Temporal Question Answering in News Article Collections
Adam Jatowt · Companion Proceedings of the Web Conference 2022 · 2022
The fields of automatic question answering and reading comprehension have been rapidly advancing recently [6]. Open-domain question answering, in particular, assumes answering arbitrary user questions from a large document collection. The existing QA approaches are however generally working on synchronic document collections such as Wikipedia, Web data or short-term news corpora. This talk is about our latest efforts in automatic question answering over temporal news collections which can contain millions of news articles published during several decades’ long time frames. Temporal aspects of both news articles and user questions form an additional challenge for this kind of a temporal question answering task. To correctly answer questions over such collections one usually needs first to find candidate documents that are likely to contain these answers. We will first discuss a re-ranking approach [3] for news articles which works by utilizing temporal information embedded in questions and in the underlying document collection, thus combining methods from Temporal Information Retrieval [1, 2] and Natural Language Processing. Next, we will discuss a dedicated solution for answering ”When” type questions which require finding occurrence dates of events described in input questions based on the underlying news archive [5]. Finally, we will introduce ArchivalQA [4] - a large-scale question answering dataset which has been automatically created from a two decades’ long news article collection, and which contains over 500k question-answer pairs. The dataset has been processed to remove temporally ambiguous questions for which more than one correct answer exist.