Research: genre taxonomy and relationships #39

Goal

Investigate how to model relationships between genre tags so that search, filtering, and tagging can be smarter — for example, understanding that sci-fi and fantasy are related, or that action commonly co-occurs with adventure.

Background

The current genre model is a flat array of tags per quote (e.g. ["action", "sci-fi"]). This is simple and works for exact-match filtering. It has two limitations:

1. No hierarchy — a user filtering for fiction does not get sci-fi or fantasy results, even though those are subsets of fiction.
2. No relationships — there is no way to suggest related genres or expand a search to nearby tags.

As the genre vocabulary grows (TMDB alone adds crime, documentary, family, history, war, western, kids, reality, politics, music, tv-movie to the existing set), a flat model becomes harder to navigate.

Questions to answer

Hierarchy

Is a genre hierarchy useful? Candidate parent–child relationships:

• Is a two-level hierarchy (parent + leaf) enough, or do we need arbitrary depth?
• Should the hierarchy live in a config file, a DB table, or be hardcoded in the filter logic?

Relationships (non-hierarchical)

Beyond hierarchy, are there useful sibling/affinity relationships?

• sci-fi ↔ fantasy (frequently co-tagged, often searched together)

• thriller ↔ mystery (overlapping audience)

• action ↔ adventure

• Should related genres be stored explicitly (a relationship table/map), or derived from co-occurrence in the dataset?

• Is co-occurrence analysis on the existing 780+ quotes sufficient to produce meaningful relationships, or is the dataset too small?

Impact on search and filtering

• Should the /search endpoint accept a genre filter that automatically expands to children? (e.g. ?genre=fiction returns quotes tagged sci-fi, fantasy, etc.)
• Should the Blazor UI genre filter show a grouped/hierarchical picker rather than a flat list?
• Should the MCP tool expose genre relationships to help an AI assistant suggest related quotes?

Data storage

• A genre relationship map could be a static JSON file in scripts/ or data/, a table in SQLite, or hardcoded constants in Quotinator.Constants.
• What is the lowest-maintenance option that still allows the relationships to be updated without a code deploy?

Output

A short design note (can be a comment on this issue) covering:

• Whether a hierarchy, affinity relationships, or both are worth implementing
• Proposed storage approach
• Impact on the /search endpoint and Blazor filter UI
• Whether co-occurrence analysis on the existing dataset is a useful starting point

This feeds into future issues for the search endpoint, Blazor UI genre filter, and potentially the MCP tool.