# Spotify Listening History Analysis

14 years of streaming history — 10,133 hours, 147,594 streams, 15,748 unique artists. Taste galaxy, artist loyalty scatter plots, and the sharp take on what the data actually says about who I am.

## The Data

Spotify’s extended streaming history goes back to your first play. Mine starts in 2012 — 10,133 hours, 147,594 streams, 15,748 unique artists across 14 years. 422 continuous days of listening. Here’s what the data says about who I am.

Each record has a timestamp, play duration, track URI, skip flag, shuffle flag, platform, and whether it was offline. It is enough to reconstruct what I listened to and _how_ I listened.

## What I Built

A Python + DuckDB analysis pipeline that ingests all the raw JSON files, deduplicates, cleans edge cases, and outputs a set of explorable views.

### Key Findings

**Top artist all-time:** Mac Miller — by total minutes, by years present, by genre diversity.

**Skip rate as signal:** Songs I’ve never skipped once in 500+ plays are a fundamentally different kind of favorite than songs I’ve played 1,000 times but skipped 40% of the time. The “never-skip” list is the real canon.

**Obsession cycles:** I listen in phases. There’s a clear pattern of 6–8 week immersions into a single artist, followed by a reset. I can trace every relationship, job change, and major life event to a soundtrack change.

**Listening by platform:** Desktop vs. mobile vs. offline reveal different moods. Offline listening is almost always music I trust. Desktop listening skews to new exploration.

**Discovery vs. catalog ratio:** The share of first-time listens has dropped steadily since 2021 — I’m settling into a catalog instead of exploring. Concerning.

## Tech Stack

- **Python 3.9** — ingestion, cleaning, analysis
- **DuckDB 1.4.4** — in-process analytics, fast joins across 67K records
- **pandas** — dataframe manipulation
- **matplotlib + seaborn** — chart generation
- **Parquet** — processed data persistence

## The Hard Parts

**Deduplication:** Spotify double-logs plays that span offline/online sync transitions. Same track, same timestamp, different `offline_timestamp`. Deduplicated on `(ts, track_uri, ms_played)`.

**Track identity:** The same song re-released on three different albums has three different URIs. Fell back to `(artist_name, track_name)` grouping for anything where URI-level precision caused fragmentation.

**Minimum play threshold:** 30 seconds. Below that it’s a skip regardless of what the `reason_end` field says.

**Podcast contamination:** All podcast and audiobook entries have null track fields. Filtered out for music analysis, kept separate for potential future work.

## Concert Connection

The concert history analysis used the same Spotify data as one of three corroboration sources — scrobble timestamps verified against Google Calendar events and email receipts to reconstruct 178 live shows since 2007.