flights-nlp 0.1.3

Natural language flight search parser — extract structured flight data from freeform text

flights-nlp ✈️

A robust, Django-friendly natural language flight search query parser for English and Japanese. It translates freeform text queries into structured parameters ready for flight search APIs (like Amadeus, Sabre, or Skyscanner).

Features

• 🧠 Hybrid Route Extraction: Combines spaCy NLP models (en_core_web_sm/ja_core_news_sm) with geographic database lookups to reliably extract origin and destination cities/airports, resolving sequence ambiguities.
• 🇯🇵 Spaceless Japanese Support: Fully parses spaceless Japanese query layouts (e.g., 東京から札幌明日片道).
• 📍 Multi-Airport Cities (MAC): Resolves city and airport names (e.g., 羽田, 成田) to their IATA codes and ties them back to parent metropolitan codes (e.g., TYO).
• 🔍 Airport Disambiguation: Extracts parenthesized hints like 大阪(伊丹) to target specific airports (e.g., ITM) instead of whole metropolitan areas.
• 📅 Relative Dates & Durations: Parses relative terms (today, tomorrow, next week, 来週の月曜日), relative offsets (3日後, 2週間後), and trip durations (for a week, 1週間, 3日間).
• 👥 Passenger Parsing: Normalizes numbers and counts for adults, children, and infants (e.g., 大人二人, 2 adults 1 infant).
• 🛡️ Django Ready: Fully optimized for web frameworks with startup model pre-loading (preload_models), custom exception classes, and standard Python logging hooks.

---

Installation

Requires Python 3.9+. Install via pip or uv:

pip install flights-nlp

SpaCy language pipelines (en_core_web_sm / ja_core_news_sm) are downloaded automatically on first parse (or when you call preload_models). You can also install them manually:

python -m spacy download en_core_web_sm
python -m spacy download ja_core_news_sm

---

Quick Start (Python API)

Basic Parsing

from datetime import date
from flights_nlp import parse

# Parse a query relative to a reference date
result = parse("Tokyo to Sapporo next week 3 adults", reference_date=date(2026, 7, 12))

print(result.origin)               # "Tokyo"
print(result.origin_iata)          # ["NRT", "HND"]
print(result.origin_city_code)     # "TYO"
print(result.destination)          # "Sapporo"
print(result.destination_iata)     # ["CTS", "OKD"]
print(result.departure_date)       # 2026-07-20
print(result.adults)               # 3
print(result.trip_type)            # "round_trip"

Serializing to JSON

print(result.to_json(indent=2))

Django Startup Warmup (Model Pre-loading)

To prevent the initial web request from experiencing a ~2 second model loading latency, call preload_models inside your Django app's startup hook:

# myapp/apps.py
from django.apps import AppConfig
from flights_nlp import preload_models

class MyAppConfig(AppConfig):
    name = "myapp"
    
    def ready(self):
        # Preload English and Japanese models
        preload_models(["en", "ja"])

Custom Exception Handling

The library raises structured subclasses of FlightsNLPError (which inherits from ValueError):

• RouteExtractionError: Raised when origin/destination cannot be resolved.
• DateResolutionError: Raised when date formats are invalid.
• PassengerParsingError: Raised when passenger parameters are faulty.

---

Command Line Interface (CLI)

The package installs a CLI tool flights-nlp to parse queries directly from the terminal.

JSON Output (Default)

flights-nlp "Tokyo to Sapporo next week 3 adults"

Table Output

flights-nlp "東京から札幌明日片道" --format table

Output:

┌────────────────┬─────────────────────────┐
│     Field      │          Value          │
├────────────────┼─────────────────────────┤
│ Origin         │ Tokyo [TYO] (NRT,HND)   │
│ Destination    │ Sapporo [SPK] (CTS,OKD) │
│ Departure Date │ 2026-07-13              │
│ Return Date    │ —                       │
│ Trip Type      │ one_way                 │
│ Adults         │ 1                       │
│ Children       │ 0                       │
│ Infants        │ 0                       │
└────────────────┴─────────────────────────┘

---

Interactive Playground (Streamlit)

For developer testing, run the Streamlit visual debugger locally:

uv run streamlit run app.py

This starts a dashboard where you can type queries, inspect spaCy tokenization / entity tags, check JSON representations, and test various reference dates reactively.

---

Running Tests

To run the full unit test suite:

pytest tests/