REMINDS project provides an automatic learning system on relevance detection
The REMINDS project (Relevance Mining and Detection System) aims to make an automatic learning system with web interface, capable of detecting potentially relevant content on social networks.
07th June 2018
This work was developed internationally through the collaboration of Portugal with the UT Austin programme and had inputs from INESC TEC’s Centre for Research in Advanced Computing Systems (CRACS), Laboratory of Artificial Intelligence and Decision Support (LIAAD) and Centre of Information Systems and Computer Graphics (CSIG), in partnership with the Faculty of Sciences of the University of Porto, the Faculty of Arts of the University of Porto, the Polytechnic Institute of Porto and CISUC (Centre for Informatics and Systems) of the University of Coimbra. From the american side, there was a collaboration from Texas Advanced Computing Center (TACC), the School of Information, both from UT@Austin, and the IBM-NewYork.
Aimed especially towards journalists but also to the public in general, this system automatically predicts if a post on Facebook or Twitter is relevant or not, using more than 30 features like: the contents of the post and its statistics, the feeling that it originates through feedback, the feeling associated to the entities mentioned in it, the dynamics of the feedback that it receives and the type of external links that it uses. Besides all this, this new approach integrates the use of surrogate features for journalistic criteria for relevance detection which otherwise would be too hard to be computed.
During the active period of the project, several models of relevance were created. There is already a fully functional prototype that integrates one of these models and that is hosted on CRACS. Even though being finished in 2017, this project is still generating results since the team is currently working on making databases freely available and studying new ways to dynamically change the model used for the prototype, according to the characteristics of the post to be analysed.
When looking to improve these filters in information, other models were also created. Some examples are: disambiguation of the mentioned entities, identification of matches on similar news, evolution over time of the feeling associated to entities and consensus of human opinion on relevance.
The team from INESC TEC is composed by Álvaro Figueira (CRACS; PI of the project), Nuno Guimarães (CRACS), Pedro Ribeiro (CRACS), Luís Torgo (LIAAD) and Paula Fortuna (CSIG).