1 Motivation

This project is developed as part of the VAST Mini Challenge 2025. The chosen topic is Mini-Challenge 3, which focuses on the investigation of rising tensions and suspicious activities in Oceanus—an island community undergoing economic and regulatory transformations. Following crackdowns on illegal fishing, many suspects are believed to have redirected their interests into the tourism sector, potentially exploiting it for hidden agendas.

Led by investigative journalist Clepper Jessen, the case revolves around intercepted radio communications tied to the temporary closure of Nemo Reef. These communications hint at expedited approvals, pseudonyms, and covert logistics involving public officials, influential families, and celebrities such as Sailor Shift. Our task is to apply novel visual analytics approaches to decode patterns, relationships, and suspicious behaviors hidden in these interactions to help uncover the truth behind Oceanus’s unfolding story.

2 Objectives

The objective of this report is to analyze intercepted radio communications over a two-week period to uncover patterns, relationships, and potentially illicit activities among people and vessels operating in Oceanus. Using a knowledge graph and visual analytics, the investigation aims to:

Identify daily temporal communication patterns and their evolution.
Detect and interpret relationship clusters among individuals and vessels, based on shared communication or common topics.
Uncover the use of pseudonyms and determine real identities or recurring patterns behind them.
Evaluate the activities of Nadia Conti for potential ongoing illegal behavior.

The ultimate goal is to assist Clepper in gaining actionable insights into key actors, their interactions, and suspicious behaviors within the Oceanus network.

3 Data

The primary dataset for this project is a knowledge graph constructed from intercepted radio communications collected over a two-week period in Oceanus. In this graph, each node represents an entity—such as a person, vessel, pseudonym, or organization—while the edges represent interactions or co-occurrences in communications. Each edge carries contextual attributes such as the timestamp, message topic, communication type, and potential pseudonym use.

Additional contextual data from the Mini Challenge background supports the analysis, including: - Entity affiliations and roles (e.g., Green Guardians, Sailor Shift’s entourage). - Identified pseudonyms (e.g., “Boss”, “The Lookout”). - Records of previous illicit activities (e.g., Nadia Conti’s involvement).

Refer to the reference for a detailed breakdown of the dataset.

4 Methodology & Analytical Approach

4.1 Data Preparation & EDA

For this project, the following approaches will be taken by the team:

Exploring Graph Structure: Understanding the types, subtypes, and attributes of nodes and edges.
Extracting & Tidying Tables: Converting nodes and edges into tabular tibbles using jsonlite and tibble.
Filtering Communication Events: Isolating only Event nodes with subtype Communication, which contain timestamps and message content.
Identifying Sender–Receiver Relationships: Filtering edge types for “sent” and “received” and reshaping into a unified format.
Merging Communication Metadata: Joining timestamps and content to each sender–receiver pair.
Standardizing Entity Labels: Resolving pseudonyms and inconsistent naming using the node label and sub_typefields.
Enhancing with Time Features: Deriving fields like weekday, hour, and week group for temporal analysis.

4.2 Temporal Pattern & Influence Network Analysis

This is to solve Q1a–c: Daily communication trends, how they evolve over time, and identification of influence hubs.

Techniques and outputs include:

Time Series Visualization: Plotting communication volume by day and hour, revealing routine versus reactive patterns across the two-week timeline.
Week-over-Week Comparison: Faceted line and heatmap visualizations of message intensity by weekday and hour, identifying shifts during the closure of Nemo Reef.
Influence Network Graphs: Directed visNetwork plots mapping who contacts whom, segmented by entity types (e.g., person-to-vessel, person-to-organization).
Insights into Influence: Identification of influential entities like “Boss,” “Green Guardians,” or “V. Miesel Shipping,” based on volume, directionality, and centrality of communications.

4.3 Relationship Clusters & Pseudonym Mapping

This is to solve:

Q2a–b: Who interacts frequently and which groups emerge?
Q3a–b: Who is hiding behind pseudonyms, and what can we infer about their real identities?

Key methods include:

Graph-Based Clustering: Using tidygraph, ggraph, and visNetwork to visualize groupings based on shared communication edges.
Topic Labeling by Cluster: Assigning thematic labels (e.g., “Conservationists,” “Permit Brokers,” “Sailor Shift’s Crew”) based on common keywords and communication targets.
Pseudonym Detection: Mining the content column to extract frequent aliases (e.g., “The Lookout,” “Boss,” “Mrs. Money”), and mapping them to known actors via message patterns, directionality, and co-occurrence.
Clustered Network with Alias Highlighting: Annotating group-level graphs with nodes styled by pseudonym or known identity.

5 Prototype

The prototype is splitted as follows:

For detail, please refer to Prototype tab

6 Storyboard

To support our investigation, we have designed and developed a prototype that visually outlines the structure of the R Shiny Application. This storyboard illustrates both the user interface (UI) layout and the corresponding server-side logic, offering a clear representation of how shareholders in Oceanus interact with the visual analytics workflow.

For detail, please refer to Storyboard page

7 Packages

To leverage the investigation, the following R packages has been used:

tidyverse
lubridate -igraph
ggraph
wordcloud
tidytext
stopwords
dplyr
tibble
visNetwork
stringr
knitr
ggplot2
readr
jsonlite
janitor
forcats
ggrepel

8 Reference

MC 3 Data Description