Open Argument Mining

• Applicants: Professorin Dr. Iryna Gurevych; Professor Dr. Steffen Staab
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2018 to 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 413534432

• Final Report Year: 2024

Final Report Abstract

In this project, we deal with three main challenges in open argument mining: (C1) Dealing with arguments that introduce new aspects to the debate, (C2) Dealing with incomplete arguments at the surface, and (C3) Constructing an open knowledge graph for argument mining. To account for the first challenge, we set all experiments in this project on heterogeneous data from various topics and sources. First, we deal with open-domain argument search on two benchmark datasets with different foundation model structures. While the gap to human performance remains, we are able to significantly reduce it and also to show the generalizability of modern approaches to unseen topics. Then, we concentrated on the task of stance detection. On a collective benchmark, we discovered models trained on heterogeneous sources are in danger of accumulating bias from individual sources over time, such that an evaluation of a static test dataset alone is not enough. Finally, we adopted a Bert-based model for argument aspect detection. For the second challenge, we discovered that the argument linking was more efficient to deal with incomplete arguments. Among the approaches we tested for knowledge injection, unstructured knowledge seemed to help more, in particular when applied for argument (stance) detection. We also constructed the ASPECT corpus, which contains pairs of automatically classified sentential arguments for argument linking. As for the third challenge, we constructed a dynamic open knowledge graph to support the argument mining process and an entity linking dataset targeting unseen novel entities. We also proposed a novel temporal knowledge graph embedding method to incorporate relevant knowledge into downstream argument mining applications. To conclude, we have contributed many new algorithms and datasets to the various aspects of open argument mining tasks, such as open-domain argument search, stance detection, argument linking, and dynamic open knowledge graph. An important open question arising from our project pertains to the effective detection of evolving arguments on social media and the formulation of appropriate knowledge graph schemas to integrate these evolving arguments into the open knowledge graph.

Publications

• Classification and Clustering of Arguments with Contextualized Word Embeddings. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics.
  Reimers, Nils; Schiller, Benjamin; Beck, Tilman; Daxenberger, Johannes; Stab, Christian & Gurevych, Iryna
  
• Common Sense or World Knowledge? Investigating Adapter-Based Knowledge Injection into Pretrained Transformers. Proceedings of Deep Learning Inside Out (DeeLIO): The First Workshop on Knowledge Extraction and Integration for Deep Learning Architectures. Association for Computational Linguistics.
  Lauscher, Anne; Majewska, Olga; Ribeiro, Leonardo F. R.; Gurevych, Iryna; Rozanov, Nikolai & Glavaš, Goran
  
• AdapterFusion: Non-Destructive Task Composition for Transfer Learning. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, 487-503. Association for Computational Linguistics.
  Pfeiffer, Jonas; Kamath, Aishwarya; Rücklé, Andreas; Cho, Kyunghyun & Gurevych, Iryna
  
• Aspect-Controlled Neural Argument Generation. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 380-396. Association for Computational Linguistics.
  Schiller, Benjamin; Daxenberger, Johannes & Gurevych, Iryna
  
• Knowledge Graphs. ACM Computing Surveys, 54(4), 1-37.
  Hogan, Aidan; Blomqvist, Eva; Cochez, Michael; D.’amato, Claudia; De Melo, Gerard; Gutierrez, Claudio; Kirrane, Sabrina; Gayo, José Emilio Labra; Navigli, Roberto; Neumaier, Sebastian; Ngomo, Axel-Cyrille Ngonga; Polleres, Axel; Rashid, Sabbir M.; Rula, Anisa; Schmelzeisen, Lukas; Sequeda, Juan; Staab, Steffen & Zimmermann, Antoine
  
• Stance Detection Benchmark: How Robust is Your Stance Detection?. KI - Künstliche Intelligenz, 35(3-4), 329-341.
  Schiller, Benjamin; Daxenberger, Johannes & Gurevych, Iryna
  
• NILK: Entity Linking Dataset Targeting NIL-linking Cases. Proceedings of the 31st ACM International Conference on Information & Knowledge Management, 4069-4073. ACM.
  Iurshina, Anastasiia; Pan, Jiaxin; Boutalbi, Rafika & Staab, Steffen
  
• The challenges of temporal alignment on Twitter during crises. Findings of the Association for Computational Linguistics: EMNLP 2022, 2658-2672. Association for Computational Linguistics.
  Pramanick, Aniket; Beck, Tilman; Stowe, Kevin & Gurevych, Iryna
  
• Do Temporal Knowledge Graph Embedding Models Learn or Memorize Shortcuts? In Temporal Graph Learning Workshop@ NeurIPS
  an, J., Nayyeri, M., Li, Y. & Staab, S.
  
• HGE: Embedding Temporal Knowledge Graphs in a Product Space of Heterogeneous Geometric Subspaces. Proceedings of the AAAI Conference on Artificial Intelligence, 38(8), 8913-8920.
  Pan, Jiaxin; Nayyeri, Mojtaba; Li, Yinan & Staab, Steffen