30th Symposium of Applied Computing, SAC 2015

Author: Andreas Lommatzsch

Last week I attended the 30th Symposium of Applied Computing (SAC 2015) in Salamanca, Spain. The conference was held in the Hospedería del Colegio Arzobispo Fonseca (Universidad de Salamanca), a historical edifice in Salamanca, Spain, founded in 1519 by Alonso de Fonseca.

The Symposium of Applied Computing is a big international conference with 37 technical tracks, covering almost all relevant aspects of applied computing. Overall there were 1218 paper submissions and an acceptance rate of 23.9% for this years SAC.

I presented the paper “Real-Time Recommendations for User-Item streams”. The paper describes a new approach for compute recommendations based on stream data under real-time constraints and is part of our work in the CrowdRec project. The evaluation is done based on the news recommendation scenario defined in the NewsREEL challenge.

There were two keynotes at the conference. On Tuesday, Professor Il-Yeol Song from College of Computing & Informatics (Drexel University, Philadelphia, PA, USA) presented a survey on Big Data technologies, pointing out that “Big data is dirty” and “diverse.” The second keynote was given by PD. Dr. Matthias Klusch from the German Research Center for Artificial Intelligence (DFKI), Saarbrücken, Germany. In his interesting talk he explained how intelligent agents and semantic technologies will be the enabler for smart future internet applications. As an intelligent recommender system the system SERUM was mentioned developed in a joint project from the Neofonie GmbH and CC IRML.

There were many high quality presentations and posters at the conference. I would like to mention here three of them.

• Predictability in Human-Agent Cooperation: Adapting to Humans’ Personalities: The paper analyzes the questions how artificial agents influence the constitution of teams of humans and robots. Based on a repeatedly played game, an agent model is developed able to learn human personality defined using the so called Five-Factor Model personality type model. The authors conclude that that only some, but not all characteristics of a personality can be learned accurately.
• On the Spectrum between Binary Relevance and Classifier Chains in Multi-Label Classification: This paper analyzes block-wise classifier chains in comparison with classical classifier chains. It is shown that the intelligent choice of the block size leads to a noticeable but modest improvement of the classification quality.
• Visual Detection of Singularities in Review Platforms: This paper analyzes anomalies in hotel rating extracted from Tripadvisor and Booking.com. They proposed an approach for detecting misleading reviews in a large stream of noising reviews.

And, as the last point, some impressions from the conference and Salamanca: