Artificial Intelligence and Robotics Commons^™

Advancing Discourse Analysis In Multiparty Meetings: Comprehensive Classification Of Argument And Relation Types, Vishal Vaitla

Master's Projects

In multi-party meetings, accurately analyzing dialogue is crucial for enhancing communication effectiveness and decision-making. However, the informal and dynamic nature of these discussions presents complex challenges for computational analysis. Dialogues in such settings often include non-standard language, interruptions, and rapid topic changes, making it difficult to extract useful information with conventional text analysis tools. To tackle this challenge, two specific methods were developed:

Argument Classification: We use machine learning models like Gradient Boosting to identify and categorize the main points people make in their discussions. This helps us understand what each person is trying to say, making it easier to …

Short, Full, Best: Analysis Of Different Conference Papers, Miguel Williams

Graduate Research Theses & Dissertations

Research, publish, repeat. This is the basic cycle of anyone in academia. Individuals in academia conduct research, write up your research into a paper or journal article, submit to a conference or journal, and repeat the process. If you're skilled you may even obtain the coveted best paper award from the conference. In this research, I compare full papers to short papers and full papers to best papers. I start by fine-tuning three transformer models for classification capabilities. After I calculate lexical diversity and readability metrics of the papers, I use the features to train three traditional machine learning models. …

Mitigating Safety Issues In Pre-Trained Language Models: A Model-Centric Approach Leveraging Interpretation Methods, Weicheng Ma

Dartmouth College Ph.D Dissertations

Pre-trained language models (PLMs), like GPT-4, which powers ChatGPT, face various safety issues, including biased responses and a lack of alignment with users' backgrounds and expectations. These problems threaten their sociability and public application. Present strategies for addressing these safety concerns primarily involve data-driven approaches, requiring extensive human effort in data annotation and substantial training resources. Research indicates that the nature of these safety issues evolves over time, necessitating continual updates to data and model re-training—an approach that is both resource-intensive and time-consuming. This thesis introduces a novel, model-centric strategy for understanding and mitigating the safety issues of PLMs by …

Outsourcing Voting To Ai: Can Chatgpt Advise Index Funds On Proxy Voting Decisions?, Chen Wang

Fordham Journal of Corporate & Financial Law

Released in November 2022, Chat Generative Pre-training Transformer (“ChatGPT”), has risen rapidly to prominence, and its versatile capabilities have already been shown in a variety of fields. Due to ChatGPT’s advanced features, such as extensive pre-training on diverse data, strong generalization ability, fine-tuning capabilities, and improved reasoning, the use of AI in the legal industry could experience a significant transformation. Since small passive funds with low-cost business models generally lack the financial resources to make informed proxy voting decisions that align with their shareholders’ interests, this Article considers the use of ChatGPT to assist small investment funds, particularly small passive …

Improving Semantic Document Classification Accuracy By Integrating Human-Crafted Knowledge, Zachary Weinfeld, Lubomir Stanchev

College of Engineering Summer Undergraduate Research Program

Document classification is a pivotal task in various domains, warranting the development of robust algorithms. Among these, the Bidirectional Encoder Representations from Transformers (BERT) algorithm, introduced by Google, has proven to perform well when fine-tuned for the task at hand. Leveraging transformer architecture, BERT demonstrates stellar language understanding capabilities. However, the integration of BERT with a range of techniques has shown potential for further enhancing classification accuracy. This work investigates several techniques that leverage semantic understanding to improve the performance of document classification models trained with BERT. Specifically, we explore three methods. First, we will balance corpuses afflicted by imbalanced …

Gpachov At Checkthat! 2023: A Diverse Multi-Approach Ensemble For Subjectivity Detection In News Articles, Georgi Pachov, Dimitar Dimitrov, Ivan Koychev, Preslav Nakov

Natural Language Processing Faculty Publications

The wide-spread use of social networks has given rise to subjective, misleading, and even false information on the Internet. Thus, subjectivity detection can play an important role in ensuring the objectiveness and the quality of a piece of information. This paper presents the solution built by the Gpachov team for the CLEF-2023 CheckThat! lab Task 2 on subjectivity detection. Three different research directions are explored. The first one is based on fine-tuning a sentence embeddings encoder model and dimensionality reduction. The second one explores a sample-efficient few-shot learning model. The third one evaluates fine-tuning a multilingual transformer on an altered …

Decoding The Underlying Meaning Of Multimodal Hateful Memes, Ming Shan Hee, Wen Haw Chong, Roy Ka-Wei Lee

Research Collection School Of Computing and Information Systems

Recent studies have proposed models that yielded promising performance for the hateful meme classification task. Nevertheless, these proposed models do not generate interpretable explanations that uncover the underlying meaning and support the classification output. A major reason for the lack of explainable hateful meme methods is the absence of a hateful meme dataset that contains ground truth explanations for benchmarking or training. Intuitively, having such explanations can educate and assist content moderators in interpreting and removing flagged hateful memes. This paper address this research gap by introducing Hateful meme with Reasons Dataset (HatReD), which is a new multimodal hateful meme …

Can You Answer This? - Exploring Zero-Shot Qa Generalization Capabilities In Large Language Models, Saptarshi Sengupta, Shreya Ghosh, Preslav Nakov, Prasenjit Mitra

Natural Language Processing Faculty Publications

The buzz around Transformer-based Language Models (TLMs) such as BERT, RoBERTa, etc. is well-founded owing to their impressive results on an array of tasks. However, when applied to areas needing specialized knowledge (closed-domain), such as medical, finance, etc. their performance takes drastic hits, sometimes more than their older recurrent/convolutional counterparts. In this paper, we explore zero-shot capabilities of large language models for extractive Question Answering. Our objective is to examine the performance change in the face of domain drift, i.e., when the target domain data is vastly different in semantic and statistical properties from the source domain, in an attempt …

Quantification Of Various Types Of Biases In Large Language Models, Sudhashree Sayenju

Doctor of Data Science and Analytics Dissertations

Natural Language Processing (NLP) systems are included everywhere on the internet from search engines, language translations to more advanced systems like voice assistant and customer service. Since humans are always on the receiving end of NLP technologies, it is very important to analyze whether or not the Large Language Models (LLMs) in use have bias and are therefore unfair. The majority of the research in NLP bias has focused on societal stereotype biases embedded in LLMs. However, our research focuses on all types of biases, namely model class level bias, stereotype bias and domain bias present in LLMs. Model class …

Chatgpt As Metamorphosis Designer For The Future Of Artificial Intelligence (Ai): A Conceptual Investigation, Amarjit Kumar Singh (Library Assistant), Dr. Pankaj Mathur (Deputy Librarian)

Library Philosophy and Practice (e-journal)

Abstract

Purpose: The purpose of this research paper is to explore ChatGPT’s potential as an innovative designer tool for the future development of artificial intelligence. Specifically, this conceptual investigation aims to analyze ChatGPT’s capabilities as a tool for designing and developing near about human intelligent systems for futuristic used and developed in the field of Artificial Intelligence (AI). Also with the helps of this paper, researchers are analyzed the strengths and weaknesses of ChatGPT as a tool, and identify possible areas for improvement in its development and implementation. This investigation focused on the various features and functions of ChatGPT that …

Practical Ai Value Alignment Using Stories, Md Sultan Al Nahian

Theses and Dissertations--Computer Science

As more machine learning agents interact with humans, it is increasingly a prospect that an agent trained to perform a task optimally - using only a measure of task performance as feedback--can violate societal norms for acceptable behavior or cause harm. Consequently, it becomes necessary to prioritize task performance and ensure that AI actions do not have detrimental effects. Value alignment is a property of intelligent agents, wherein they solely pursue goals and activities that are non-harmful and beneficial to humans. Current approaches to value alignment largely depend on imitation learning or learning from demonstration methods. However, the dynamic nature …

Wikipedia Web Table Interpretation, Keyword-Based Search, And Ranking, Kartikee Dabir

Master's Projects

Information retrieval and data interpretation on the web, for the purpose of gaining knowledgeable insights, has been a widely researched topic from the onset of the world wide web or what is today popularly known as the internet. Web tables are structured tabular data present amidst unstructured, heterogenous data on the web. This makes web tables a rich source of information for a variety of tasks like data analysis, data interpretation, and information retrieval pertaining to extracting knowledge from information present on the web. Wikipedia tables which are a subset of web tables hold a huge amount of useful data, …

Comparative Analysis Of Transformer-Based Models For Text-To-Speech Normalization, Pankti Dholakia

Master's Projects

Text-to-Speech (TTS) normalization is an essential component of natural language processing (NLP) that plays a crucial role in the production of natural-sounding synthesized speech. However, there are limitations to the TTS normalization procedure. Lengthy input sequences and variations in spoken language can present difficulties. The motivation behind this research is to address the challenges associated with TTS normalization by evaluating and comparing the performance of various models. The aim is to determine their effectiveness in handling language variations. The models include LSTM-GRU, Transformer, GCN-Transformer, GCNN-Transformer, Reformer, and a BERT language model that has been pre-trained. The research evaluates the performance …

Multi-Label Text Classification With Transfer Learning, Likhitha Yelamanchili

Master's Projects

Multi-label text categorization is a crucial task in Natural Language Processing, where each text instance can be simultaneously assigned to numerous labels. This project's goal is to assess how well several deep learning models perform on a real-world dataset for multi-label text classification. We employed data augmentation techniques like Synonym Substitution and Random Word Substitution to address the problem of data imbalance. We conducted experiments on a toxic comment classification dataset to evaluate the effectiveness of several deep learning models including Bi-LSTM, GRU, and Bi-GRU, as well as fine- tuned pre-trained BERT models. Many metrics, including log loss, recall@k, and …

Transfer Of Personality Through Text Style, Michael O'Mahony, Robert Ross

Other resources

The style of generated text is how something is said rather than what is said. We hypothesize that changing the style of generated text can change the perceived personality of the text generation agent. Dialogue systems that aim to imitate a human agent can appear to have a consistent personality through a consistent, controllable style of conversation. Some recent work on the style of generated text [1] performs impressively in the small number of domains selected for their experiments using transformer and LSTM-based models. Lin et al. [1] used weak supervised learning as their data set lacks parallel data. The …

Nusax: Multilingual Parallel Sentiment Dataset For 10 Indonesian Local Languages, Genta Indra Winata, Alham Fikri Aji, Samuel Cahyawijaya, Rahmad Mahendra, Fajri Koto, Ade Romadhony, Kemal Kurniawan, David Moeljadi, Radityo Eko Prasojo, Pascale Fung, Timothy Baldwin, Jey Han Lau

Natural Language Processing Faculty Publications

Natural language processing (NLP) has a significant impact on society via technologies such as machine translation and search engines. Despite its success, NLP technology is only widely available for high-resource languages such as English and Chinese, while it remains inaccessible to many languages due to the unavailability of data resources and benchmarks. In this work, we focus on developing resources for languages in Indonesia. Despite being the second most linguistically diverse country, most languages in Indonesia are categorized as endangered and some are even extinct. We develop the first-ever parallel resource for 10 low-resource languages in Indonesia. Our resource includes …

The Behaviors Of Bert Attention Heads In Stereotype Detection, Joseph H. Hajjar

Dartmouth College Master’s Theses

We are living in the age of information, where it has become increasingly easy to share ideas, news, and content which are seen by an increasingly large number of people. This increasing scope of the increasing amount of data that is being shared lends itself to the question: how can we determine whether what we are reading promotes a stereotype? Previous work has applied transformer based models in this domain yielding impressive performance, but few studies exist interpreting the nature of attention heads in this task. Our work explores the feature encoding and extraction behaviors of attention heads in transformer …

A Machine Learning And Deep Learning Framework For Binary, Ternary, And Multiclass Emotion Classification Of Covid-19 Vaccine-Related Tweets, Aditya Dubey

Honors Scholar Theses

My research mines public emotion toward the Covid-19 vaccine based on Twitter data collected over the past 6-12 months. This project is centered around building and developing machine learning and deep learning models to perform natural language processing of short-form text, which in our case tweets. These tweets are all vaccine-related tweets and the goal of the classification task is for our models to accurately classify a tweet into one of four emotion groups: Apprehension/Anticipation, Sadness/Anger/Frustration, Joy/Humor/Sarcasm, and Gratitude/Relief. Given this data and the goal of the paper, we aim to answer the following questions: (1) Can a framework be …

Using A Bert-Based Ensemble Network For Abusive Language Detection, Noah Ballinger

Computer Science and Computer Engineering Undergraduate Honors Theses

Over the past two decades, online discussion has skyrocketed in scope and scale. However, so has the amount of toxicity and offensive posts on social media and other discussion sites. Despite this rise in prevalence, the ability to automatically moderate online discussion platforms has seen minimal development. Recently, though, as the capabilities of artificial intelligence (AI) continue to improve, the potential of AI-based detection of harmful internet content has become a real possibility. In the past couple years, there has been a surge in performance on tasks in the field of natural language processing, mainly due to the development of …

A Study On Developing Novel Methods For Relation Extraction, Darshini Mahendran

Theses and Dissertations

Relation Extraction (RE) is a task of Natural Language Processing (NLP) to detect and classify the relations between two entities. Relation extraction in the biomedical and scientific literature domain is challenging as text can contain multiple pairs of entities in the same instance. During the course of this research, we developed an RE framework (RelEx), which consists of five main RE paradigms: rule-based, machine learning-based, Convolutional Neural Network (CNN)-based, Bidirectional Encoder Representations from Transformers (BERT)-based, and Graph Convolutional Networks (GCNs)-based approaches. RelEx's rule-based approach uses co-location information of the entities to determine whether a relation exists between a selected entity …

Man-In-The-Middle Attacks On Mqtt Based Iot Networks, Henry C. Wong

Masters Theses

“The use of Internet-of-Things (IoT) devices has increased a considerable amount in recent years due to decreasing cost and increasing availability of transistors, semiconductor, and other components. Examples can be found in daily life through smart cities, consumer security cameras, agriculture sensors, and more. However, Cyber Security in these IoT devices are often an afterthought making these devices susceptible to easy attacks. This can be due to multiple factors. An IoT device is often in a smaller form factor and must be affordable to buy in large quantities; as a result, IoT devices have less resources than a typical computer. …

Caption And Image Based Next-Word Auto-Completion, Meet Patel

Master's Projects

With the increasing number of options or choices in terms of entities like products, movies, songs, etc. which are now available to users, they try to save time by looking for an application or system that provides automatic recommendations. Recommender systems are automated computing processes that leverage concepts of Machine Learning, Data Mining and Artificial Intelligence towards generating product recommendations based on a user’s preferences. These systems have given a significant boost to businesses across multiple segments as a result of reduced human intervention. One similar aspect of this is content writing. It would save users a lot of time …

Enhancing Zero-Shot And Few-Shot Text Classification Using Pre-Trained Language Models, Yanan Chen

Theses and Dissertations (Comprehensive)

In recent years, the community of natural language processing (NLP) has seen amazing progress in the development of pre-trained language models (PLMs). The novel paradigm of PLMs does not require labeled data, allowing us to experiment with increased training scale through employing freely available colossal online self-training corpus to push the limits. Language models (LMs), such as GPT, BERT and T5, have achieved high performance on a wide range of NLP tasks. Meanwhile, research on zero-shot and few-shot text classification has received increasing attention. As labelling can be costly and time-consuming, how to perform data augmentation (DA) and enhance the …

The Detection Of Sexual Harassment And Chat Predators Using Artificial Neural Network, Noor Amer Hamzah, Ban N. Dhannoon

Karbala International Journal of Modern Science

The vast increase in using social media sites like Twitter and Facebook led to frequent sexual_harassment on the Internet, which is considered a major societal problem. This paper aims to detect sexual_harassment and cyber_predators in early phase. We used deeplearning like Bidirectionally-long-short-term memory. Word representations are carefully reviewed in text specific to mapping to real number vectors. The chat sexual predators Detection_approach with the proposed_model. The best results obtained by the performance measured with F0.5-score were the result is_0.927 with proposed_models. The accuracy measured is_97.27% in the proposed_model. The comments sexual_harassment Detection_approach the result is_0.925 F0.5-score, and accuracy measured is_99.12%.

Lexical Complexity Prediction With Assembly Models, Aadil Islam

Dartmouth College Undergraduate Theses

Tuning the complexity of one's writing is essential to presenting ideas in a logical, intuitive manner to audiences. This paper describes a system submitted by team BigGreen to LCP 2021 for predicting the lexical complexity of English words in a given context. We assemble a feature engineering-based model and a deep neural network model with an underlying Transformer architecture based on BERT. While BERT itself performs competitively, our feature engineering-based model helps in extreme cases, eg. separating instances of easy and neutral difficulty. Our handcrafted features comprise a breadth of lexical, semantic, syntactic, and novel phonetic measures. Visualizations of BERT …

Fine-Grained Detection Of Hate Speech Using Bertoxic, Yakoob Khan

Dartmouth College Undergraduate Theses

This thesis describes our approach towards the fine-grained detection of hate speech using deep learning. We leverage the transformer encoder architecture to propose BERToxic, a system that fine-tunes a pre-trained BERT model to locate toxic text spans in a given text and utilizes additional post-processing steps to refine the prediction boundaries. The post-processing steps involve (1) labeling character offsets between consecutive toxic tokens as toxic and (2) assigning a toxic label to words that have at least one token labeled as toxic. Through experiments, we show that these two post-processing steps improve the performance of our model by 4.16% on …

Automated Analysis Of Rfps Using Natural Language Processing (Nlp) For The Technology Domain, Sterling Beason, William Hinton, Yousri A. Salamah, Jordan Salsman

SMU Data Science Review

Much progress has been made in text analysis, specifically within the statistical domain of Term Frequency (TF) and Inverse Document Frequency (IDF). However, there is much room for improvement especially within the area of discovering Emerging Trends. Emerging Trend Detection Systems (ETDS) depend on ingesting a collection of textual data and TF/IDF to identify new or up-trending topics within the Corpus. However, the tremendous rate of change and the amount of digital information presents a challenge that makes it almost impossible for a human expert to spot emerging trends without relying on an automated ETD system. Since the U.S. Government …

Machine Learning Models For Deciphering Regulatory Mechanisms And Morphological Variations In Cancer, Saman Farahmand

Graduate Doctoral Dissertations

The exponential growth of multi-omics biological datasets is resulting in an emerging paradigm shift in fundamental biological research. In recent years, imaging and transcriptomics datasets are increasingly incorporated into biological studies, pushing biology further into the domain of data-intensive-sciences. New approaches and tools from statistics, computer science, and data engineering are profoundly influencing biological research. Harnessing this ever-growing deluge of multi-omics biological data requires the development of novel and creative computational approaches. In parallel, fundamental research in data sciences and Artificial Intelligence (AI) has advanced tremendously, allowing the scientific community to generate a massive amount of knowledge from data. Advances …

Exploring The Use Of Neural Transformers For Psycholinguistics, Antonio Laverghetta Jr.

USF Tampa Graduate Theses and Dissertations

Deep learning has the potential to help solve numerous problems in cognitive science andeducation, by providing us a way to model the cognitive profiles of individual people. If this were possible, it would allow us to design targeted tests and suggest specific remediation based on each individual’s needs. On the flip side, employing techniques from psychology can give us insight into the underlying skillsets neural networks have acquired during training, addressing the interpretability concern. This thesis explores these ideas in the context of transformer language models, which have achieved state-of-the-art results on virtually every natural language processing (NLP) task. First, …

Improving Space Efficiency Of Deep Neural Networks, Aliakbar Panahi

Theses and Dissertations

Language models employ a very large number of trainable parameters. Despite being highly overparameterized, these networks often achieve good out-of-sample test performance on the original task and easily fine-tune to related tasks. Recent observations involving, for example, intrinsic dimension of the objective landscape and the lottery ticket hypothesis, indicate that often training actively involves only a small fraction of the parameter space. Thus, a question remains how large a parameter space needs to be in the first place — the evidence from recent work on model compression, parameter sharing, factorized representations, and knowledge distillation increasingly shows that models can be …