A5046895021- Topic Modeling
- Natural Language Processing Techniques
- Advanced Text Analysis Techniques
- Machine Learning and Data Classification
- Text and Document Classification Technologies
- Explainable Artificial Intelligence (XAI)
- Sentiment Analysis and Opinion Mining
- Semantic Web and Ontologies
- Speech and dialogue systems
- Adversarial Robustness in Machine Learning
- Multimodal Machine Learning Applications
- Anomaly Detection Techniques and Applications
- Biomedical Text Mining and Ontologies
- Text Readability and Simplification
- Advanced Graph Neural Networks
- Data Quality and Management
- Bayesian Modeling and Causal Inference
- Software Engineering Research
- Hate Speech and Cyberbullying Detection
- Machine Learning in Healthcare
- Domain Adaptation and Few-Shot Learning
- Image Retrieval and Classification Techniques
- Misinformation and Its Impacts
- Scientific Computing and Data Management
- Generative Adversarial Networks and Image Synthesis
Heidelberg University
2017-2025
University of Vienna
2021-2024
Friedrich-Alexander-Universität Erlangen-Nürnberg
2024
Faculty (United Kingdom)
2023
Ludwig-Maximilians-Universität München
2016-2021
German Cancer Research Center
2017-2020
DKFZ-ZMBH Alliance
2017
University of Massachusetts Amherst
2014-2016
Amherst College
2016
Consol Energy (United States)
2016
Arvind Neelakantan, Benjamin Roth, Andrew McCallum. Proceedings of the 53rd Annual Meeting Association for Computational Linguistics and 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.
Patrick Verga, David Belanger, Emma Strubell, Benjamin Roth, Andrew McCallum. Proceedings of the 2016 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2016.
In this work, we propose a new model for aspect-based sentiment analysis. contrast to previous approaches, jointly the detection of aspects and classification their polarity in an end-to-end trainable neural network. We conduct experiments with different architectures word representations on recent GermEval 2017 dataset. were able show considerable performance gains by using joint modeling approach all settings compared pipeline approaches. The combination convolutional network fasttext...
The behavior of deep neural networks (DNNs) is hard to understand. This makes it necessary explore post hoc explanation methods. We conduct the first comprehensive evaluation methods for NLP. To this end, we design two novel paradigms that cover important classes NLP problems: small context and large problems. Both require no manual annotation are therefore broadly applicable. also introduce LIMSSE, an method inspired by LIME designed show empirically LRP DeepLIFT most effective recommend...
We survey recent approaches to noise reduction in distant supervision learning for relation extraction. group them according the principles they are based on: at-least-one constraints, topic-based models, or pattern correlations. Besides describing them, we illustrate fundamental differences and attempt give an outlook potentially fruitful further research. In addition, identify related work sentiment analysis which could profit from reduction.
Calibration, the alignment between model confidence and prediction accuracy, is critical for reliable deployment of large language models (LLMs). Existing works neglect to measure generalization their methods other prompt styles different sizes LLMs. To address this, we define a controlled experimental setting covering 12 LLMs four styles. We additionally investigate if incorporating response agreement multiple an appropriate loss function can improve calibration performance. Concretely,...
Abstract Precision cancer medicine aims to improve patient outcomes by providing individually tailored recommendations for clinical management based on the evaluation of biological disease profiles in multidisciplinary molecular tumor boards (MTBs). The quality MTB decisions depends comprehensive, reliable, and reproducible interpretation increasingly complex data. We developed implemented, as part a multicenter precision oncology program, Knowledge Connector (KC), decision support system...
We address relation classification in the context of slot filling, task finding and evaluating fillers like "Steve Jobs" for X "X founded Apple".We propose a convolutional neural network which splits input sentence into three parts according to arguments compare it state-ofthe-art traditional approaches classification.Finally, we combine different methods show that combination is better than individual approaches.We also analyze effect genre differences on performance.
Distant supervision is a scheme to generate noisy training data for relation extraction by aligning entities of knowledge base with text. In this work we combine the output discriminative at-least-one learner that generative hierarchical topic model reduce noise in distant data. The combination significantly increases ranking quality extracted facts and achieves state-of-the-art performance an end-to-end setting. A simple linear interpolation scores performs better than parameter-free based...
Spoken Language Systems at Saarland University (LSV) participated this year with 5 runs the TAC KBP English slot filling track. Effective algorithms for all parts of pipeline, from document retrieval to relation prediction and response post-processing, are bundled in a modular end-to-end extraction system called RelationFactory. The main run solely focuses on shallow techniques achieved significant improvements over LSV's last year's system, while using same training data patterns....
Interpretability of machine learning (ML) models becomes more relevant with their increasing adoption. In this work, we address the interpretability ML based question answering (QA) on a combination knowledge bases (KB) and text documents. We adapt post hoc explanation methods such as LIME input perturbation (IP) compare them self-explanatory attention mechanism model. For purpose, propose an automatic evaluation paradigm for in context QA. also conduct study human annotators to evaluate...
This paper describes how to apply self-attention with relative positional encodings the task of relation extraction. We propose use encoder layer together an additional position-aware attention that takes into account positions query and object in sentence. The also uses a custom implementation which allow each word sentence take its left right context. evaluation model is done on TACRED dataset. proposed relies only (no recurrent or convolutional layers are used), while improving...
Input optimization methods, such as Google Deep Dream, create interpretable representations of neurons for computer vision DNNs. We propose and evaluate ways transferring this technology to NLP. Our results suggest that gradient ascent with a gumbel softmax layer produces n-gram outperform naive corpus search in terms target neuron activation. The highlight differences syntax awareness between the language visual models Imaginet architecture.
Benjamin Roth, Tassilo Barth, Grzegorz Chrupała, Martin Gropp, Dietrich Klakow. Proceedings of the Demonstrations at 14th Conference European Chapter Association for Computational Linguistics. 2014.
Pankaj Gupta, Benjamin Roth, Hinrich Schütze. Proceedings of the 2018 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.
Supervised relation extraction from text relies on annotated data. Distant supervision is a scheme to obtain noisy training data by using knowledge base of relational tuples as the ground truth and finding entity pair matches in corpus. We propose evaluate two feature-based models for increasing quality distant patterns.
In the weakly supervised learning paradigm, labeling functions automatically assign heuristic, often noisy, labels to data samples. this work, we provide a method for from weak by separating two types of complementary information associated with functions: related target label and specific one function only. Both are reflected different degrees all labeled instances. contrast previous works that aimed at correcting or removing wrongly instances, learn branched deep model uses as-is, but...
Image clustering divides a collection of images into meaningful groups, typically interpreted post-hoc via human-given annotations. Those are usually in the form text, begging question using text as an abstraction for image clustering. Current methods, however, neglect use generated textual descriptions. We, therefore, propose Text-Guided Clustering, i.e., generating captioning and visual question-answering (VQA) models subsequently text. Further, we introduce novel approach to inject task-...