Relevance analysis of the RTBF's data : metadata describing the content of the Auvio platform

Details

Supervisors

Delvenne, Jean-Charles

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en science des données

Abstract

Nowadays, new technologies are expanding rapidly and it’s getting easier and easier to access a wide variety of content thanks to the internet. The "Radio-Television Belge de la Communaute Francaise" or ’RTBF’ is one of the main public-service broadcasting organization for the French-speaking community of Belgium. Like other companies in the same sector, the RTBF diffuses online video and audio content through its free Video-On-Demand platform called ’Auvio’. The content of this platform are referred by a meta-dataset. The different records of this dataset are provided by many platforms (internal or not to the RTBF). We are interested in analyzing the relevance of these records, understand why some of them are less relevant than others and from which providers the not relevant records come from. We perform a descriptive analysis to understand the composition of the dataset and establish that the text fields describe best the content of Auvio. Complete and specific titles and descriptions are essential to assure the relevance of a record. We use text mining techniques to quantify the relevance of these text fields. We show that relevant texts are specific to the content they refer. If the relevance is mainly determined by the text fields. We can also find other characteristics that affect the relevance. We build a classifier to determine how relevant the content is classified in categorical attributes. Using the different techniques to quantify the relevance of the records, we conclude that the relevance of the dataset is relatively good. If most of the provides furnish records of suitable relevance, some providers are particularly efficient in terms of providing relevant records. On the other hand, one big provider affects strongly negatively the global relevance of the dataset. The relevance analysis of the dataset is interested to know why some records lack relevance and to understand the reasons. After the analysis, our final objective is to provide to the RTBF a software using the different techniques we used to evaluate the records during our analysis. This software would compute the relevance of an eventual new record that is not added to the dataset yet. This could prevent an inappropriate record to be added to the dataset.