NLP Conference Success for Danae
It is a busy time of year for most, not least our PhD students. We caught up with Danae, from our first cohort, to find out about her recent success and attendance at three different NLP conferences.
‘Attending academic conferences is an important way to build skills as a PhD student. You get the opportunity to ask questions, get new ideas, ask for practical advice and meet professionals in the field. Given the COVID-19 pandemic, this year most conferences took place online. I recently attended 3 virtual conferences: ACL, EMNLP, and AACL.’
The 58th Annual Meeting of the Association for Computational Linguistics (ACL) took place online from 5th to 10th July 2020. ACL is the leading conference in the field of computational linguistics, covering a wide variety of research areas dealing with computational approaches to natural language.
In this conference, I presented the paper Analyzing Political Parody in Social Media. This paper is co-authored with Antonis Maronikolakis, my supervisor Nikolaos Aletras (University of Sheffield), and with Daniel Preoţiuc-Pietro (Bloomberg). We present the first study of parody using methods from computational linguistics and machine learning. We introduce a freely available large-scale dataset containing a total of 131,666 English tweets from 184 real and corresponding parody accounts, and evaluate and analyze a range of neural models achieving high predictive accuracy.
Also, I attended a mentoring session led by Dr. Alona Fyshe from University of Alberta, where PhD students could ask general questions. A common concern among students is how to balance work and social life. Dr. Fyshe recommended the book Deep Work: Rules for Concentrated Success in a Distracted World, a guide to intense concentration in a distraction-free environment that leads to rapid, efficient learning and results. She reminded us to have fun, too!
The 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) took place online from 16th to 20th November 2020. The Keynote by Dr. Janet Pierrehumbert, from University of Oxford, discussed the importance of the effects of context when determining the efficiency of NLP systems, as well as the variability in language among individuals. She reminded us to be cautious of what we are evaluating. For example, when using annotators, we should take into account that annotation depends on the point of view and lived experience of the participants, thus, we shouldn’t treat annotators as having access to ground truth.
The 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (AACL) took place online from 4th to 7th December 2020. In this conference I presented the paper Point-of-Interest Type Inference from Social Media Text. This work is co-authored with Daniel Preoţiuc-Pietro (Bloomberg), and my supervisor Nikolaos Aletras (University of Sheffield). We present the first study on the relationship between language of a social media message, and the type information associated with the point of interest (POI) the message was sent from. We develop a large-scale dataset of tweets mapped to their POI category, and conduct an analysis to uncover characteristics specific to place type. Also, we train predictive models to infer the POI category using only the text of the tweet and the posting time. Inferring the place type from the text could help geographers and social scientists research mobility trends, and how people interact with places in real-time.