A Framework for Argument Retrieval

Abstract

Computational argumentation has recently become a fast growing field of research. An argument consists of a claim, such as "We should abandon fossil fuels", which is supported or attacked by at least one premise, for example "Burning fossil fuels is one cause for global warming". From an information retrieval perspective, an interesting task within this setting is finding the best supporting and attacking premises for a given query claim from a large corpus of arguments. Since the same logical premise can be formulated differently, the system needs to avoid retrieving duplicate results and thus needs to use some form of clustering. In this paper we propose a principled probabilistic ranking framework for premises based on the idea of tf-idf that, given a query claim, first identifies highly similar claims in the corpus, and then clusters and ranks their premises, taking clusters of claims as well as the stances of query and premises into account. We compare our approach to a baseline system that uses BM25F which we outperform even with a primitive implementation of our framework utilising BERT.

Quality-Aware Ranking of Arguments

“…Probabilistic Framework for Argument Retrieval. In this paper we extend the probabilistic framework of our prior work [10], which clusters claims and premises according to their meaning and then ranks premises similar to the principle of to a query. A premise is ranked high if similar premises frequently support (or attack) similar claims, and similar premises support as few other claim clusters as possible.…”

“…While the approach of Wachsmuth et al [37] has the advantage of quality dimensional diversity, so that arguments can be ranked according to speci c dimensions, because di erent dimensions have di erent e ects on people, the approach of Habernal and Gurevych [15] has the advantage that a ranking can easily be extended when a new argument arrives, because a new argument can always have a better rating than the current best one which might already have the highest possible score. We extend the framework of our prior work [10] by quality dimensions, but we do not want to assign nal scores. Therefore we use the dataset of Wachsmuth et al [37], but do not learn any scores, but transform the dataset to (premise 1 , premise 2 ) pairs and then learn which argument is better with regard to a speci c dimension similar to the approach of Habernal and Gurevych [15].…”

“…In this section we summarize the main components of the probabilistic ranking framework of our prior work [10] on which our method builds. This framework nds good premises for a query from a large corpus of already mined arguments using a two-step retrieval approach.…”

“…To the best of our knowledge, our prior work [10] introduced the rst probabilistic framework for retrieving premises from already mined argument graphs, given a query claim. This framework operates on clusters of claims and clusters of premises where clusters are formed based on similarity.…”

“…Next, we discuss related work in Section 2. Then we introduce the foundations of the probabilistic framework of our prior work [10] in Section 3 and extend it to include argument quality in Section 4. Afterwards, we evaluate our approach in Section 5.Section 6 concludes the paper and discusses ideas for future work.…”