🤖 AI

Exploring ESC Winners with Nested Diagrams

This paper introduces ConceptFlow, a Python library for Formal Concept Analysis that generates interactive nested line diagrams to reveal the relationship between voting patterns and musical characteristics in Eurovision Song Contest winners from 1975 to 2025.

Original authors: Anurag Sharma, Marcel Nöhre, Gerd Stumme

Published 2026-08-17

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Anurag Sharma, Marcel Nöhre, Gerd Stumme

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery, but instead of fingerprints and alibis, your clues are patterns in data. This is the world of Formal Concept Analysis (FCA), a branch of mathematics that helps us organize messy information into neat, logical structures. Think of FCA as a super-powered filing cabinet that doesn't just sort files by name, but automatically figures out which files belong together based on the things they have in common. If you have a list of songs and a list of countries, FCA can tell you, "Hey, all the songs from Country A share this specific musical trait," or "Every time a song has a fast beat, it gets votes from this specific group of neighbors."

The paper we are exploring today uses this mathematical filing system to look at the Eurovision Song Contest, the massive annual music competition where countries vote for each other's songs. Everyone knows Eurovision has a reputation for "voting blocs"—where neighbors or culturally similar countries tend to give each other high scores, sometimes regardless of the song's quality. But what if the music itself plays a role? Do certain types of songs only win when they come from specific voting alliances? The authors built a new digital tool to untangle these two worlds: the political and cultural reasons people vote, and the actual musical notes and speeds of the songs. They want to see if the way a song is voted for is secretly linked to how it sounds.

The Musical Detective Tool: ConceptFlow

The authors, Anurag Sharma, Marcel Nöhre, and Gerd Stumme, created a new software tool called ConceptFlow. You can think of this tool as a "nested diagram generator." Imagine a set of Russian nesting dolls, but instead of wooden figures, each doll is a map of relationships. The outer doll represents one big picture (like voting patterns), and inside every single part of that outer doll, there is a smaller, detailed doll representing a different picture (like musical traits).

This tool is built to work with Python, a popular programming language, and it follows a standard set of rules (called the "scikit-learn" interface) so that other scientists can easily plug it into their own work. The main job of ConceptFlow is to take a giant, complicated list of data—where every entry has many different types of information—and break it down into two layers: an outer scale and an inner scale.

The Case of the 50 Winners

To test their tool, the team looked at the winners of the Eurovision Song Contest from 1975 to 2025. They skipped the year 2020 because the contest was cancelled that year, leaving them with 50 winning songs to analyze.

They gathered two types of information for each winner:

The Voting Support (The Outer Scale): They looked at how much support each winner got from countries that share ties with them. They grouped these ties into four categories: regional (neighbors), cultural (shared history or language), historical (past connections), and political (alliances). If a winner got an average of at least 8 points from a specific group of countries, they marked that as a "strong tie."
The Musical Traits (The Inner Scale): They looked at the songs' tempo (how fast the beat is) and key (whether the song sounds happy/major or sad/minor). They turned the speed into simple categories: songs with a tempo of at least 100 BPM (beats per minute) and songs with a tempo of at least 150 BPM.

The Magic of the Nested Diagram

Usually, if you try to draw a map showing both voting patterns and musical traits at the same time, it becomes a messy, tangled spiderweb that is impossible to read. ConceptFlow solves this by using a "nested line diagram."

Imagine a large tree (the outer diagram) where each branch represents a type of voting support. Now, imagine that inside the leaves of that tree, there are tiny, identical maps (the inner diagrams) showing the musical traits. The software fills in the spots on these tiny maps where a real winner actually exists. If a spot is empty (hollow), it means that combination of voting support and musical traits never happened in the data.

The authors used a method called "conceptual scaling" to turn their raw data into these two layers. Then, they used a mathematical trick to figure out exactly which spots in the nested maps should be filled. They didn't just draw a picture; they calculated the "filled nodes" (the real winners) and the "unfilled nodes" (the missing combinations).

What the Diagram Revealed

When they looked at the final diagram, some interesting patterns popped up, showing how voting and music are linked in the winners they studied:

Fast Songs Need Cultural Friends: Every single winning song that was very fast (at least 150 BPM) also had strong cultural support. In the diagram, there was no spot for a "fast song without cultural support" that was filled. This suggests that in this dataset, if a song is super fast, it seems to need a cultural connection to win.
Sad Songs Have a Minimum Speed: Among the winners who got support from their region or culture, every song with a "sad" or minor key had a tempo of at least 100 BPM. The diagram showed that a "slow, sad song" with regional or cultural support simply didn't exist among the winners. The spot for that combination was empty.
The Ultimate Combo: There was a specific combination that only appeared with all types of support. If a winner had a minor key, a fast tempo (at least 150 BPM), and cultural support, they also had political and historical support. The diagram showed that this specific musical mix never happened without the extra political and historical backing.

Why This Matters (And What It Doesn't)

The authors are careful to say that this isn't a complete study of why people vote the way they do. Instead, it's a demonstration of what their new tool, ConceptFlow, can do. They aren't claiming to have solved the mystery of Eurovision forever. They are showing that by using nested diagrams, you can see hidden rules and connections that are hard to spot in a spreadsheet.

The diagram acts like a spotlight, showing you exactly where certain musical combinations are possible and where they are blocked by the rules of voting. For example, it highlights that you can't have a slow, sad song win with just regional support; something else is needed.

In the future, the team hopes to add more features to ConceptFlow, like automatically listing the "rules" (implications) that the diagram reveals, so users can click on a spot and see exactly why a certain combination is missing. For now, they have successfully built a bridge between the math of voting patterns and the art of song structure, proving that sometimes, the best way to understand a complex system is to look at it through a set of nested, mathematical dolls.

Technical Summary: Exploring ESC Winners with Nested Diagrams

Problem Statement
The Eurovision Song Contest (ESC) presents a complex dataset where voting patterns are influenced by regional, cultural, historical, and political alliances, while simultaneously being determined by the musical characteristics of the entries (e.g., tempo and key). Traditional analysis often treats these dimensions separately. The challenge lies in visualizing the dependencies between these distinct conceptual scales—specifically, how voting support clusters relate to specific musical properties within a single conceptual structure. Standard Formal Concept Analysis (FCA) line diagrams can become cluttered when dealing with many-valued contexts containing diverse attribute types, making it difficult to discern implications across different domains.

Methodology
The authors propose ConceptFlow, a Python library compatible with the scikit-learn estimator interface, designed to construct and render nested line diagrams from many-valued formal contexts. The methodology proceeds as follows:

Data Preparation and Scaling:
- The study utilizes a dataset of 50 ESC winners (1975–2025, excluding 2020).
- Outer Scale (Voting): Attributes are derived from four manual clusterings of participating countries based on regional, cultural, historical, and political ties. A winning entry is associated with a cluster if it received an average of at least eight points from eligible countries in that cluster.
- Inner Scale (Musical): Attributes are derived from song metadata. Tempo is converted into ordinal Boolean attributes ( $\ge$ 100 BPM, $\ge$ 150 BPM), and key is represented as a dichotomic attribute (Major/Minor).
- This process transforms the many-valued context into an apposition of two formal contexts, $K_{outer}$ and $K_{inner}$ , sharing the same object set (the winners).
Lattice Construction and Computation:
- Instead of explicitly constructing the concept lattice of the full apposition, the authors compute the filled nodes of the subdirect product directly from the factor lattices of $K_{outer}$ and $K_{inner}$ .
- The algorithm identifies atomic coordinate pairs $(\gamma_{outer}(g), \gamma_{inner}(g))$ for each object $g$ , adds the bottom element, and iteratively computes componentwise joins until a fixpoint is reached. This identifies the specific inner concepts that must be "filled" within the outer diagram.
Visualization:
- The outer lattice is laid out using DimFlux, a force-based model that projects coordinates for readability.
- The inner lattice layout is computed once and reused as a fixed template inside every node of the outer lattice.
- The final output is a JSON representation rendered as an interactive D3.js visualization, separating the rendering layer from the FCA construction logic.

Key Contributions

ConceptFlow Library: The primary contribution is the release of ConceptFlow, a tool that integrates conceptual scaling, factor lattice computation, and nested line diagram generation into a scikit-learn-compatible workflow.
Nested Visualization Technique: The paper demonstrates a specific application of nested line diagrams where an outer scale (voting support) contains copies of an inner scale (musical properties). This allows for the visualization of the subdirect product of two concept lattices without the visual complexity of a single, massive lattice.
Algorithmic Efficiency: The approach avoids constructing the full concept lattice of the combined context, instead calculating the necessary filled nodes via join operations on the factor lattices.

Results
Applying ConceptFlow to the ESC dataset revealed specific implications between voting patterns and musical characteristics:

Tempo and Cultural Support: Every winning song with a tempo of at least 150 BPM received cultural support.
Tempo, Key, and Regional/Cultural Support: Among winners receiving regional or cultural support, every song in a minor key had a tempo of at least 100 BPM. The diagram visually confirms this by the absence of a node for "slow minor-key songs" in those specific regions.
Complex Dependencies: Winners combining cultural support, a minor key, and a tempo of $\ge$ 150 BPM invariably also received political and historical support. The absence of a node representing this combination without the additional political/historical support attributes indicates a strict dependency in the dataset.

Significance and Claims
The paper positions this work as a demonstration of the ConceptFlow library's capabilities rather than a comprehensive statistical study of ESC voting behavior. The significance lies in the ability of nested line diagrams to "localize implications within particular nodes of the subdirect product." By decomposing the data into scales, the visualization makes relationships and dependencies between voting patterns and musical properties directly accessible, showing where specific musical combinations are possible and where they are excluded by voting conditions.

The authors note that future work will extend the library to include canonical implication bases and integrate them into the interactive interface, allowing users to inspect implications associated with concepts and unfilled regions directly.