Skip to main content

Implementing Full-Text Search for Bilingual Political Data

· 2 min read
samundrak
JavaScript Dev

Overview​

Search is a critical part of political platforms—especially when users are trying to find leaders, elections, or scandals across decades of data. I implemented a full-text search engine that:

  • Supports both English and Nepali
  • Works across multiple models
  • Handles fuzzy and partial input

This article details how I designed and deployed this system using PostgreSQL, Prisma, and language-aware search logic.

Models Supported​

Search spans multiple resource types:

  • Leaders
  • Parties
  • Elections
  • Governments
  • Content (scandals, comments, ratings)

Each model has fields like name, localName, title, titleLocal, and description indexed.

PostgreSQL Full-Text Indexing​

I used to_tsvector and tsquery over both English and Nepali fields.

Example SQL:

CREATE INDEX idx_leader_search ON leaders USING GIN (
  to_tsvector('simple', coalesce(name, '') || ' ' || coalesce(localName, ''))
);

Prisma doesn’t natively support fulltext search across multiple fields, so I used raw queries:

await prisma.$queryRaw`SELECT * FROM leaders WHERE to_tsvector('simple', name || ' ' || localName) @@ plainto_tsquery(${term})`

Language Detection​

On the frontend, I check user input and dynamically route the query to the correct tsvector config:

const lang = /[\u0900-\u097F]/.test(query) ? 'nepali' : 'english';

Synonyms + Transliterations​

I created a synonym map and fuzzy transliteration helper to allow Hindi → Nepali → Romanized matching. For example, "Prachanda" maps to "Pushpa Kamal Dahal".

Future work includes weighted ranking and stemming improvements.

Search API​

All results are returned with type, ID, and snippet. The frontend displays contextual matches:

{
  type: 'LEADER',
  id: 1023,
  name: 'Sher Bahadur Deuba',
  localName: 'शेर बहादुर देउवा',
  snippet: 'Nepali Congress | 5-time PM | Dadeldhura'
}

Summary​

Bilingual full-text search makes the platform usable across linguistic lines and political generations. With PostgreSQL GIN indexing, custom query handling, and fuzzy logic, users can now explore the archive effortlessly.

Next, I’ll cover how I structured the government timeline and prime ministerial records, including coalition breakdowns and ruling parties.