Browsing by Author "Lin, Jiayue"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
- ImageSI: Interactive Deep Learning for Image Semantic InteractionLin, Jiayue (Virginia Tech, 2024-06-04)Interactive deep learning frameworks are crucial for effectively exploring and analyzing complex image datasets in visual analytics. However, existing approaches often face challenges related to inference accuracy and adaptability. To address these issues, we propose ImageSI, a framework integrating deep learning models with semantic interaction techniques for interactive image data analysis. Unlike traditional methods, ImageSI directly incorporates user feedback into the image model, updating underlying embeddings through customized loss functions, thereby enhancing the performance of dimension reduction tasks. We introduce three variations of ImageSI, ImageSI$_{text{MDS}^{-1}}$, prioritizing explicit pairwise relationships from user interaction, and ImageSI$_{text{DRTriplet}}$ and ImageSI$_{text{PHTriplet}}$, emphasizing clustering by defining groups of images based on user input. Through usage scenarios and quantitative analyses centered on algorithms, we demonstrate the superior performance of ImageSI$_{text{DRTriplet}}$ and ImageSI$_{text{MDS}^{-1}}$ in terms of inference accuracy and interaction efficiency. Moreover, ImageSI$_{text{PHTriplet}}$ shows competitive results. The baseline model, WMDS$^{-1}$, generally exhibits lower performance metrics.
- Topic Modeling ToolkitLin, Jiayue; Pang, Mingkai; Liu, Yulong (Virginia Tech, 2023-05-08)The Topic Modeling Toolkit project began with an existing text mining toolkit and aimed to enhance its functionality by incorporating cutting-edge topic modeling techniques. Specifically, BERTopic, CTM, and LDA were used to extract pertinent topics from a corpus of text documents. The resulting web-based platform provides users with a search engine, a recommendation system, and a usable interface for browsing and exploring these topics. In addition to these enhancements, our team also implemented a text-filtering framework and redesigned the user interface using Tailwind CSS. The final deliverables of the project include a fully functional website, user documentation, and an open-source toolkit that can be used to train machine learning models and support browsing and searching for various text datasets. While the current version of the toolkit includes BERTopic, CTM, and LDA, there is potential for future work to incorporate additional topic modeling methods. It is important to note that while the project originally focused on electronic theses and dissertations (ETDs), the resulting platform can be used to explore and comprehend complex subjects within any corpus of text documents. The topic modeling toolkit is available as an open-source package that users can install and use on their own computers. It is available for use and can be used to support browsing and searching for various text datasets. The intended user group for the platform includes researchers, students, and other users interested in exploring and understanding complex topics within a given corpus of text documents. The resulting topic modeling toolkit offers features that facilitate the exploration and comprehension of intricate topics within text document collections. This tool has the potential to aid researchers, students, and other users in their respective fields.