CoolFace
Apppublic

dgeni2/HFThematicAnalysis

sourceHugging Facemitupdated 7mo agoView on Hugging Face
0likes
App README

HarmonyForge Thematic Analysis Dashboard

An interactive Streamlit dashboard that synthesizes findings from multiple AI-assisted thematic analysis methods (KriukowTA, ZambranoTA, and Braun-Clarke TA) to ground research questions for HarmonyForge—a Glass Box system that bridges the Repair Phase in music arrangement through axiomatic generation and LLM-driven explainability.

Overview

This dashboard presents comprehensive thematic analysis evidence from 21 musician interviews, revealing critical insights about part availability barriers, manual adaptation pain points, AI tool integration perceptions, and human-centric design requirements. The analysis addresses our main research question: To what extent does replacing probabilistic generation with deterministic logic and explainable critique enhance musician agency and trust when arranging lead sheets for non-standard ensembles?

Key Insights

Part availability creates logistical filters that exclude musicians and diminishes musical connection, while manual adaptation practices create Repair Phase bottlenecks requiring tools that reduce Mechanical Toil while preserving creative control. Musicians demand Edit-Authority and note-level control—they need a Jumping Pad (not full automation) that maintains Expressive Sovereignty, and trust requires Glass Box explainability with Red Line theory validation to build Deductive Trust rather than opaque Black Box outputs that increase skepticism.

Methodology

All qualitative analyses were conducted using Google Gemini as our LLM of choice, following established methodological frameworks:

  • —KriukowTA: Kriukow's AI-assisted thematic analysis method (originally demonstrated with ChatGPT)
  • —ZambranoTA: Zambrano et al.'s hybrid human-AI codebook development approach
  • —Braun-Clarke TA: Braun & Clarke's six-phase thematic analysis process

Human researchers maintained oversight, refinement, and validation throughout to ensure rigor and defensibility.

Features

  • —Cross-source analysis overview: Compare themes and evidence across all three analysis methods
  • —Comprehensive quote integration: Direct participant quotes organized by theme and source
  • —MoSCoW prioritization: Aggregated feature prioritization from all analyses
  • —Interactive visualizations: Plotly charts with dark theme styling
  • —Research question mapping: Clear connections between findings and research questions
  • —Raw data access: Full datasets available for further analysis

Data Structure

  • —KriukowTA/: Initial coding, focused coding, mapping & evidence, MoSCoW prioritization
  • —ZambranoTA/: Coded dataset, initial/refined codebooks, theme frequency, MoSCoW prioritization
  • —BraunClarkeTA/: Thematic analysis and MoSCoW prioritization
  • —GoogleForm/: Survey responses (quantitative and qualitative)

Usage

The dashboard automatically loads available data and presents it in organized tabs:

  • —Key Takeaways: Main insights, research questions, and cross-source analysis
  • —KriukowTA: Themes from user interviews using Kriukow's method
  • —ZambranoTA: Refined framework analysis
  • —Braun-Clarke TA: Thematic analysis validation
  • —Survey (Google Form): Quantitative and categorical survey responses

Each visualization includes question prompts explaining its purpose and how it answers research questions.

Research Questions

Main RQ: To what extent does replacing probabilistic generation with deterministic logic and explainable critique enhance musician agency and trust when arranging lead sheets for non-standard ensembles?

Sub-Questions:

  • —RQ1: How do limitations in part availability affect musicians' participation and satisfaction in ensembles?
  • —RQ2: What are the current manual practices and pain points for adapting music?
  • —RQ3: How do musicians perceive the integration of algorithmic/AI tools in the arrangement process?
  • —RQ4: What features and guardrails are required for a human-centric music arrangement tool?

Technical Details

  • —Built with Streamlit
  • —Visualizations using Plotly and Altair
  • —Dark theme optimized for readability
  • —Responsive design with interactive charts
  • —Docker deployment ready for Hugging Face Spaces

License

MIT