Large Language Models (LLMs) have demonstrated significant potential in revolutionizing clinical applications. In this study, we investigate the efficacy of four techniques in adapting LLMs for clinical use-cases: continuous pretraining, instruct fine-tuning, NEFTune, and prompt engineering. We employ these methods on Mistral 7B and Mixtral 8x7B models, leveraging a large-scale clinical pretraining dataset of 50 billion tokens and an instruct fine-tuning dataset of 500 million tokens. Our evaluation across various clinical tasks reveals nuanced insights. While continuous pretraining beyond 250 billion tokens yields marginal improvements, instruct fine-tuning emerges as a more influential factor. Notably, NEFTune, designed primarily to enhance generation quality, surprisingly demonstrates additional gains on our benchmark. These findings underscore the importance of tailoring fine-tuning strategies and exploring innovative techniques to optimize LLM performance in the clinical domain.

This paper describes our approach for SemEval-2024 Task 9: BRAINTEASER: A Novel Task Defying Common Sense. The BRAINTEASER task comprises multiple-choice Question Answering designed to evaluate the models’ lateral thinking capabilities. It consists of Sentence Puzzle and Word Puzzle subtasks that require models to defy default commonsense associations and exhibit unconventional thinking. We propose a unique strategy to improve the performance of pre-trained language models, notably the Gemini 1.0 Pro Model, in both subtasks. We employ static and dynamic few-shot prompting techniques and introduce a model-generated reasoning strategy that utilizes the LLM’s reasoning capabilities to improve performance. Our approach demonstrated significant improvements, showing that it performed better than the baseline models by a considerable margin but fell short of performing as well as the human annotators, thus highlighting the efficacy of the proposed strategies.

This paper describes our systems for CLPsych?s 2022 Shared Task. Subtask A involves capturing moments of change in an individual?s mood over time, while Subtask B asked us to identify the suicidality risk of a user. We explore multiple machine learning and deep learning methods for the same, taking real-life applicability into account while considering the design of the architecture. Our team achieved top results in different categories for both subtasks. Task A was evaluated on a post-level (using macro averaged F1) and on a window-based timeline level (using macro-averaged precision and recall). We scored a post-level F1 of 0.520 and ranked second with a timeline-level recall of 0.646. Task B was a user-level task where we also came in second with a micro F1 of 0.520 and scored third place on the leaderboard with a macro F1 of 0.380.

The number of depression and suicide risk cases on social media platforms is ever-increasing, and the lack of depression detection mechanisms on these platforms is becoming increasingly apparent. A majority of work in this area has focused on leveraging linguistic features while dealing with small-scale datasets. However, one faces many obstacles when factoring into account the vastness and inherent imbalance of social media content. In this paper, we aim to optimize the performance of user-level depression classification to lessen the burden on computational resources. The resulting system executes in a quicker, more efficient manner, in turn making it suitable for deployment. To simulate a platform agnostic framework, we simultaneously replicate the size and composition of social media to identify victims of depression. We systematically design a solution that categorizes post embeddings, obtained by fine-tuning transformer models such as RoBERTa, and derives user-level representations using hierarchical attention networks. We also introduce a novel mental health dataset to enhance the performance of depression categorization. We leverage accounts of depression taken from this dataset to infuse domain-specific elements into our framework. Our proposed methods outperform numerous baselines across standard metrics for the task of depression detection in text.

This paper provides a comparison of different deep learning methods for identifying misogynous memes for SemEval-2022 Task 5: Multimedia Automatic Misogyny Identification. In this task, we experiment with architectures in the identification of misogynous content in memes by making use of text and image-based information. The different deep learning methods compared in this paper are: (i) unimodal image or text models (ii) fusion of unimodal models (iii) multimodal transformers models and (iv) transformers further pretrained on a multimodal task. From our experiments, we found pretrained multimodal transformer architectures to strongly outperform the models involving the fusion of representation from both the modalities.

This paper describes our approach (IIITH) for SemEval-2021 Task 5: HaHackathon: Detecting and Rating Humor and Offense. Our results focus on two major objectives: (i) Effect of task adaptive pretraining on the performance of transformer based models (ii) How does lexical and hurtlex features help in quantifying humour and offense. In this paper, we provide a detailed description of our approach along with comparisions mentioned above.

Code-mixed texts are widespread nowadays due to the advent of social media. Since these texts combine two languages to formulate a sentence, it gives rise to various research problems related to Natural Language Processing. In this paper, we try to excavate one such problem, namely, Parts of Speech tagging of code-mixed texts. We have built a system that can POS tag English-Bengali code-mixed data where the Bengali words were written in Roman script. Our approach initially involves the collection and cleaning of English-Bengali code-mixed tweets. These tweets were used as a development dataset for building our system. The proposed system is a modular approach that starts by tagging individual tokens with their respective languages and then passes them to different POS taggers, designed for different languages (English and Bengali, in our case). Tags given by the two systems are later joined together and the final result is then mapped to a universal POS tag set. Our system was checked using 100 manually POS tagged code-mixed sentences and it returned an accuracy of 75.29%.