<p dir="ltr">Sexual harassment is regarded as a serious issue in society, with a particularly negative impact on young children and adolescents. Online sexual harassment has recently gained prominence as a significant number of communications have taken place online. Online sexual harassment can happen anywhere in the world because of the global nature of the internet, which transcends geographical barriers and allows people to communicate electronically. Online sexual harassment can occur in a wide variety of environments such as through work mail or chat apps in the workplace, on social media, in online communities, and in games (Chawki & El Shazly, 2013).<br>However, especially for non-native English speakers, due to cultural differences and language barriers, may vary in their understanding or interpretation of text-based sexual harassment (Welsh, Carr, MacQuarrie, & Huntley, 2006). To bridge this gap, previous studies have proposed large language models to detect and classify online sexual harassment, prompting a need to explore how language models comprehend the nuanced aspects of sexual harassment data. Prior to exploring the role of language models, it is critical to recognize the current gaps in knowledge that these models could potentially address in order to comprehend and interpret the complex nature of sexual harassment.</p><p><br></p><p dir="ltr">The Large Language Model (LLM) has attracted significant attention recently due to its exceptional performance on a broad spectrum of tasks. However, these models are characterized by being very sensitive to input data (Fujita et al., 2022; Wei, Wang, et al., 2022). Thus, the purpose of this study is to examine how various LLMs interpret data that falls under the domain of sexual harassment and how they comprehend it after replacing Out-of-Vocabulary words.</p><p dir="ltr"><br>This research examines the impact of Out-of-Vocabulary words on the performance of LLMs in classifying sexual harassment behaviors in text. The study compares the story classification abilities of cutting-edge LLM, before and after the replacement of Out-of-Vocabulary words. Through this investigation, the study provides insights into the flexibility and contextual awareness of LLMs when managing delicate narratives in the context of sexual harassment stories as well as raises awareness of sensitive social issues.</p>

<p dir="ltr">This research explores the application of large language models (LLMs) to comprehension of puns. Leveraging the expansive capabilities of LLMs, this study delves into the domain of pun classification by examining it through the prism of two humor theories: the Computational Model of Humor and the Benign Violation theory, which is an extension of the N+V Theory. The computational model posits that for a phrase to qualify as a pun, it must possess both ambiguity and distinctiveness, characterized by a word that can be interpreted in two plausible ways, each interpretation being supported by at least one unique word. On the other hand, the Benign Violation theory posits that puns work by breaching one linguistic rule while conforming to another, thereby creating a "benign violation." By leveraging the capabilities of large language models (LLMs), this research endeavors to scrutinize a curated collection of English language puns. Our aim is to assess the validity and effectiveness of the use of these theoretical frameworks in accurately classifying puns. We undertake controlled experiments on the dataset, selectively removing a condition specific to one theory and then evaluating the puns based on the criteria of the other theory to see how well it classifies the altered inputs. This approach allows us to uncover deeper insights into the processes that facilitate the recognition of puns and to explore the practical implications of applying humor theories. The findings of our experiments, detailed in the subsequent sections, sheds light on how the alteration of specific conditions impacts the ability of the LLMs to accurately classify puns, according to each theory, where each component of the theory does not influence the result to the same extent, thereby contributing to our understanding of humor mechanics through the eyes of LLMs.</p>

<p dir="ltr">Natural Language Processing (NLP), a vital branch of artificial intelligence, is designed to equip computers with the ability to comprehend and manipulate human language, facilitating the extraction and utilization of textual data. NLP plays a crucial role in harnessing the vast quantities of textual data generated daily, facilitating meaningful information extraction. Among the various techniques, keyphrase extraction stands out due to its ability to distill concise information from extensive texts, making it invaluable for summarizing and navigating content efficiently. The process of keyphrase extraction usually begins by generating candidates first and then ranking them to identify the most relevant phrases. Keyphrase extraction can be categorized into supervised and unsupervised approaches. Supervised methods typically achieve higher accuracy as they are trained on labeled data, which allows them to effectively capture and utilize patterns recognized during training. However, the dependency on extensive, well-annotated datasets limits their applicability in scenarios where such data is scarce or costly to obtain. On the other hand, unsupervised methods, while free from the constraints of labeled data, face challenges in capturing deep semantic relationships within text, which can impact their effectiveness. Despite these challenges, unsupervised keyphrase extraction holds significant promise due to its scalability and lower barriers to entry, as it does not require labeled datasets. This approach is increasingly favored for its potential to aid in building extensive knowledge bases from unstructured data, which can be particularly useful in domains where acquiring labeled data is impractical. As a result, unsupervised keyphrase extraction is not only a valuable tool for information retrieval but also a pivotal technology for the ongoing expansion of knowledge-driven applications in NLP.</p><p dir="ltr">In this dissertation, we introduce three innovative unsupervised keyphrase extraction methods: AttentionRank, AGRank, and LLMRank. Additionally, we present a method for constructing knowledge graphs from unsupervised keyphrase extraction, leveraging the self-attention mechanism. The first study discusses the AttentionRank model, which utilizes a pre-trained language model to derive underlying importance rankings of candidate phrases through self-attention. This model employs a cross-attention mechanism to assess the semantic relevance between each candidate phrase and the document, enhancing the phrase ranking process. AGRank, detailed in the second study, is a sophisticated graph-based framework that merges deep learning techniques with graph theory. It constructs a candidate phrase graph using mutual attentions from a pre-trained language model. Both global document information and local phrase details are incorporated as enhanced nodes within the graph, and a graph algorithm is applied to rank the candidate phrases. The third study, LLMRank, leverages the strengths of large language models (LLMs) and graph algorithms. It employs LLMs to generate keyphrase candidates and then integrates global information through the text's graphical structures. This process reranks the candidates, significantly improving keyphrase extraction performance. The fourth study explores how self-attention mechanisms can be used to extract keyphrases from medical literature and generate query-related phrase graphs, improving text retrieval visualization. The mutual attentions of medical entities, extracted using a pre-trained model, form the basis of the knowledge graph. This, coupled with a specialized retrieval algorithm, allows for the visualization of long-range connections between medical entities while simultaneously displaying the supporting literature. In summary, our exploration of unsupervised keyphrase extraction and biomedical data analysis introduces novel methods and insights in NLP, particularly in information extraction. These contributions are crucial for the efficient processing of large text datasets and suggest avenues for future research and applications.</p>

<p dir="ltr">This research investigates the use of Large Language Model (LLM) embeddings to capture the unique stylistic features of authors in Authorship Attribution (AA) tasks. Specifically, the focus of this research is on evaluating whether LLM-generated embeddings can effectively capture stylistic nuances that distinguish different authors, ultimately assessing their utility in tasks such as authorship attribution and clustering.The dataset comprises news articles from The Guardian authored by multiple writers, and embeddings were generated using OpenAI's text-embedding-ada-002 model. These embeddings were subsequently passed through a Siamese network with the objective of determining whether pairs of texts were authored by the same individual. The resulting model was used to generate style embeddings for unseen articles, which were then evaluated through classification and cluster analysis to assess their effectiveness in identifying individual authors across varying text samples. The classification task tested the model's accuracy in distinguishing authors, while the clustering analysis examined whether style embeddings primarily captured authorial identity or reflected domain-specific topics.</p><p dir="ltr">Our findings demonstrate that the proposed architecture achieves high accuracy for authors not previously encountered, outperforming traditional stylometric features and highlighting the effectiveness of LLM-based style embeddings. Additionally, our experiments reveal that authorship attribution accuracy decreases as the number of authors increases, yet improves with longer text lengths. </p><p dir="ltr"><br></p>

<p dir="ltr">Easy access, variety of content, and fast widespread interactions are some of the reasons that have made social media increasingly popular in our society. This has lead to many people use social media everyday for a variety of reasons, such as interacting with friends or consuming news content. Thus, understanding content on social media is more important than ever.</p><p dir="ltr">An increased understanding on social media can lead to improvements on a large number of important tasks. In this work, we particularly focus on fake news detection and political bias detection. Fake news, text published by news sources with an intent to spread misinformation and sway beliefs, is ever prevalent in today's society. Detecting it is an important and challenging problem to prevent large scale misinformation and maintain a healthy society. In a similar way, detecting the political bias of news content can provide insights about the different perspectives on social media.</p><p dir="ltr">In this work, we view the problem of understanding social media as reasoning over the relationships between sources, the articles they publish, and the engaging users. We start by analyzing these relationships in a graph-based framework, and then use Large Language Models to do the same. We hypothesize that the key to understanding social media is understanding these relationships, such as identifying which users have similar perspectives, or which articles are likely to be shared by similar users.</p><p dir="ltr">Throughout this thesis, we propose several frameworks to capture the relationships on social media better. We initially tackle this problem using supervised learning systems, improving them to achieve strong performance. However, we find that automatedly modeling the complexities of the social media landscape is challenging. On the contrary, having humans analyze and interact with all news content to find relationships, is not scalable. Thus, we then propose to approach enhance our supervised approaches by approaching the social media understanding problem \textit{interactively}, where humans can interact to help an automated system learn a better social media representation quality.</p><p dir="ltr">On real world events, our experiments show performance improvements in detecting the factuality and political bias of news sources, both when trained with and without minimal human interactions. We particularly focus on one of the most challenging setups of this task, where test data is unseen and focuses on new topics when compared with the training data. This realistic setting shows the real world impact of our work in improving social media understanding.</p>

<p dir="ltr">Neurological disorders present a considerable challenge due to their variety and diagnostic complexity especially for older adults. Early prediction of the onset and ongoing assessment of the severity of these disease conditions can allow timely interventions. Currently, most of the assessment tools are time-consuming, costly, and not suitable for use in primary care. To reduce this burden, the present thesis introduces passive digital markers for different disease conditions that can effectively automate the severity assessment and risk prediction from different modalities of electronic health records (EHR). The focus of the first phase of the present study in on developing passive digital markers for the functional assessment of patients suffering from Bipolar disorder and Schizophrenia. The second phase of the study explores different architectures for passive digital markers that can predict patients at risk for dementia. The functional severity PDM uses only a single EHR modality, namely medical notes in order to assess the severity of the functioning of schizophrenia, bipolar type I, or mixed bipolar patients. In this case, the input of is a single medical note from the electronic medical record of the patient. This note is submitted to a hierarchical BERT model which classifies at-risk patients. A hierarchical attention mechanism is adopted because medical notes can exceed the maximum allowed number of tokens by most language models including BERT. The functional severity PDM follows three steps. First, a sentence-level embedding is produced for each sentence in the note using a token-level attention mechanism. Second, an embedding for the entire note is constructed using a sentence-level attention mechanism. Third, the final embedding is classified using a feed-forward neural network which estimates the impairment level of the patient. When used prior to the onset of the disease, this PDM is able to differentiate between severe and moderate functioning levels with an AUC of 76%. Disease-specific severity assessment PDMs are only applicable after the onset of the disease and have AUCs of nearly 85% for schizophrenia and bipolar patients. The dementia risk prediction PDM considers multiple EHR modalities including socio-demographic data, diagnosis codes and medical notes. Moreover, the observation period and prediction horizon are varied for a better understanding of the practical limitations of the model. This PDM is able to identify patients at risk of dementia with AUCs ranging from 70% to 92% as the observation period approaches the index date. The present study introduces methodologies for the automation of important clinical outcomes such as the assessment of the general functioning of psychiatric patients and the prediction of risk for dementia using only routine care data.</p>

The development of artificial intelligence (AI) has revolutionised the way we interact with technology and each other, both in society and in professional careers. Although they come with great potential for productivity and automation, AI systems have been found to exhibit biases that reflect and perpetuate existing societal inequalities. With the recent rise of artificial intelligence tools exploiting the large language model (LLM) technology, such as ChatGPT, Bing Chat and Bard AI, this research project aims to investigate the extent of AI bias in said tools and explore its ethical implications. By reviewing and analysing responses to carefully crafted prompts generated by three different AI chatbot tools, the author will intend to determine whether the content generated by these tools indeed exhibits patterns of bias related to various social identities, as well as compare the extent to which such bias is present across all three tools. This study will contribute to the growing body of literature on AI ethics and inform efforts to develop more equitable and inclusive AI systems. By exploring the ethical dimensions of AI bias in selected LLMs, this research will shed light on the broader societal implications of AI and the role of technology in shaping our future.

<p dir="ltr">Sentiment analysis is a computational technique employed to extract and interpret subjective information from textual data. It involves the identification and classification of sentiments, opinions, and emotions expressed within the text. By analyzing linguistic cues, such as word choice, syntax, and sentiment lexicons, sentiment analysis can discern a range of emotions, from positive to negative, as well as more nuanced sentiments, such as anger, joy, or surprise. This powerful tool has the potential to unlock valuable insights from vast amounts of unstructured text data, which enables informed decision-making and effective communication in various domains, including education. </p><p dir="ltr">Recent advances in sentiment analysis have leveraged the power of deep neural networks, particularly general-purpose Large Language Models (LLMs) trained on extensive labeled datasets. However, real-world applications frequently encounter challenges related to the availability of large, high-quality labeled data and the computational resources necessary for training such models. </p><p dir="ltr">This research addresses these challenges by investigating effective strategies for utilizing LLMs in scenarios with limited data and computational resources. Specifically, this study explores three techniques: zero-shot learning, <i>N</i>-shot learning and fine-tuning. By evaluating these methods, this research aims to demonstrate the feasibility of employing general-purpose LLMs for sentiment analysis within educational contexts even when access to computational resources and labeled data is limited. The findings of this study reveal that different adaptation methods lead to significantly different LLM performance outcomes.</p>

2024 / This thesis empirically evaluates the impact of voluntary minimum wage changes to firm reputation using data from the hiring website Indeed.com. As a starting point, I show that when Target and Amazon unilaterally raised their minimum wages, their ratings on Indeed.com improved substantially across multiple dimensions: work-life balance, compensation, job security, management, and culture. Next, I examine the impact of a focal firm voluntarily raising its minimum wage on the ratings of similar firms in proximal locations. Using a differences-in-differences (DiD) design, I present preliminary evidence that competitors that are located near the focal firm may expe- rience a negative reputational shock relative to similar firms that are geographically distant. Additionally, this thesis applies novel sentiment analysis techniques to eval- uate minimum wage policies on review text. Using state-of-the-art NLP models such as Claude, ChatGPT, and RoBERTa, I identify and score two topics that are im- portant to job reviewers, but do not receive star ratings on Indeed.com: ”Scheduling and Hours,” and ”Workload and Compensation.” Finally, I use LLMs to perform zero-shot fine-grained sentiment analysis to investigate how a company’s reputation in regards to these topics is impacted by voluntary minimum wage policies. In these analyses, I am not able to refute the null hypothesis, though the method demostrates promise for further development.

Large Language Models (LLMs) have become ever larger in recent years, typically demonstrating improved performance as the number of parameters increases. This thesis investigates how the probability distributions output by language models differ depending on the size of the model. For this purpose, three features for capturing the differences between the distributions are defined, namely the difference in entropy, the difference in probability mass in different slices of the distribution, and the difference in the number of tokens covering the top-p probability mass. The distributions are then put into different distribution classes based on how they differ from the distributions of the differently-sized model. Finally, the distributions are transformed to be more similar to the distributions of the other model. The results suggest that classifying distributions before transforming them, and adapting the transformations based on which class a distribution is in, improves the transformation results. It is also shown that letting a classifier choose the class label for each distribution yields better results than using random labels. Furthermore, the findings indicate that transforming the distributions using entropy and the number of tokens in the top-p probability mass makes the distributions more similar to the targets, while transforming them based on the probability mass of individual slices of the distributions makes the distributions more dissimilar.