The Fact Extraction and VERification (FEVER) shared task was launched to support the development of systems able verify claims by extracting supporting or refuting facts from raw text. organizers provide a large-scale dataset for consecutive steps involved in claim verification, particular, document retrieval, fact extraction, classification. In this paper, we present our verification pipeline approach, which, according preliminary results, scored third task, out 23 competing systems. For...

We experiment with two recent contextualized word embedding methods (ELMo and BERT) in the context of open-domain argument search. For first time, we show how to leverage power embeddings classify cluster topic-dependent arguments, achieving impressive results on both tasks across multiple datasets. classification, improve state-of-the-art for UKP Sentential Argument Mining Corpus by 20.8 percentage points IBM Debater - Evidence Sentences dataset 7.4 points. understudied task clustering,...

Argument mining is a core technology for automating argument search in large document collections. Despite its usefulness this task, most current approaches are designed use only with specific text types and fall short when applied to heterogeneous texts. In paper, we propose new sentential annotation scheme that reliably applicable by crowd workers arbitrary Web We source annotations over 25,000 instances covering eight controversial topics. show integrating topic information into...

Christian Stab, Johannes Daxenberger, Chris Stahlhut, Tristan Miller, Benjamin Schiller, Christopher Tauchmann, Steffen Eger, Iryna Gurevych. Proceedings of the 2018 Conference North American Chapter Association for Computational Linguistics: Demonstrations. 2018.

Abstract Stance detection (StD) aims to detect an author’s stance towards a certain topic and has become key component in applications like fake news detection, claim validation, or argument search. However, while is easily detected by humans, machine learning (ML) models are clearly falling short of this task. Given the major differences dataset sizes framing StD (e.g. number classes inputs), ML trained on single usually generalize poorly other domains. Hence, we introduce benchmark that...

Benjamin Schiller, Johannes Daxenberger, Iryna Gurevych. Proceedings of the 2021 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2021.

Molecular communication presents a new approach for data transmission between miniaturised devices, especially in the context of medical applications. A link is established using molecules, or other particles nanoscale, to modulate information. Due lack changing physical parameters, information channel often cannot be modelled accurately. Deep Learning provides solution receive transmitted sequence without need an analytical description channel. We present proof-of-concept application...

Abstract The ArgumenText project creates argument mining technology for big and heterogeneous data aims to evaluate its use in real-world applications. mines clusters arguments from a variety of textual sources large range topics multiple languages. Its main strength is generalization very different including web crawls, news data, or customer reviews. We validated the with focus on supporting decisions innovation management as well feedback analysis. Along public search engine API, has...

Concise and reliable graph-theoretic analysis of complex networks today is a cumbersome task, consisting essentially the adaptation intricate libraries for each specific problem instance. The growing number metrics that have been proposed in last years, which mainly gain significance due to increasing computational capabilities at hand, led important new insights field. However, they solely implemented as single algorithms, specialized purpose calculating exactly targeted metric selected...

Self-certifying names provide the property that any entity in a distributed system can verify binding between corresponding public key and self-certifying name without relying on trusted third party. However, lack with real-world identity. In this demonstration, we present implementation of concrete mechanism for using Web-of-Trust conjunction to missing binding. Our prototype runs Android devices demonstrates decentralised message authentication scheme kind content-oriented architecture....

Routing in complex networks is increasingly optimized towards situation and properties of the underlying network. Quick hypothesis testing with respect to performance different strategies, however, posing be an unnecessarily complicated task. To this end we propose GTNA-2, enhanced second version Graph-Theoretic Network Analyzer. Based on broadly used GTNA, it allows both for efficient simple analysis a large set graph metrics, but additionally has been extended support rapid prototyping...

The purpose of this article is to understand how Triple Helix linkages foster study program innovation at the micro-level and entrepreneurial university shapes support structures processes meso-level. We draw on case cooperative programs from a German applied sciences. selected business administration nursing as two different disciplinary examples. Cooperative are delivered partly in industry illustrate hybridity that knowledge transfer university. Our draws semi-structured interviews with...

Content-Centric Networking (CCN) promises to yield large efficiency gains for Internet content distribution. Its autonomous cache management, however, raises doubts about achieving the intended goals optimally. A coordinated based on timely usage information, will help fully leverage efficiency. In this poster we introduce CoMon, a system architecture that implements Coordinated caching Monitoring of and its stability. CoMon aims at improving CCN with low monitoring communication overheads.

With the rise of online social networks and other highly dynamic system, need for analysis their structural properties has grown in last years. While re-computation graph-theoretic metrics is feasible investigating a small set static system snapshots, this approach unfit application systems where we aim at frequent property updates. Based on concept data streams, new algorithms have been developed that update computed based changes instead recomputing them regularly.While there exists...

Using Peer-to-Peer technology to deliver live video streams mobile devices is a promising approach. It allows service providers scale their distribution without increasing cost. As the resources are replicated at edge of network, can take advantage close-by peers in order get required data faster. This, however, challenging due highly dynamic nature participating devices. Hence, overlay needs adapt quickly changes available bandwidth as well location peers. Also, it should be resilient...

Most tasks in NLP require labeled data. Data labeling is often done on crowdsourcing platforms due to scalability reasons. However, publishing data public can only be if no privacy-relevant information included. Textual contains sensitive like person names or locations. In this work, we investigate how removing personally identifiable (PII) as well applying differential privacy (DP) rewriting enable text with used for crowdsourcing. We find that DP-rewriting before preserve while still...

Decision-making tasks usually follow five steps: identifying the problem, collecting data, extracting evidence, iden-tifying arguments, and making decision. This paper focuses on two steps of decision-making: evidence by building knowledge graphs (KGs) specialized topics sentences' arguments through sentence-level argument mining. We present a hybrid model that combines topic modeling using latent Dirichlet allocation (LDA) word embeddings to obtain external from structured unstructured...

Decision-making usually takes five steps: identifying the problem, collecting data, extracting evidence, pro and con arguments, making decisions. Focusing on this paper presents a hybrid model that combines latent Dirichlet allocation word embeddings to obtain external knowledge from structured unstructured data. We study task of sentence-level argument mining, as arguments mostly require some degree world be identified understood. Given topic sentence, goal is classify whether sentence...