creation of this post, the website was not maintained anymore, and I only could document it with old screenshots)
Evolution of the facets data feed along the development
Front-end development of the video items, with a focus on avoiding the sense of loading time for the users
