Prof. Dr. Andreas Hotho

Head of Data Science Chair and Founding Spokesman of CAIDAS

Chair of Data Science (Informatik X)
University of Würzburg
Am Hubland
97074 Würzburg
Germany

Email: hotho[at]informatik.uni-wuerzburg.de
Phone:(+49 931) 31 - 88453
Mobile: (+49) 173 259 40 52

Office: Room B112 (Computer Science Building M2)
Office Hours: By appointment only

About me

I am a professor at the University of Würzburg and the head of the data science chair (former DMIR group) and the founding spokesman of the Center for Artificial Intelligence and Data Science. Prior, I was a senior researcher at the University of Kassel. I started my research at the AIFB Institute at the University of Karlsruhe where I was working on text mining, ontology learning and semantic web related topics. My previous work also involved working at the KDE group of the University of Kassel on topics like data mining, semantic web mining and social media analysis. For a couple of years I've been a member of the L3S Research Center located in Hannover.

Research interests

In general, my current research focus is on data science (formerly known as data mining), text mining and semantic web.

Meanwhile, for many years I have followed the idea to combine the processing of natural language with the explicitly represented knowledge known today as knowledge graphs. This naturally leads to research on a combination of Text Mining and NLP methods like representation learning, information extraction, metric learning and ontology learning with research on Semantic Web, or Web Science. To reach these goals, I use and adopt NLP, machine learning and data mining methods. Beside that, I also work on Sentiment Analysis, genre classification and quotation detection. I have applied these methods on historic literature, but also on Social Media data, most recently on chat messages from Twitch.tv.

Other areas I’m interested in and working on are ranking, recommendation and behavior analysis methods. Additionally, my research interests include Anomaly Detection and the analysis of Time Series mostly on the web but recently also on ERP and environmental data, for example modeling states of bee colonies by analyzing sensor data obtained from smart beehives. Since many of these problems can be approached by black box machine learning and deep learning methods, another research area of mine is on explainable AI, to gather insights and understand the models.

To demonstrate my results, my group is working on different application systems: BibSonomy, Everyaware and We4Bee.

Projects

Awards

Best ML Innovation Award: "Deep Learning for Climate Model Output Statistics", Michael Steininger, Daniel Abel, Katrin Ziegler, Anna Krause, Heiko Paeth, Andreas Hotho at Tackling Climate Change with Machine Learning Workshop at NeurIPS 2020 (link)
Best Student Paper Award: "Evaluating the multi-task learning approach for land use regression modelling of air pollution", Andrzej Dulny, Michael Steininger, Florian Lautenschlager, Anna Krause, Andreas Hotho at FAIML 2020
Best Paper Award: "Financial Fraud Detection with Improved Neural Arithmetic Logic Units" by Daniel Schlör, Markus Ring, Anna Krause, Andreas Hotho on the Fifth Workshop on MIning DAta for financial applicationS Co-Hosted by ECML- PKDD 2020
SWSA Ten-Year Award: "Semantic Grounding of Tag Relatedness in Social Bookmarking Systems", Ciro Cattuto, Dominik Benz, Andreas Hotho, Gerd Stumme at the International Semantic Web Conference 2018 (link )
Best Paper Award: "HypTrails: A Bayesian Approach for Comparing Hypotheses About Human Trails on the Web” Philipp Singer, Denis Helic, Andreas Hotho and Markus Strohmaier, at WWW Conference 2015 (link)
Honorable mention of the paper: “Semantic Grounding of Tag Relatedness in Social Bookmarking Systems” Ciro Cattuto, Dominik Benz, Andreas Hotho and Gerd Stumme at ISWC 2008 (link)
The 7 years most influential paper award: “Information Retrieval in Folksonomies: Search and Ranking”, Andreas Hotho, Robert Jäschke, Christoph Schmitz, Gerd Stumme at ESWC 2013 (link )

Activities

Founding spokesman of the Center for Artificial Intelligence and Data Science (CAIDAS)
Member of the collegial leadership of the Zentrum für Philologie und Digitalität (Kallimachos)
Spokesperson of the Fachgruppe Knowledge Discovery, Data Mining und Maschinelles Lernen at GI
BAFög coordinator for the computer science department at the University of Würzburg
Executive director of the Institute of Computer Science at the University of Würzburg (2015 - 2016)

Research Track Chair: International Semantic Web Conference 2021
PC Chair: European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases 2019
Editor in Chief: Journal of Web Semantics since 2018
Editor in Chief: Transactions on Graph Data and Knowledge since 2023
Editorial boards: Semantic Web Journal 2009 - 2018, Journal of Web Semantics (data mining area chair) 2013 -2018, Transaction on Internet Technology 2013 -2018
PC-Chair Hypertext 2013, Track Co-Chair ESWC 2013, Demo Co-Chair ECML PKDD 2013, Track Co-Chair Hypertext 2011, Local Co-Chair GI–Workshopwoche “Lernen – Lehren – Wissen – Adaptivität” 2010, Track Co-Chair Hypertext 2009, Workshops and Tutorial Chair KCap 2009, Local Co-Chair GI–Workshopwoche “Lernen – Lehren – Wissen – Adaptivität” 2003, PC Co-Chair for a variety of workshops, e.g. RSWeb at RecSys 2012-2015, MUSE at ECML PKDD 2010-2015 or a workshop series on semantic web mining at the ECML PKDD 2001-2005
Selected PC memberships for conferences: ACM SIGKDD (regularly), AAAI (regularly), WWW (regularly), ISWC (regularly), ESWC (regularly), ECML PKDD (regularly), WebSci 2014, WSDM 2013, CIKM 2011
Reviewer for journals, e.g., International Journal of Information Security (IJISS), Journal on Data Semantics (JoDS), ACM Transactions on the Web (TWEB), Machine Learning Journal, Data and Knowledge Engineering (DKE)
Reviewer for a variety of workshops
Reviewer of research grants for the DFG and the European Union

Selected Publications

ConvMOS: Climate Model Output Statistics with Deep Learning M. Steininger; D. Abel; K. Ziegler; A. Krause; H. Paeth; A. Hotho in Data Mining and Knowledge Discovery, (P. Cellier; K. Dembczynski; A. Zimmermann; E. Devijver, Eds.) (2022).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, particularly systematic and location-specific representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we assess the use of convolutional Deep Learning climate MOS approaches and present the ConvMOS architecture which is specifically designed based on the observation that there are systematic and location-specific errors in the precipitation estimates of climate models. We apply ConvMOS models to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters for reducing location-specific errors and global model parameters for reducing systematic errors is indeed beneficial for MOS performance. We find that ConvMOS models can reduce errors considerably and perform significantly better than three commonly used MOS approaches and plain ResNet and U-Net models in most cases. Our results show that non-linear MOS models underestimate the number of extreme precipitation events, which we alleviate by training models specialized towards extreme precipitation events with the imbalanced regression method DenseLoss. While we consider climate MOS, we argue that aspects of ConvMOS may also be beneficial in other domains with geospatial data, such as air pollution modeling or weather forecasts.

@article{steininger2022convmos, abstract = {Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, particularly systematic and location-specific representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we assess the use of convolutional Deep Learning climate MOS approaches and present the ConvMOS architecture which is specifically designed based on the observation that there are systematic and location-specific errors in the precipitation estimates of climate models. We apply ConvMOS models to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters for reducing location-specific errors and global model parameters for reducing systematic errors is indeed beneficial for MOS performance. We find that ConvMOS models can reduce errors considerably and perform significantly better than three commonly used MOS approaches and plain ResNet and U-Net models in most cases. Our results show that non-linear MOS models underestimate the number of extreme precipitation events, which we alleviate by training models specialized towards extreme precipitation events with the imbalanced regression method DenseLoss. While we consider climate MOS, we argue that aspects of ConvMOS may also be beneficial in other domains with geospatial data, such as air pollution modeling or weather forecasts.}, author = {Steininger, Michael and Abel, Daniel and Ziegler, Katrin and Krause, Anna and Paeth, Heiko and Hotho, Andreas}, editor = {Cellier, Peggy and Dembczynski, Krzysztof and Zimmermann, Albrecht and Devijver, Emilie}, journal = {Data Mining and Knowledge Discovery}, keywords = {statistics}, month = {October}, publisher = {Springer}, title = {ConvMOS: Climate Model Output Statistics with Deep Learning}, year = 2022 }

%0 Journal Article %1 steininger2022convmos %A Steininger, Michael %A Abel, Daniel %A Ziegler, Katrin %A Krause, Anna %A Paeth, Heiko %A Hotho, Andreas %D 2022 %E Cellier, Peggy %E Dembczynski, Krzysztof %E Zimmermann, Albrecht %E Devijver, Emilie %I Springer %J Data Mining and Knowledge Discovery %R 10.1007/s10618-022-00877-6 %T ConvMOS: Climate Model Output Statistics with Deep Learning %X Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, particularly systematic and location-specific representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we assess the use of convolutional Deep Learning climate MOS approaches and present the ConvMOS architecture which is specifically designed based on the observation that there are systematic and location-specific errors in the precipitation estimates of climate models. We apply ConvMOS models to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters for reducing location-specific errors and global model parameters for reducing systematic errors is indeed beneficial for MOS performance. We find that ConvMOS models can reduce errors considerably and perform significantly better than three commonly used MOS approaches and plain ResNet and U-Net models in most cases. Our results show that non-linear MOS models underestimate the number of extreme precipitation events, which we alleviate by training models specialized towards extreme precipitation events with the imbalanced regression method DenseLoss. While we consider climate MOS, we argue that aspects of ConvMOS may also be beneficial in other domains with geospatial data, such as air pollution modeling or weather forecasts. %@ 1573-756X
InDiReCT: Language-Guided Zero-Shot Deep Metric Learning for Images K. Kobs; M. Steininger; A. Hotho (2022).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
- [ Download ]
Common Deep Metric Learning (DML) datasets specify only one notion of similarity, e.g., two images in the Cars196 dataset are deemed similar if they show the same car model. We argue that depending on the application, users of image retrieval systems have different and changing similarity notions that should be incorporated as easily as possible. Therefore, we present Language-Guided Zero-Shot Deep Metric Learning (LanZ-DML) as a new DML setting in which users control the properties that should be important for image representations without training data by only using natural language. To this end, we propose InDiReCT (Image representations using Dimensionality Reduction on CLIP embedded Texts), a model for LanZ-DML on images that exclusively uses a few text prompts for training. InDiReCT utilizes CLIP as a fixed feature extractor for images and texts and transfers the variation in text prompt embeddings to the image embedding space. Extensive experiments on five datasets and overall thirteen similarity notions show that, despite not seeing any images during training, InDiReCT performs better than strong baselines and approaches the performance of fully-supervised models. An analysis reveals that InDiReCT learns to focus on regions of the image that correlate with the desired similarity notion, which makes it a fast to train and easy to use method to create custom embedding spaces only using natural language.

@misc{kobs2022indirect, abstract = {Common Deep Metric Learning (DML) datasets specify only one notion of similarity, e.g., two images in the Cars196 dataset are deemed similar if they show the same car model. We argue that depending on the application, users of image retrieval systems have different and changing similarity notions that should be incorporated as easily as possible. Therefore, we present Language-Guided Zero-Shot Deep Metric Learning (LanZ-DML) as a new DML setting in which users control the properties that should be important for image representations without training data by only using natural language. To this end, we propose InDiReCT (Image representations using Dimensionality Reduction on CLIP embedded Texts), a model for LanZ-DML on images that exclusively uses a few text prompts for training. InDiReCT utilizes CLIP as a fixed feature extractor for images and texts and transfers the variation in text prompt embeddings to the image embedding space. Extensive experiments on five datasets and overall thirteen similarity notions show that, despite not seeing any images during training, InDiReCT performs better than strong baselines and approaches the performance of fully-supervised models. An analysis reveals that InDiReCT learns to focus on regions of the image that correlate with the desired similarity notion, which makes it a fast to train and easy to use method to create custom embedding spaces only using natural language.}, author = {Kobs, Konstantin and Steininger, Michael and Hotho, Andreas}, keywords = {zero}, note = {cite arxiv:2211.12760Comment: Accepted to WACV 2023}, title = {InDiReCT: Language-Guided Zero-Shot Deep Metric Learning for Images}, year = 2022 }

%0 Generic %1 kobs2022indirect %A Kobs, Konstantin %A Steininger, Michael %A Hotho, Andreas %D 2022 %R 10.48550/arXiv.2211.12760 %T InDiReCT: Language-Guided Zero-Shot Deep Metric Learning for Images %U http://arxiv.org/abs/2211.12760 %X Common Deep Metric Learning (DML) datasets specify only one notion of similarity, e.g., two images in the Cars196 dataset are deemed similar if they show the same car model. We argue that depending on the application, users of image retrieval systems have different and changing similarity notions that should be incorporated as easily as possible. Therefore, we present Language-Guided Zero-Shot Deep Metric Learning (LanZ-DML) as a new DML setting in which users control the properties that should be important for image representations without training data by only using natural language. To this end, we propose InDiReCT (Image representations using Dimensionality Reduction on CLIP embedded Texts), a model for LanZ-DML on images that exclusively uses a few text prompts for training. InDiReCT utilizes CLIP as a fixed feature extractor for images and texts and transfers the variation in text prompt embeddings to the image embedding space. Extensive experiments on five datasets and overall thirteen similarity notions show that, despite not seeing any images during training, InDiReCT performs better than strong baselines and approaches the performance of fully-supervised models. An analysis reveals that InDiReCT learns to focus on regions of the image that correlate with the desired similarity notion, which makes it a fast to train and easy to use method to create custom embedding spaces only using natural language.
Density-based weighting for imbalanced regression M. Steininger; K. Kobs; P. Davidson; A. Krause; A. Hotho in Machine Learning, (A. Appice; S. Escalera; J. A. Gamez; H. Trautmann, Eds.) (2021).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
- [ Download ]
In many real world settings,imbalanced data impedes model performance of learning algorithms, like neural networks, mostly for rare cases. This is especially problematic for tasks focusing on these rare occurrences. For example, when estimating precipitation, extreme rainfall events are scarce but important considering their potential consequences. While there are numerous well studied solutions for classification settings, most of them cannot be applied to regression easily. Of the few solutions for regression tasks, barely any have explored cost-sensitive learning which is known to have advantages compared to sampling-based methods in classification tasks. In this work, we propose a sample weighting approach for imbalanced regression datasets called DenseWeight and a cost-sensitive learning approach for neural network regression with imbalanced data called DenseLoss based on our weighting scheme. DenseWeight weights data points according to their target value rarities through kernel density estimation (KDE). DenseLoss adjusts each data point’s influence on the loss according to DenseWeight, giving rare data points more influence on modeltraining compared to common data points. We show on multiple differently distributed datasets that DenseLoss significantly improves model performance for rare data points through its density-based weighting scheme. Additionally, we compare DenseLoss to the state-of-the-art method SMOGN, finding that our method mostly yields better performance. Our approach provides more control over model training as it enables us to actively decide on the trade-off between focusing on common or rare cases through a single hyperparameter, allowing the training of better models for rare data points.

@article{steininger2021densitybased, abstract = {In many real world settings,imbalanced data impedes model performance of learning algorithms, like neural networks, mostly for rare cases. This is especially problematic for tasks focusing on these rare occurrences. For example, when estimating precipitation, extreme rainfall events are scarce but important considering their potential consequences. While there are numerous well studied solutions for classification settings, most of them cannot be applied to regression easily. Of the few solutions for regression tasks, barely any have explored cost-sensitive learning which is known to have advantages compared to sampling-based methods in classification tasks. In this work, we propose a sample weighting approach for imbalanced regression datasets called DenseWeight and a cost-sensitive learning approach for neural network regression with imbalanced data called DenseLoss based on our weighting scheme. DenseWeight weights data points according to their target value rarities through kernel density estimation (KDE). DenseLoss adjusts each data point’s influence on the loss according to DenseWeight, giving rare data points more influence on modeltraining compared to common data points. We show on multiple differently distributed datasets that DenseLoss significantly improves model performance for rare data points through its density-based weighting scheme. Additionally, we compare DenseLoss to the state-of-the-art method SMOGN, finding that our method mostly yields better performance. Our approach provides more control over model training as it enables us to actively decide on the trade-off between focusing on common or rare cases through a single hyperparameter, allowing the training of better models for rare data points.}, author = {Steininger, Michael and Kobs, Konstantin and Davidson, Padraig and Krause, Anna and Hotho, Andreas}, editor = {Appice, Annalisa and Escalera, Sergio and Gamez, Jose A. and Trautmann, Heike}, journal = {Machine Learning}, keywords = {selected}, title = {Density-based weighting for imbalanced regression}, year = 2021 }

%0 Journal Article %1 steininger2021densitybased %A Steininger, Michael %A Kobs, Konstantin %A Davidson, Padraig %A Krause, Anna %A Hotho, Andreas %D 2021 %E Appice, Annalisa %E Escalera, Sergio %E Gamez, Jose A. %E Trautmann, Heike %J Machine Learning %R 10.1007/s10994-021-06023-5 %T Density-based weighting for imbalanced regression %U https://doi.org/10.1007/s10994-021-06023-5 %X In many real world settings,imbalanced data impedes model performance of learning algorithms, like neural networks, mostly for rare cases. This is especially problematic for tasks focusing on these rare occurrences. For example, when estimating precipitation, extreme rainfall events are scarce but important considering their potential consequences. While there are numerous well studied solutions for classification settings, most of them cannot be applied to regression easily. Of the few solutions for regression tasks, barely any have explored cost-sensitive learning which is known to have advantages compared to sampling-based methods in classification tasks. In this work, we propose a sample weighting approach for imbalanced regression datasets called DenseWeight and a cost-sensitive learning approach for neural network regression with imbalanced data called DenseLoss based on our weighting scheme. DenseWeight weights data points according to their target value rarities through kernel density estimation (KDE). DenseLoss adjusts each data point’s influence on the loss according to DenseWeight, giving rare data points more influence on modeltraining compared to common data points. We show on multiple differently distributed datasets that DenseLoss significantly improves model performance for rare data points through its density-based weighting scheme. Additionally, we compare DenseLoss to the state-of-the-art method SMOGN, finding that our method mostly yields better performance. Our approach provides more control over model training as it enables us to actively decide on the trade-off between focusing on common or rare cases through a single hyperparameter, allowing the training of better models for rare data points.
Detecting Scenes in Fiction: A new Segmentation Task A. Zehe; L. Konle; L. Dümpelmann; E. Gius; A. Hotho; F. Jannidis; L. Kaufmann; M. Krug; F. Puppe; N. Reiter; A. Schreiber; N. Wiedmer (2021).
- [ BibTeX ]
- [ EndNote ]
- [ Download ]
@inproceedings{zehe2021detecting, author = {Zehe, Albin and Konle, Leonard and Dümpelmann, Lea and Gius, Evelyn and Hotho, Andreas and Jannidis, Fotis and Kaufmann, Lucas and Krug, Markus and Puppe, Frank and Reiter, Nils and Schreiber, Annekea and Wiedmer, Nathalie}, booktitle = {Proceedings of the 16th Conference of the {E}uropean Chapter of the Association for Computational Linguistics: Volume 1, Long Papers}, keywords = {task}, publisher = {ACL}, title = {Detecting Scenes in Fiction: A new Segmentation Task}, year = 2021 }

%0 Conference Paper %1 zehe2021detecting %A Zehe, Albin %A Konle, Leonard %A Dümpelmann, Lea %A Gius, Evelyn %A Hotho, Andreas %A Jannidis, Fotis %A Kaufmann, Lucas %A Krug, Markus %A Puppe, Frank %A Reiter, Nils %A Schreiber, Annekea %A Wiedmer, Nathalie %B Proceedings of the 16th Conference of the {E}uropean Chapter of the Association for Computational Linguistics: Volume 1, Long Papers %D 2021 %I ACL %T Detecting Scenes in Fiction: A new Segmentation Task
LM4KG: Improving Common Sense Knowledge Graphs with Language Models J. Omeliyanenko; A. Zehe; L. Hettinger; A. Hotho J. Z. Pan, V. Tamma, C. d’Amato, K. Janowicz, B. Fu, A. Polleres, O. Seneviratne, L. Kagal (Eds.) (2020). 456–473.
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
- [ Download ]
Language Models (LMs) and Knowledge Graphs (KGs) are both active research areas in Machine Learning and Semantic Web. While LMs have brought great improvements for many downstream tasks on their own, they are often combined with KGs providing additionally aggregated, well structured knowledge. Usually, this is done by leveraging KGs to improve LMs. But what happens if we turn this around and use LMs to improve KGs?

@inproceedings{omeliyanenko2020lm4kg, abstract = {Language Models (LMs) and Knowledge Graphs (KGs) are both active research areas in Machine Learning and Semantic Web. While LMs have brought great improvements for many downstream tasks on their own, they are often combined with KGs providing additionally aggregated, well structured knowledge. Usually, this is done by leveraging KGs to improve LMs. But what happens if we turn this around and use LMs to improve KGs?}, address = {Cham}, author = {Omeliyanenko, Janna and Zehe, Albin and Hettinger, Lena and Hotho, Andreas}, booktitle = {The Semantic Web – ISWC 2020}, editor = {Pan, Jeff Z. and Tamma, Valentina and d'Amato, Claudia and Janowicz, Krzysztof and Fu, Bo and Polleres, Axel and Seneviratne, Oshani and Kagal, Lalana}, keywords = {semantic}, pages = {456–473}, publisher = {Springer International Publishing}, title = {LM4KG: Improving Common Sense Knowledge Graphs with Language Models}, year = 2020 }

%0 Conference Paper %1 omeliyanenko2020lm4kg %A Omeliyanenko, Janna %A Zehe, Albin %A Hettinger, Lena %A Hotho, Andreas %B The Semantic Web -- ISWC 2020 %C Cham %D 2020 %E Pan, Jeff Z. %E Tamma, Valentina %E d'Amato, Claudia %E Janowicz, Krzysztof %E Fu, Bo %E Polleres, Axel %E Seneviratne, Oshani %E Kagal, Lalana %I Springer International Publishing %P 456--473 %T LM4KG: Improving Common Sense Knowledge Graphs with Language Models %U https://www.informatik.uni-wuerzburg.de/datascience/news/single/news/our-paper-lm4kg-improving-common-sense-knowledge-graphs-with-language-models-has-been-presented-a/ %X Language Models (LMs) and Knowledge Graphs (KGs) are both active research areas in Machine Learning and Semantic Web. While LMs have brought great improvements for many downstream tasks on their own, they are often combined with KGs providing additionally aggregated, well structured knowledge. Usually, this is done by leveraging KGs to improve LMs. But what happens if we turn this around and use LMs to improve KGs? %@ 978-3-030-62419-4

List of all publications: http://dmir.org/staff/hotho/pubs

Google Scholar profile: https://scholar.google.de/citations?user=eWTzXFAAAAAJ

DBLP Profile: http://dblp.uni-trier.de/pers/hd/h/Hotho:Andreas