MaNLP@SMM4H22: BERT for Classification of Twitter Posts

TL;DR

This paper presents a BERT-based binary classifier for identifying tweets reporting exact age, achieving around 80-81% F1 score in the SMM4H shared task.

Contribution

It introduces a straightforward BERT-based approach with different preprocessing strategies for classifying age-related tweets.

Findings

F1 scores of 0.80 and 0.81 achieved

Effective binary classification of age-related tweets

Simple preprocessing variations impact performance

Abstract

The reported work is our straightforward approach for the shared task Classification of tweets self-reporting age organized by the Social Media Mining for Health Applications (SMM4H) workshop. This literature describes the approach that was used to build a binary classification system, that classifies the tweets related to birthday posts into two classes namely, exact age(positive class) and non-exact age(negative class). We made two submissions with variations in the preprocessing of text which yielded F1 scores of 0.80 and 0.81 when evaluated by the organizers.