Multi-Target Prediction and Applications in the Publishing, Energy and Retail Industries

Abstract

This talk is structured in two parts. In the first one, I will very briefly introduce the area of multi-target prediction, highlight its challenges and summarize some of our recent work on exploiting dependencies among multiple targets. In particular, I will briefly touch upon our work on: (i) discovering deterministic positive entailment and mutual exclusion relationships among multiple labels, representing these relationships as a Bayesian network and exploiting them for post-processing the output of multi-label models through probabilistic inference [1], and (ii) transferring successful multi-label classification approaches that use target variables as inputs (stacking, classifier chains) to the multi-target regression setting and extending them in order to deal with the discrepancy of the values of the target variables between training and prediction [2]. In the second part of the talk, I will discuss three ongoing data science collaborations with the private sector. In particular, I will talk about our work on: (i) semantic indexing of scientific publications [3], (ii) natural gas consumption forecasting, and (iii) business intelligence for the retail industry [4].

References

1. Papagiannopoulou, G. Tsoumakas, I. Tsamardinos (2015) Discovering and Exploiting Deterministic Label Relationships in Multi-Label Learning. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD ’15). ACM, New York, NY, USA, 915-924.
2. Spyromitros-Xioufis, G. Tsoumakas, W. Groves, I. Vlahavas (2016) Multi-Target Regression via Input Space Expansion: Treating Targets as Inputs. Machine Learning Journal.
3. Papanikolaou, D. Dimitriadis, G. Tsoumakas, M. Laliotis, N. Markantonatos, I. Vlahavas (2014) Ensemble Approaches for Large-Scale Multi-Label Classification and Question Answering in Biomedicine, Proceedings BioASQ 2014 Workshop, Sheffield, UK, 2014.
4. Fachantidis, A. Tsiaras, G. Tsoumakas, I. Vlahavas, (2016) Segmento: An R-based Visualization-rich System for Customer Segmentation and Targeting. In Proceedings of the 12^th Hellenic Conference on Artificial Intelligence.

Slides

About the lecturer

Grigorios Tsoumakas is an Assistant Professor of Machine Learning and Knowledge Discovery at the Department of Informatics of the Aristotle University of Thessaloniki (AUTH) in Greece. He received a degree in computer science from AUTH in 1999, an MSc in artificial intelligence from the University of Edinburgh, United Kingdom, in 2000 and a PhD in computer science from AUTH in 2005. His research expertise focuses on supervised learning techniques (ensemble methods, multi-target prediction) and text mining (semantic indexing, sentiment analysis, topic modelling). He has published more than 80 research papers and according to Google Scholar he has more than five thousand citations and an h-index of 30. Dr. Tsoumakas is a member of the Data Mining and Big Data Analytics Technical Committee of the Computational Intelligence Society of the IEEE and a member of the editorial board of the Data Mining and Knowledge Discovery journal. He is an advocate of applied research that matters and has worked as a machine learning and data mining developer, researcher and consultant in several national and private sector funded R&D projects

Website: http://intelligence.csd.auth.gr/people/tsoumakas