Pavithra Rajendran is studying for a PhD in the Department of Computer Science:
“My PhD project, which is entitled ‘Computational argumentation of online discussions’, aims to bridge the gap between theoretical argumentation models and computational approaches using natural language processing (NLP) techniques.
I am funded by the departmental GTA studentship and supervised by Dr Danushka Bollegala here in Liverpool and Simon Parsons from Kings College London.
The Internet has become an important platform for people to share and gather information. This vast amount of information is highly unstructured such as what you find in forums, blogs, user reviews and other online discussions.
Discussions on controversial issues within online debating systems are also becoming quite popular. Theoretical argumentation models can help in understanding the inferential arguments present in them. The challenge is to be able to automatically identify and analyse those texts that need to be tackled with the help of natural language processing techniques.
Natural language processing is a branch in artificial intelligence that deals with computational linguistics and analysis. This state of the art research has gone beyond just text extraction to analysing the sentiment or opinions present and also the semantic relationships between phrases or text.
Several theoretical analyses in the field of argumentation have been developed based on abstract models that need to be practically developed on real data. Argumentation can give us a deeper analysis of reasoning of the textual content that has yet to be achieved by the NLP and argumentation community.
My project will use available resources to develop argumentation models that can work on real data. This will give an insight into the drawbacks that are faced and how future new NLP state of the art methods can be developed to analyse real arguments.
Our analysis of opinions in online reviews has revealed two interesting results — both presented in the ACL argument mining workshop 2016 and the COMMA conference 2016. We have shown how argument structures can be identified in reviews based on the structure of the opinions and also on how such opinions can be aggregated based on their argumentative properties to automatically assess whether they are negative or positive reviews.
Our results have also shown the difficulties that we face with the current NLP techniques that fail to identify the argumentative properties behind these texts.
This is an interesting research area which is gaining a lot of attention from both the argumentation and NLP community.”