32584 research outputs found
Sort by
Enhancing Recommender Systems Using Graph Neural Networks, Long Short-Term Memory and Textual Embeddings
No full textRecommender systems surround us. They shape what we watch, how we buy, and even what our future might look like next. The Amazon Review Dataset and Movielens, those two datasets will help this project explore how to improve recommender systems through the user’s preferences. Two methods were combined: sequence-based models and graph-based models. Sequence models, such as LSTMs and Transformers, look at the order of user actions to find patterns by their sequence. On the other hand, Graphbased models focus on relationships between users, items, and their attributes. Textual embeddings added depth and context. Both methods offer something special, according to metrics such as Precision@K and NDCG@K. They also offer even more scope for improving recommendations, with together offering them a further boost. This project highlights how combining time-based insights with Relational data provides more user experience
MediLightRAG: A System for Medical Query Response Using Fine-Tuned LLMs and Graph Based Retrieval
No full textThe exponential increase in medical data has created a greater demand for precise and efficient information retrieval systems. Existing Large Language Models (LLMs) face domain-specific difficulties such as sophisticated medical jargon, situational comprehension, and the continual advancement of healthcare knowledge. To tackle these challenges, we present MediLightRAG, an innovative two-stage system which integrates parameter-efficient fine-tuning of Large Language models with LightRAG’s graph-based retrieval. The first stage focuses on enabling accurate resource-efficient model adaptation for the medical domain through QLoRA fine-tuning. In the second stage, LightRAG’s two-tiered retrieval architecture that combines graph-based indexing with dynamic knowledge retrieval is employed to enhance information coverage and precision. MediLightRAG’s incremental knowledge update feature allows newer medical research to be seamlessly integrated. Using sophisticated retrieval techniques alongside compute-efficient tuning, this new system greatly enhances the accuracy of medical query responses and clinical decision-making tasks
Mitigating Cold Start Problem through Metadata Integration and User Preference Analysis
No full textRecommendation systems power the most popular platforms in the world: from content catalogs on Netflix to custom feeds on TikTok – the importance of recommendation systems is significant. Collaborative filtering, the most popular recommendation technique, is essentially based on the idea of leveraging collective user intelligence i.e., creating recommendations by finding similar users. But this technique suffers when there is not enough data in the profiles of users, formally termed as the cold start problem. This research focuses on this problem by introducing an approach that integrates metadata-driven similarity measures with profile expansion techniques. Our approach combines traditional collaborative filtering with profile expansion and additional signals derived from user and item metadata to enrich sparse user profiles. We use profile expansion strategies to generate pseudo-ratings that when integrated with metadata-based similarities, create a robust hybrid model. Experiments conducted on MovieLens 100k, 1M and MovieDex datasets show that our approach significantly improves performance under severe cold start conditions. This work not only provides a comprehensive framework for enhancing recommendation accuracy in cold start but also outlines key insights for dynamic adjustment of metadata and expansion influence as user profiles & the overall system evolve
An Evidence-Based Approach to Predicting Pancreatic Ductal Adenocarcinoma
No full textPancreatic ductal adenocarcinoma (PDAC) is a complex disease with hidden clinical indicators, so a reliable diagnosis of PDAC requires high precision and sophisticated analysis. Traditional probabilistic methods often rely on making unwarranted assumptions or undesirable approximations about probabilistic estimates, limiting their ability to provide the precision needed for correct diagnosis and treatment planning. In contrast, Dempster–Shafer Theory offers a formal framework for integrating uncertain and potentially conflicting evidence. This makes it well-suited for analyzing incomplete and ambiguous data typically associated with PDAC. By employing an evidential reasoning (ER) model based on Dempster-Shafer Theory, this approach systematically combines and evaluates imperfect evidence, eliminating the requirement to make unwarranted assumptions or approximations and enhancing analytical fidelity. This project aims to develop a model to improve prediction accuracy and fidelity and support more reliable diagnostic decisions for PDAC
VisionMate: AI-Powered Image Captioning Web Application
No full textVisionMate is a web application that generates captions for camera-captured images. It is designedto assist users with visual impairments by converting visual input into spoken and written text. Theapplication uses the GIT-base model from Hugging Face, which processes the image and returns adescriptive caption. Users can take a picture using the device camera—either via webcam ondesktop or the native camera interface on mobile. The app provides audio output using theSpeechSynthesis API and uses full-screen tap interaction to simplify accessibility.The frontend is implemented in React.js, and the backend is built with FastAPI. The backend callsHugging Face’s Inference API to perform model inference without loading large models locally,reducing memory usage during deployment. On average, captions are generated in 5 to 8 seconds.The GIT-base model was selected after comparative testing with BLIP-base, BLIP-large, and GIT-large. Testing was conducted on Chrome, Safari, and Firefox browsers, using devices such as theMacBook Pro (M1) and iPhone 15.This report outlines the system architecture, model comparisons, deployment on Vercel (frontend)and Render (backend), and evaluation of performance across speed, model accuracy, and device andbrowser compatibility
Understanding Residential Development in a High-Quality Transit Area (HQTA): An Application of Deep Learning
Get PDFConsiderable research addresses the dynamics between land use and public transportation, including the impacts of transit systems on urban development and vice versa. However, this research has underestimated the heterogeneity of contributing factors to urban developments near transit by property type: vacant lot and occupied property. That is, what drives development in vacant lots differs from what drives development in already-occupied property. A lack of comprehensive data that differentiates development records based on the type of property is one explanation for this research gap. This project fills this research gap by employing a deep learning algorithm and separately examining the contributing factors to vacant lot and occupied property development near transit. A deep learning model—a foundation model—detects and classifies residential development on vacant and occupied properties in high-quality transit areas (HQTAs) in Los Angeles County. Taking the classification as the dependent variable, this research constructs two multi-level logistic regression models: vacant and occupied models. The findings confirm the heterogeneity of the contributing factors to the development of vacant and occupied parcels. The findings also indicate that while the factors at the property level are more significant than the neighborhood-level factors in both models, the significance of the property characteristics in vacant lot development is more significant than in occupied lot development. While infill development on occupied lots in core cities is more likely to occur, the inner suburban cities tend to experience residential development on both vacant and occupied properties much less than the other areas. Another interesting contrast between vacant and occupied lot development is that occupied lot development is more likely to be associated with urban functions/amenities, while the vacant lot development tends to occur in areas with tranquility (where amenities are fewer) near transit. These insights highlight the need for strategies specific to property type in transit-oriented development planning and suggest that nuanced, data-driven approaches are essential to promoting efficient urban growth around transit hubs
Tsao, H.-S. Jacob
Get PDFUniversity of California, Berkeley, Operations Research, Ph.D. 1984
University of Texas at Dallas, Mathematical Statistics, MS, 1979
National Chiao-Tung University, Taiwan, Applied Mathematics, BS, 1976https://scholarworks.sjsu.edu/erfa_bios/1400/thumbnail.jp
Spartan Daily, January 28, 2025
Get PDFVolume 164, Issue 2https://scholarworks.sjsu.edu/spartan_daily_2025/1001/thumbnail.jp
Spartan Daily, February 13, 2025
Get PDFVolume 164, Issue 10https://scholarworks.sjsu.edu/spartan_daily_2025/1009/thumbnail.jp
Examining the Impact of Registered Nurse Engagement in Medical Assistant Clinical Competency Development and Team Effectiveness in Ambulatory Care
Get PDFBackground
As ambulatory care grows in complexity, there is a growing need to ensure that staff possess a high level of competence and team effectiveness to provide safe, efficient, and cost-effective care. Leveraging the expertise and leadership skills of Registered Nurses (RN) to support the team of unlicensed, Medical Assistants (MA) will be incentivized. With limited knowledge of MA preparation to meet the rapidly changing demands, managers will struggle to build effective teams and look to the impact of RN- supported MA onboarding to bridge skill gaps.
Purpose
To examine this issue, a quasi-experimental, quality improvement study was done to explore MA competence and RN perceptions of work effectiveness using two MA onboarding models: one that utilizes an RN to support MA onboarding and one that utilizes a peer MA.
Methods
Previously collected blood pressure competency data obtained from direct observation of MAs were obtained for the two groups. Organizational relationship and global empowerment survey scores were collected for the RNs in the two groups.
Results
No significant difference was found between the two groups in MA blood pressure competency nor the RNs organizational effectiveness scores. A strong correlation was identified between an RNs perception of work effectiveness with their manager and other support staff. RN employment tenure was strongly linked to improved team empowerment.
Implications for Practice
Findings suggest that RN tenure with the employer and manager-RN collaboration strength may be more predictive of team empowerment and RN-MA work effectiveness than direct involvement in MA onboarding or competency development alone. Efforts to strengthen the Manager-RN relationship in leadership partnership may be key to improving team effectiveness in the rapidly changing healthcare environment