Machine learning approaches to rank news feed updates on social media

Résumé

Today, social media such as Facebook and Twitter are used by hundreds of millions of users worldwide. Due to the large number of members and the large amount of data posted and shared (messages, articles, videos, music, photos, etc.), users find themselves overwhelmed by a large volume of updates in their news feed, also known as the news stream. The news feed is typically displayed in reverse chronological order from the most recent to the least recent update. In addition, several research works have shown that the majority of the updates are considered irrelevant. Therefore, large data volume and irrelevance make it difficult for users to quickly catch up and interact with updates that may be of interest to them. To this end, based on the prediction of a relevance score between a user and a new update unread in the news feed, research work has proposed approaches to rank and display news feed updates in descending relevance order. These approaches aim to provide recommendations and help users quickly find relevant updates. In this thesis work, we first carry out a state-of-the-art of the proposed approaches in the field of ranking news feed updates according to several criteria: the features that may influence the relevance of updates, the relevance prediction models, the training and evaluation of prediction models, the target social media platforms, etc. The goal is first to show the advantages of the proposed approaches, their limitations, and identify open research issues. Thereafter, we propose and implement several intelligent machine-learning-based models to address the limitations of the approaches proposed in the literature. The objective of the proposed approaches is to collect and preprocess real data, extract the features that may influence relevance, train personalized prediction models for each user, predict relevance scores of news feed updates, and finally conduct in-depth experiments to evaluate and validate the proposed models.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 4

Téléchargements : 0