GraDVis: A Visualization Tool for a Visual Data Management System
Author: Jarrett Zapata
Faculty Supervisor: Shahrukh Humayoun
Department: Computer Science
The exponential growth of complex data, encompassing social media interactions, criminal records, and location data, necessitates efficient visualization techniques. Graph databases, such as Neo4j excel at managing relationships within such large datasets, with systems like VDMS, a novel data management solution, that treats visual data such as images, videos, and feature descriptors as first class objects in the system. The goal of VDMS is to enable efficient access of big-visual data to support visual analytics This is achieved by searching for relevant visual data via metadata stored as a graph. Although VDMS supports the multi-modality of the data (image, feature-vectors, meta-data and video), the task of inferring useful insights about this data is tedious. VDMS keeps the relation between this complex data point using a graph database, but the intricate nature of graph data structures poses challenges in analysis, hindering pattern discovery and potentially overwhelming users.
This work addresses the above-mentioned gap by proposing an interactive graph database visualization tool, called GraDVis (Graph Database Visualizer), tailored to VDMS main features. GraDVis visualizes the relationship between different visual data points (images, descriptors, and meta-data), enabling user to get insights about the stored graph data efficiently and interactively. We built a web-application UI to interact with VDMS through HTTP-server. The client application may send JSON request to find-entities (node, images, or descriptors) and retrieve a JSON response attached with a blob data as answer to the sent-query. Then, we can visualize the whole graph and on-demand retrieve more insights about the selected nodes. We show the visualization of the retail scenario to get shopper insights in a retail store use case scenario.