Merge pull request #3995 from Itz-Agasta/exp

Add Language-aware country penalty in forward geocoding
This commit is contained in:
Sarah Hoffmann
2026-03-04 17:24:46 +01:00
committed by GitHub
4 changed files with 46 additions and 7 deletions

View File

@@ -170,11 +170,20 @@ class ForwardGeocoder:
if qword not in words: if qword not in words:
wdist = max(difflib.SequenceMatcher(a=qword, b=w).quick_ratio() for w in words) wdist = max(difflib.SequenceMatcher(a=qword, b=w).quick_ratio() for w in words)
distance += len(qword) if wdist < 0.4 else 1 distance += len(qword) if wdist < 0.4 else 1
# Compensate for the fact that country names do not get a # Countries with high importance can dominate results when matched
# match penalty yet by the tokenizer. # via an alternate-language name. Apply a language-aware penalty
# Temporary hack that needs to be removed! # to offset this.
if result.rank_address == 4: if result.rank_address == 4:
distance *= 2 if self.params.locales and result.names:
loc_names = [result.names[t] for t in self.params.locales.name_tags
if t in result.names]
if loc_names:
norm_loc = self.query_analyzer.normalize_text(' '.join(loc_names))
loc_words = set(w for w in re.split('[-,: ]+', norm_loc) if w)
if loc_words and loc_words.isdisjoint(qwords):
result.accuracy += result.calculated_importance() * 0.5
else:
distance *= 2
result.accuracy += distance * 0.3 / sum(len(w) for w in qwords) result.accuracy += distance * 0.3 / sum(len(w) for w in qwords)
async def lookup_pois(self, categories: List[Tuple[str, str]], async def lookup_pois(self, categories: List[Tuple[str, str]],

View File

@@ -8,7 +8,7 @@
Complex datatypes used by the Nominatim API. Complex datatypes used by the Nominatim API.
""" """
from typing import Optional, Union, Tuple, NamedTuple, TypeVar, Type, Dict, \ from typing import Optional, Union, Tuple, NamedTuple, TypeVar, Type, Dict, \
Any, List, Sequence Any, List, Sequence, TYPE_CHECKING
from collections import abc from collections import abc
import dataclasses import dataclasses
import datetime as dt import datetime as dt
@@ -17,6 +17,8 @@ import math
from struct import unpack from struct import unpack
from binascii import unhexlify from binascii import unhexlify
if TYPE_CHECKING:
from .localization import Locales
from .errors import UsageError from .errors import UsageError
@@ -573,6 +575,13 @@ class SearchDetails(LookupDetails):
viewbox_x2: Optional[Bbox] = None viewbox_x2: Optional[Bbox] = None
locales: Optional['Locales'] = dataclasses.field(
default=None, metadata={'transform': lambda v: v})
""" Locale preferences of the caller.
Used during result re-ranking to prefer results that match the
caller's locale over results that only match in an alternate language.
"""
def __post_init__(self) -> None: def __post_init__(self) -> None:
if self.viewbox is not None: if self.viewbox is not None:
xext = (self.viewbox.maxlon - self.viewbox.minlon)/2 xext = (self.viewbox.maxlon - self.viewbox.minlon)/2

View File

@@ -334,6 +334,8 @@ async def search_endpoint(api: NominatimAPIAsync, params: ASGIAdaptor) -> Any:
details['layers'] = DataLayer.ADDRESS details['layers'] = DataLayer.ADDRESS
else: else:
details['layers'] = get_layers(params) details['layers'] = get_layers(params)
details['locales'] = Locales.from_accept_languages(get_accepted_languages(params),
params.config().OUTPUT_NAMES)
# unstructured query parameters # unstructured query parameters
query = params.get('q', None) query = params.get('q', None)
@@ -359,8 +361,7 @@ async def search_endpoint(api: NominatimAPIAsync, params: ASGIAdaptor) -> Any:
except UsageError as err: except UsageError as err:
params.raise_error(str(err)) params.raise_error(str(err))
Locales.from_accept_languages(get_accepted_languages(params), details['locales'].localize_results(results)
params.config().OUTPUT_NAMES).localize_results(results)
if details['dedupe'] and len(results) > 1: if details['dedupe'] and len(results) > 1:
results = helpers.deduplicate_results(results, max_results) results = helpers.deduplicate_results(results, max_results)

View File

@@ -80,3 +80,23 @@ Feature: Searching of simple objects
| Chicago | Illinois | IL | | Chicago | Illinois | IL |
| Auburn | Alabama | AL | | Auburn | Alabama | AL |
| New Orleans | Louisiana | LA | | New Orleans | Louisiana | LA |
# github #3210
Scenario: Country with alternate-language name does not dominate when locale differs
Given the 1.0 grid with origin DE
| 1 | | 2 |
| | 10 | |
| 4 | | 3 |
Given the places
| osm | class | type | admin | name+name | name+name:fi | name+name:de | country | geometry |
| R1 | boundary | administrative | 2 | Turgei | Turgi | Testland | de | (1,2,3,4,1) |
Given the places
| osm | class | type | name+name | geometry |
| N10 | place | village | Turgi | 10 |
When importing
And geocoding "Turgi"
| accept-language |
| de |
Then result 0 contains
| object |
| N10 |