Context-Independent Claim Detection for Argument Mining

Authors: Marco Lippi, Paolo Torroni

IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 5 we present experiments and results. We built an SVM classifier that exploits a kernel on the constituency parse trees obtained using the Stanford Core NLP 3.5.0 software4. Table 1: Results obtained on the IBM Wikipedia corpus.
Researcher Affiliation Academia Marco Lippi and Paolo Torroni DISI Universit a degli Studi di Bologna {marco.lippi3,p.torroni}@unibo.it
Pseudocode No The paper describes its methodology in prose and through figures, but does not include structured pseudocode or algorithm blocks.
Open Source Code No Dataset and results are available on the following website: http://lia.disi.unibo.it/ ml/argumentationmining/ijcai2015.html. (The website content states: 'The SVM classifier which implements the Tree Kernel is available upon request (the Partial Tree Kernel is part of the SVM-Light-TK software distribution, available here).')
Open Datasets Yes A smaller publicly available data set consists in a collection of 90 persuasive essays [Stab and Gurevych, 2014a] on a variety of different topics, where a team of annotators provided information regarding claims, major claims, premises (a synonym for evidence) and attack/support relations between such components.
Dataset Splits No The paper describes a 'leave-one-topic-out procedure' and '10-fold cross validation' for evaluation, which implies training and testing, but does not explicitly provide percentages or counts for distinct training, validation, and test dataset splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies Yes We built an SVM classifier that exploits a kernel on the constituency parse trees obtained using the Stanford Core NLP 3.5.0 software4.
Experiment Setup Yes To summarize, the methodology we propose consists of the following claim detection pipeline: 1. the given text document is split into sentences using a tokenizer that detects sentence boundaries; 2. each sentence is parsed, to obtain a constituency tree; 3. sentences not containing a verb (VP tag) are discarded; 4. in order to improve generalization, words at leaves are substituted with their stemmed versions; 5. an SVM classifies each sentence as possibly containing a claim or not.