E8-IJS@LT-EDI-ACL2022 - BERT, AutoML and Knowledge-graph backed Detection of Depression

Abstract

Depression is a mental illness that negatively affects a person's well-being and can, if left untreated, lead to serious consequences such as suicide. Therefore, it is important to recognize the signs of depression early. In the last decade, social media has become one of the most common places to express one's feelings. Hence, there is a possibility of text processing and applying machine learning techniques to detect possible signs of depression. In this paper, we present our approaches to solving the shared task titled Detecting Signs of Depression from Social Media Text. We explore three different approaches to solve the challenge: finetuning BERT model, leveraging AutoML for the construction of features and classifier selection and finally, we explore latent spaces derived from the combination of textual and knowledge-based representations. We ranked 9th out of 31 teams in the competition. Our best solution, based on knowledge graph and textual representations, was 4.9% behind the best model in terms of Macro F1, and only 1.9% behind in terms of Recall.

Multi-label topic classification model of COVID-19 literature

“…Micro F-score SINAI [2] 82.54% Team E8 [3] 84.30% pre-trained Bert 89.85% pre-trained+fine-tuned Bert 91.14%…”

“…Chizhikova et al use the TF-IDF representation of tokenized and stemmed text data to achieve a micro F-score of 82.54% on the test data sets [2]. Tavchioski used an automated Bag-Of-Tokens system to learn representations from token, sub-word, and sentencelevel features, achieving a micro F-score of 84.30% [3]. The goal of the shallow machine learning method described above is to design artificial features using domain expert knowledge.…”

SentiNeg

“…The method includes individual text and image classification along with AutoML-generated model to perform final sentiment classification. (Tavchioski et al, 2022) explored three different models to detect depression from social media text. They include BERT family models, AutoML approaches and knowledge-based representations based on knowledge graph concepts.…”