OVERVIEW

Argus: A Schema-Agnostic Argument Mining System with Autoregressive Language Models 

1. We develop a full argument-mining pipeline that incorporates transfer learning aligned with autoregressive language models to identify ADU boundaries and exploit categorical interrelations, forming an argumentative graph. 

2. We propose a novel framework based on LLMs that can be trained on divergent argument mining datasets, each with an arbitrary predefined schema.

3. We evaluate the capabilities of various argument mining baselines, including deep learning, transformer-based, and Language models, on multiple standard datasets, investigating aspects such as complexity, weights, and architectures.

CODE and data

The code and supplementary materials are available at the following repository:

https://github.com/taghizad3h/argus

Data Referecnes

> We evaluate our proposed method on two real-world datasets, Persuasive Essays [A] and ArgMicrotexts [B].

• [A] Christian Stab and Iryna Gurevych. Parsing Argumentation Structures in Persuasive Essays. Computing Linguistics

• [B] Andreas Peldszus and Manfred Stede. Joint prediction in mst-style discourse parsing for argumentation mining, EMNLP

Relation to previous TKDE, NCAA, and WWWJ articles

Our initial works include detecting the concepts from short texts using an external knowledge base (KB) [1] Other previous articles of ours [3][4] recommend temporal-textual embedding models to track dynamic perturbations in the short text contents. We recently published our paper in IEEE TKDE [5], which investigates temporal dynamics in the question-answering corpus, resembling the conversational AI sphere. Following our prior works on emotion understanding from brief contents [6] and in contrast to traditional and multi-modal context-oriented approaches that merely utilize elements such as dialogue acts and persona, we propose a contextualized sentiment identification approach via inferring not only sentiments but also emotion signifiers.

[1] Hua, W., Wang, Z., Wang, H., Zheng, K., and Zhou, X. (2016). Understand short texts by harvesting and analyzing semantic knowledge. IEEE transactions on Knowledge and data Engineering, 29(3), 499-512.

[2] Kamran, S., Hosseini, S., Esmailzadeh, S., Kangavari, M. R., and Hua, W. (2024). Cognition2Vocation: meta-learning via ConvNets and continuous transformers. Neural Computing and Applications, 36(21), 12935-12950.

[3] Kamran, S., Zall, R., Hosseini, S., Kangavari, M., Rahmani, S., & Hua, W. (2023). EmoDNN: understanding emotions from short texts through a deep neural network ensemble. Neural Computing and Applications, 35(18), 13565-13582.

[4] Zhu, T., Hua, W., Qu, J., Hosseini, S. and Zhou, X., 2023. Auto-regressive extractive summarization with replacement. World Wide Web, 26(4), pp.2003-2026.

CONTACT US

Ali Taghizadeh: a_taghizadeh@comp.iust.ac.ir

Dr. Saeid Hosseini:  saeid.hosseini@uq.net.au