Publishing type：Research paper (scientific journal)   Kind of work：Joint Work  

The understanding of narrative stories by computer is an important task for their automatic generation. To date, high-performance neural-network technologies such as BERT have been applied to tasks such as the Story Cloze Test and Story Completion. In this study, we focus on the text segmentation of novels into paragraphs, which is an important writing technique for readers to deepen their understanding of the texts. This type of segmentation, which we call “paragraph boundary recognition”, can be considered to be a binary classification problem in terms of the presence or absence of a boundary, such as a paragraph between target sentences. However, in this case, the data imbalance becomes a bottleneck because the number of paragraphs is generally smaller than the number of sentences. To deal with this problem, we introduced several cost-sensitive loss functions, namely. focal loss, dice loss, and anchor loss, which were robust for imbalanced classification in BERT. In addition, introducing the threshold-moving technique into the model was effective in estimating paragraph boundaries. As a result of the experiment on three newly created datasets, BERT with dice loss and threshold moving obtained a higher F1 than the original BERT had using cross-entropy loss as its loss function (76% to 80%, 50% to 54%, 59% to 63%).

DOI： 10.3390/app11125632

Publishing type：Research paper (scientific journal)  

Publishing type：Research paper (scientific journal)  

Kind of work：Joint Work  

Kind of work：Joint Work  

DOI： 10.1016/S0306-4573(00)00050-9

Total pages：196   Responsible for pages：62-76   Book type：Textbook, survey, introduction Participation form：Corresponding Author

Authorship：Lead author,　Corresponding author   Publishing type：Article, review, commentary, editorial, etc. (scientific journal)   Kind of work：Single Work   International / domestic magazine：Domestic journal  

<p>2022年4月より筆者の所属だった「大阪府立大学」は大阪市が設置運営している「大阪市立大学」と合併して，日本で最大級の公立大学「大阪公立大学」が誕生することと</p>

DOI： 10.1541/ieejjournal.142.721

Presentation type：Oral presentation (general)  

Presentation type：Oral presentation (general)  

<p>Nowadays, there is a large amount of text data such as SNS posts and reviews on the Internet. Such text data contain a lot of useful information such as evaluations and impressions of various objects. However, in order to make use of such information, it is essential to label them appropriately and to automate the labeling process. Sentiment analysis is a task in the field of natural language processing that classifies text into positive or negative polarity, and in particular aspect-based sentiment analysis extracts multiple objects in a text and classifies their polarity, which is effective for automating label assignment in that it can extract detailed information. In this study, based on Mpm+T, a model proposed by Kusumoto for Japanese aspect-based sentiment analysis, we propose an extension of Mpm+T to improve Mpm+T to handle data in which a positive and a negative label are assigned simultaneously to a single target, which is not handled by Mpm+T. We also verify the effectiveness of the proposed method through experiments.</p>

DOI： 10.11517/pjsai.jsai2024.0_2g6gs602

Presentation type：Oral presentation (general)  

<p>Machine learning-based image recognition has gained significant attention, mainly using Convolutional Neural Networks (CNNs). As the complexity of problems increases, so does the complexity of CNN architectures. This makes finding the optimal CNN structure a challenging combinatorial optimization problem. Manual settings are time-consuming and labor-intensive. To address this, the field of AutoML has introduced gaCNN, which uses a genetic algorithm for CNN structure search, and tdgaCNN, which applies thermodynamic selection rules. These methods have shown superiority over traditional ones. In this study, we propose a tdgaCNN extension that incorporates skip connections to enhance performance. Its effectiveness is demonstrated on an image benchmark dataset.</p>

DOI： 10.11517/pjsai.jsai2024.0_2d4gs203

Presentation type：Oral presentation (general)  

<p>Recent advancements in Large Language Models (LLMs) within the artificial intelligence domain have shown exceptional performance across various natural language processing tasks. Amidst these developments, aligning the values and objectives of LLMs with human perspectives has become increasingly important. Reinforcement Learning from Human Feedback (RLHF) has gained notable interest as a method for such alignment adjustments. This study explored a learning approach for LLMs using RLHF, employing scenarios from the romance simulation game 'Tokimeki Memorial 3' as the game scenario data. Specifically, the research involved an experiment where sentences were generated following five Japanese characters, tailored to align with the personalities of the game characters. While subjective, this evaluation demonstrated the capability of producing sentences that appropriately matched the distinct characters in the game.</p>

DOI： 10.11517/pjsai.jsai2024.0_4a1gs602

Presentation type：Oral presentation (general)  

<p>In recent years, with the rapid development of artificial intelligence technology, Knowledge Graph, which systematically connects various kinds of human knowledge and expresses their relationships in a graph structure, has attracted much attention and is used as a fundamental technology for artificial intelligence in various fields. In this context, there is a need for an automatic complementation method of Knowledge Graphs to meet the demand for adding new knowledge to existing Knowledge Graphs. The problem with conventional Knowledge Graph completion methods such as TransE and ComplEx is that they focus on knowledge relationships and do not effectively capture the semantic information of the knowledge itself. In this study, we proposed an automatic Knowledge Graph completion method using Masked Language Modeling by BERT, which is a deep language model, to effectively capture the semantic information of knowledge itself, and verified its effectiveness through evaluation experiments.</p>

DOI： 10.11517/pjsai.jsai2024.0_4f1gs304

Audience：High school students,　College students,　General

Keyword：自然言語処理, 自然言語理解, 談話理解 

ヒトと会話ができるコンピュータはどうしたら作れるか、また、現在のコンピュータとはどこまで会話できるのか、さらにどのような問題がまだ残っているのかについて皆さんと考えてみましょう．