Deep Neural Ranking Models for Argument Retrieval (original) (raw)

Conversational argument retrieval is the problem of ranking argumentative texts in a collection of focused arguments, ordered based on their quality. This study investigates eight different deep neural ranking models proposed in the literature with respect to their suitability to this task. In order to incorporate the insights from multiple models into an argument ranking, we further investigate a simple linear aggregation strategy. The Touché @ CLEF shared task on Conversational Argument Retrieval (Task 1) targets a retrieval scenario in a focused argument collection composed of roughly 387740 arguments to support argumentative conversation. We approach the problem of argument retrieval as finding relevant premises to the input query. Considering the lack of annotated data for the purpose of ranking, we take a distant supervision approach. Based on this approach, the conclusion (claim) of the arguments play the role of queries and the corresponding premises of the argument are the ...