Towards an argumentative content search engine using weak supervision

Proceedings of the 27th International Conference on Computational Linguistics, pages 2066–2081 Santa Fe, New Mexico, USA, August 20-26, 2018. 2066 Towards an argumentative content search engine using weak supervision Ran Levy,∗ Ben Bogin∗, Shai Gretz∗, Ranit Aharonov, Noam Slonim IBM Research {ranl,boginb,avishaig,ranita,noams}@il.ibm.com Abstract Searching for sentences containing claims in a large text corpus is a key component in devel- oping an argumentative content search engine. Previous works focused on detecting claims in a small set of documents or within documents enriched with argumentative content. However, pinpointing relevant claims in massive unstructured corpora, received little attention. A step in this direction was taken in (Levy et al., 2017), where the authors suggested using a weak signal to develop a relatively strict query for claim–sentence detection. Here, we leverage this work to define weak signals for training DNNs to obtain significantly greater performance. This ap- proach allows to relax the query and increase the potential coverage. Our results clearly indicate that the system is able to successfully generalize from the weak signal, outperforming previously reported results in terms of both precision and coverage. Finally, we adapt our system to solve a recent argument mining task of identifying argumentative sentences in Web texts retrieved from heterogeneous sources, and obtain F1 scores comparable to the supervised baseline. 1 Introduction The arguments raised during a decision making process, will often determine its outcome. A common component in all argument models (e.g., (Toulmin, 2003)) is the claim, i.e. the assertion the argument aims to prove. The problem of automatically detecting claims supporting or contesting a given contro- versial topic 1 (Levy et al., 2014) is considered a fundamental task in the emerging field of computational argumentation (Lippi and Torroni, 2016; Palau and Moens, 2009). We refer to their definition of a Topic and a Claim; Topic - a short phrase that frames the discussion and Context Dependent Claim - a gen- eral, concise statement that directly supports or contests the given Topic (we henceforth use the term claim instead of Context Dependent Claim). Previous works have focused on detecting claims within a small set of documents related to the topic (Levy et al., 2014), or within documents enriched with argumentative content (Stab and Gurevych, 2014). However, pinpointing relevant claims within massive unstructured corpora, received relatively little at- tention. While this problem is obviously more challenging, its potential value is also much higher. For a widely discussed topic, one should expect many relevant claims to be mentioned across a widespread set of articles in the given corpus. The remaining issue is to develop a technology to swiftly detect these claims and present the results to potential users, similarly to search engines that retrieve information in response to a query. A step in this direction was taken in (Levy et al., 2017). They suggested a relatively strict sentence– level query (strict in the sense that it considerably limits the set of potential answers hence reduces the coverage). Their query combines three query parts that must appear in order, …

Content truncated to 500 words