Repository with annotated research papers from all sorts of conferences. Papers with high utility and edge learnings are included here.
-
Contextualizing Hate Speech Classifiers with Post hoc Explanation : In Hate Speeches, there is no clear cut definition of what is considered offensive, mostly prejudiced and offensive words which are not used in context of hate are classified which results in high false positive. This technique trains BERT with SOC(Sampling and Occlusion) regularization so the additional weight provided to prejudiced or abusive words can be removed and the model can be forced to look around the "concerned word" so it learns that in the right context the "concerned word" becomes part of speech not without it.
-
Give Me Convenience and Give Her Death: Who Should Decide WhatUses of NLP are Appropriate, and on What Basis? : This paper argues that unethically or unverfied data oriented researches are continously occuring in Top Research Generals of Machine Learning Domain. This is concerning and the paper has illustrated it using a case study. Case Study : Previously a paper presented automation of legal decisions, he argued about the bias but in fact the data collected had no consent from the citizens and in fact had no updates on the over-rulings of the decisions. The data was never updated, which takes things south real quick. The data had some personal information which enabled some of the criminals/accused/falsely accused people to be traced in the sparsely populated areas in China.
-
Exploratory Analysis of Covid-19 Tweets usingTopic Modeling, UMAP, and DiGraphs: This paper is build on top of COVID-19 tweet data hydrated by Chen(which I possess). This paper explores Topic Modelling using LDA, UMAP(TSne killer) for dimensionality reducation and cluster plotting. It also implements Di-Graphs to see the information flow from original tweets to retweets. Code Information missing but still paper is implementable on our own.