Search

Fine-tuning a pre-trained sequence-to-sequence-based language model has significantly advanced the field of abstractive summarization. However, the early models of abstractive summarization were limited by the gap between training and inference, and they did not fully utilize the potential of the language model. Recent studies have introduced a two-stage framework that allows the second-stage model to re-rank the candidate summary generated by the first-stage model, to resolve these limitations. In this study, we point out that the supervision method performed in the existing re-ranking model of the two-stage abstractive summarization framework cannot learn detailed and complex information of the data. In addition, we present the problem of positional bias in the existing encoder–decoder-based re-ranking model. To address these two limitations, this study proposes a hierarchical supervision method that jointly performs summary and sentence-level supervision. For sentence-level supervision, we designed two sentence-level loss functions: intra- and inter-intra-sentence ranking losses. Compared to the existing abstractive summarization model, the proposed method exhibited a performance improvement for both the CNN/DM and XSum datasets. The proposed model outperformed the baseline model under a few-shot setting. [ABSTRACT FROM AUTHOR]

Problem solving applications require users to exercise caution in their data usage practices. Prior to installing these applications, users are encouraged to read and comprehend the terms of service, which address important aspects such as data privacy, processes, and policies (referred to as information elements). However, these terms are often lengthy and complex, making it challenging for users to fully grasp their content. Additionally, existing transparency analytics tools typically rely on the manual extraction of information elements, resulting in a time-consuming process. To address these challenges, this paper proposes a novel approach that combines information visualization and machine learning analyses to automate the retrieval of information elements. The methodology involves the creation and labeling of a dataset derived from multiple software terms of use. Machine learning models, including naïve Bayes, BART, and LSTM, are utilized for the classification of information elements and text summarization. Furthermore, the proposed approach is integrated into our existing visualization tool TranspVis to enable the automatic detection and display of software information elements. The system is thoroughly evaluated using a database-connected tool, incorporating various metrics and expert opinions. The results of our study demonstrate the promising potential of our approach, serving as an initial step in this field. Our solution not only addresses the challenge of extracting information elements from complex terms of service but also provides a foundation for future research in this area. [ABSTRACT FROM AUTHOR]

In this paper, we present a rigorous mathematical examination of generative pre-trained transformer (GPT) models and their autoregressive self-supervised learning mechanisms. We begin by defining natural language space and knowledge space, which are two key concepts for understanding the dimensionality reduction process in GPT-based large language models (LLMs). By exploring projection functions and their inverses, we establish a framework for analyzing the language generation capabilities of these models. We then investigate the GPT representation space, examining its implications for the models' approximation properties. Finally, we discuss the limitations and challenges of GPT models and their learning mechanisms, considering trade-offs between complexity and generalization, as well as the implications of incomplete inverse projection functions. Our findings demonstrate that GPT models possess the capability to encode knowledge into low-dimensional vectors through their autoregressive self-supervised learning mechanism. This comprehensive analysis provides a solid mathematical foundation for future advancements in GPT-based LLMs, promising advancements in natural language processing tasks such as language translation, text summarization, and question answering due to improved understanding and optimization of model training and performance. [ABSTRACT FROM AUTHOR]

In the area of text summarization, there have been significant advances recently. In the meantime, the current trend in text summarization is focused more on news summarization. Therefore, developing a synthesis approach capable of extracting, comparing, and ranking sentences is vital to create a summary of various news articles in the context of erroneous online data. It is necessary, however, for the news summarization system to be able to deal with multi-document summaries due to content redundancy. This paper presents a method for summarizing multi-document news web pages based on similarity models and sentence ranking, where relevant sentences are extracted from the original article. English-language articles are collected from five news websites that cover the same topic and event. According to our experimental results, our approach provides better results than other recent methods for summarizing news. [ABSTRACT FROM AUTHOR]

Legal judgments are generally very long, and relevant information is often scattered throughout the text. To complete a legal judgment summarization, capturing important, relevant information comprehensively from a lengthy text is crucial. The existing abstractive-summarization models based on pre-trained language have restrictions on the length of an input text. Another concern is that the generated summaries have not been well integrated with the legal judgment's technical terms and specific topics. In this paper, we used raw legal judgments as information of different granularities and proposed a two-stage text-summarization model to handle different granularities of information. Specifically, we treated the legal judgments as a sequence of sentences and selected key sentence sets from the full texts as an input corpus for summary generation. In addition, we extracted keywords related to technical terms and specific topics in the legal texts and introduced them into the summary-generation model as an attention mechanism. The experimental results on the CAIL2020 and the LCRD datasets showed that our model achieved an overall 0.19–0.41 improvement in its ROUGE score, as compared to the baseline models. Further analysis also showed that our method could comprehensively capture essential and relevant information from lengthy legal texts and generate better legal judgment summaries. [ABSTRACT FROM AUTHOR]