![]() That's why we built Programiz PRO, where you can keep track of your progress, read interactive lessons with an inbuilt compiler, test your knowledge with quizzes, take challenges to gain confidence and get a certificate at the end. Our website was useful to such programmers but we were never able to really know how people were using it or where they were getting stuck in their programming journey. ![]() ![]() User feedback for results of our model.Anyone who's done any programming knows the initial dopamine hit when you print "Hello World".Īt Programiz, we found that the feeling quickly goes away once beginners start struggling with more intermediate and advanced programs. Article Summariser using Extractive Summarisationĥ. Suggested Articles using Keyword ExtractionĤ. Displaying Media Bias and Objectivity for a given siteģ. Machine Learning Predictor for News ReliabilityĢ. Keeping the point of access in mind, we built this extension to work with links shared on social media and the internet in general, as well as news sites themselves.ġ. The perils of fake news and disinformation are not just limited to the dominance of false narratives on the internet, but have far-reaching implications when these narratives are translated into our tangible world- for example vaccine disinformation during the COVID years has had damaging effects on immunisation drives worldwide.Īsatya, essentially is a suite of tools developed to tackle fake news and promote better access to trustable resources. With a growing amount of information present on the web, fake news and disinformation are huge areas of concern. The hackathon saw a participation of 3000 students from over 36 countries. I am going to be presenting my paper virtually today (14/07/22) at 8:30-10:00 AM PDT (9:00-10:30 PM IST) in the SemEval poster room on GatherTown! If you're attending the conference, I would love to discuss my paper with you and receive feedback.Įlated to announce that Asatya, a browser extension developed by Arsh Kohli, Aaryak Garg and I has been awarded the first position at the #TechForGood Hackathon organised by Enactus India along with Open Weaver. When I found that the performance of the basic BERT model I used was not satisfactory, I started focussing on methods to boost Pre Trained Language Models, and this paper represents some of those efforts. My learning curve with this paper went from simple "Bag-of-Words" models to RNNs and DNNs, to finally the modern state-of-the-art transformer-based models such as BERT. When I started working on this task, I was a beginner in the field of NLP. I participated in Task 4 of SemEval 2022 which dealt with the identification of Patronising and Condescending Language. I'm proud to announce that my first paper, Boosting Pre-trained Language Models with Task-Specific Metadata and Cost Sensitive Learning has been published at SemEval (International Workshop on Semantic Evaluation) 2022, co-located with #NAACL2022. Feature extraction and selection using the Twitter API, for a given tweet and the user tweeting it. Data augmentation using translation of Dutch, Bulgarian datasets.ģ. Developing a multimodal framework which involves pre-trained BERT for tweet text along with multilayer perceptrons (MLPs) for numerical and categorical features.Ģ. After countless literature review meets and brainstorming sessions within our team, we came up with an intuitive approach to the task, all while writing our fourth semester end-semester exams! Some key contributions of the paper are as follows:ġ. We ranked at #9, #2, #2 on the three tasks respectively. ![]() Harmfulness: This task aims to predict, given a tweet related to a news item, determine if the tweet is potentially harmful to the society. Verifiability: Given a tweet, this task requires us to predict whether it contains a verifiable factual claimģ. Check-worthiness: The goal of this task is to predict whether a given tweet is worth fact-checking.Ģ. The paper explains our team, Asatya’s approach on a task related to identifying the following claims in tweets:ġ. I am proud to announce that “Multimodal BERT for Identifying Claims in Tweets”, a paper I worked on with Prajeet Katari and Saumay Dudeja has been published in the working notes of CLEF 2022 (Conference and Labs of the Evaluation Forum), being held at Bologna, Italy from 5th-8th September, 2022.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |