Abstract

The number and accessibility of published news items have grown recently. Publishers have developed recommender systems supporting users in finding relevant news. Traditional news recommender systems focus on collaborative filtering and content-based strategies. Unlike texts, multimedia content has received little attention. However, images and other multimedia elements affect how users perceive the news. In this work, we present a system that aggregates text-based, image-based, and user interests-based features to foster recommender systems for news. The system monitors a live stream of news and interactions with them. It applies text analysis and automatic image labeling methods for enriching the news stream. A web application visualizes the collected data and statistics. We show that image features are valuable for developing news recommender systems. The created feature-rich dataset constitutes the basis for developing innovative news recommendation approaches.

@inproceedings{LommatzschEtAl:FrameworkFAnalyzingNewsImagesAndBuildingMultimediaBasedRecommenders,
author="Andreas Lommatzsch and Benjamin Kille and Kevin Styp-Rekowski and Max Karl and Jan Pommering",
editor="L\"uke, Karl-Heinz"
and Eichler, Gerald
and Erfurth, Christian
and Fahrnberger, G{\"u}nter",
title="A Framework for Analyzing News Images and Building Multimedia-based Recommenders",
booktitle="Innovations for Community Services",
year="2019",
publisher="Springer International Publishing",
address="Cham",
pages="184-201",
isbn="978-3-030-22481-3",
doi="10.1007/978-3-030-22482-0",
}