icu tokenizer: switch to matching against partial names

When matching address parts from addr:* tags against place names,
the address names where so far converted to full names and compared
those to the place names. This can become problematic with the new
ICU tokenizer once we introduce creation of different variants
depending on the place name context. It wouldn't be clear which
variant to produce to get a match, so we would have to create all of
them. To work around this issue, switch to using the partial terms
for matching. This introduces a larger fuzziness between matches but
that shouldn't be a problem because matching is always geographically
restricted.

The search terms created for address parts have a different problem:
they are already created before we even know if they are going to be
used. This can lead to spurious entries in the word table, which slows
down searching. This problem can also be circumvented by using only
partial terms for the search terms. In terms of searching that means
that the address terms would not get the full-word boost, but given
that the case where an address part does not exist as an OSM object
should be the exception, this is likely acceptable.

lib-sql/tokenizer/icu_tokenizer.sql

@@ -44,28 +44,28 @@ $$ LANGUAGE SQL IMMUTABLE;
 CREATE OR REPLACE FUNCTION token_has_addr_place(info JSONB)
   RETURNS BOOLEAN
 AS $$
-  SELECT info->>'place_match' is not null;
+  SELECT info->>'place' is not null;
 $$ LANGUAGE SQL IMMUTABLE;
 
 
 CREATE OR REPLACE FUNCTION token_matches_street(info JSONB, street_tokens INTEGER[])
   RETURNS BOOLEAN
 AS $$
-  SELECT (info->>'street')::INTEGER[] && street_tokens
+  SELECT (info->>'street')::INTEGER[] <@ street_tokens
 $$ LANGUAGE SQL IMMUTABLE STRICT;
 
 
 CREATE OR REPLACE FUNCTION token_matches_place(info JSONB, place_tokens INTEGER[])
   RETURNS BOOLEAN
 AS $$
-  SELECT (info->>'place_match')::INTEGER[] && place_tokens
+  SELECT (info->>'place')::INTEGER[] <@ place_tokens
 $$ LANGUAGE SQL IMMUTABLE STRICT;
 
 
 CREATE OR REPLACE FUNCTION token_addr_place_search_tokens(info JSONB)
   RETURNS INTEGER[]
 AS $$
-  SELECT (info->>'place_search')::INTEGER[]
+  SELECT (info->>'place')::INTEGER[]
 $$ LANGUAGE SQL IMMUTABLE STRICT;
 
 
@@ -79,14 +79,14 @@ $$ LANGUAGE SQL IMMUTABLE STRICT;
 CREATE OR REPLACE FUNCTION token_get_address_search_tokens(info JSONB, key TEXT)
   RETURNS INTEGER[]
 AS $$
-  SELECT (info->'addr'->key->>0)::INTEGER[];
+  SELECT (info->'addr'->>key)::INTEGER[];
 $$ LANGUAGE SQL IMMUTABLE STRICT;
 
 
 CREATE OR REPLACE FUNCTION token_matches_address(info JSONB, key TEXT, tokens INTEGER[])
   RETURNS BOOLEAN
 AS $$
-  SELECT (info->'addr'->key->>1)::INTEGER[] && tokens;
+  SELECT (info->'addr'->>key)::INTEGER[] <@ tokens;
 $$ LANGUAGE SQL IMMUTABLE STRICT;
 
 
@@ -146,15 +146,34 @@ BEGIN
         VALUES (term_id, term, 'w', json_build_object('count', term_count));
     END IF;
 
-    IF term_count < {{ max_word_freq }} THEN
     partial_tokens := array_merge(partial_tokens, ARRAY[term_id]);
-    END IF;
   END LOOP;
 END;
 $$
 LANGUAGE plpgsql;
 
 
+CREATE OR REPLACE FUNCTION getorcreate_partial_word(partial TEXT)
+  RETURNS INTEGER
+  AS $$
+DECLARE
+  token INTEGER;
+BEGIN
+  SELECT min(word_id) INTO token
+    FROM word WHERE word_token = partial and type = 'w';
+
+  IF token IS NULL THEN
+    token := nextval('seq_word');
+    INSERT INTO word (word_id, word_token, type, info)
+        VALUES (token, partial, 'w', json_build_object('count', 0));
+  END IF;
+
+  RETURN token;
+END;
+$$
+LANGUAGE plpgsql;
+
+
 CREATE OR REPLACE FUNCTION getorcreate_hnr_id(lookup_term TEXT)
   RETURNS INTEGER
   AS $$

nominatim/tokenizer/icu_tokenizer.py

@@ -17,7 +17,6 @@ from nominatim.tokenizer.icu_rule_loader import ICURuleLoader
 from nominatim.tokenizer.icu_name_processor import ICUNameProcessor, ICUNameProcessorRules
 from nominatim.tokenizer.base import AbstractAnalyzer, AbstractTokenizer
 
-DBCFG_MAXWORDFREQ = "tokenizer_maxwordfreq"
 DBCFG_TERM_NORMALIZATION = "tokenizer_term_normalization"
 
 LOG = logging.getLogger()
@@ -39,7 +38,6 @@ class LegacyICUTokenizer(AbstractTokenizer):
         self.data_dir = data_dir
         self.naming_rules = None
         self.term_normalization = None
-        self.max_word_frequency = None
 
 
     def init_new_db(self, config, init_db=True):
@@ -52,7 +50,6 @@ class LegacyICUTokenizer(AbstractTokenizer):
                                                              config='TOKENIZER_CONFIG'))
         self.naming_rules = ICUNameProcessorRules(loader=loader)
         self.term_normalization = config.TERM_NORMALIZATION
-        self.max_word_frequency = config.MAX_WORD_FREQUENCY
 
         self._install_php(config.lib_dir.php)
         self._save_config(config)
@@ -68,7 +65,6 @@ class LegacyICUTokenizer(AbstractTokenizer):
         with connect(self.dsn) as conn:
             self.naming_rules = ICUNameProcessorRules(conn=conn)
             self.term_normalization = get_property(conn, DBCFG_TERM_NORMALIZATION)
-            self.max_word_frequency = get_property(conn, DBCFG_MAXWORDFREQ)
 
 
     def finalize_import(self, _):
@@ -81,10 +77,8 @@ class LegacyICUTokenizer(AbstractTokenizer):
         """ Reimport the SQL functions for this tokenizer.
         """
         with connect(self.dsn) as conn:
-            max_word_freq = get_property(conn, DBCFG_MAXWORDFREQ)
             sqlp = SQLPreprocessor(conn, config)
-            sqlp.run_sql_file(conn, 'tokenizer/icu_tokenizer.sql',
+            sqlp.run_sql_file(conn, 'tokenizer/icu_tokenizer.sql')
-                              max_word_freq=max_word_freq)
 
 
     def check_database(self):
@@ -122,7 +116,7 @@ class LegacyICUTokenizer(AbstractTokenizer):
         php_file = self.data_dir / "tokenizer.php"
         php_file.write_text(dedent(f"""\
             <?php
-            @define('CONST_Max_Word_Frequency', {self.max_word_frequency});
+            @define('CONST_Max_Word_Frequency', 10000000);
             @define('CONST_Term_Normalization_Rules', "{self.term_normalization}");
             @define('CONST_Transliteration', "{self.naming_rules.search_rules}");
             require_once('{phpdir}/tokenizer/icu_tokenizer.php');"""))
@@ -135,7 +129,6 @@ class LegacyICUTokenizer(AbstractTokenizer):
         with connect(self.dsn) as conn:
             self.naming_rules.save_rules(conn)
 
-            set_property(conn, DBCFG_MAXWORDFREQ, config.MAX_WORD_FREQUENCY)
             set_property(conn, DBCFG_TERM_NORMALIZATION, self.term_normalization)
 
 
@@ -424,12 +417,12 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
             elif key in ('housenumber', 'streetnumber', 'conscriptionnumber'):
                 hnrs.append(value)
             elif key == 'street':
-                token_info.add_street(*self._compute_name_tokens({'name': value}))
+                token_info.add_street(self._compute_partial_tokens(value))
             elif key == 'place':
-                token_info.add_place(*self._compute_name_tokens({'name': value}))
+                token_info.add_place(self._compute_partial_tokens(value))
             elif not key.startswith('_') and \
                  key not in ('country', 'full'):
-                addr_terms.append((key, *self._compute_name_tokens({'name': value})))
+                addr_terms.append((key, self._compute_partial_tokens(value)))
 
         if hnrs:
             hnrs = self._split_housenumbers(hnrs)
@@ -438,6 +431,32 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         if addr_terms:
             token_info.add_address_terms(addr_terms)
 
+    def _compute_partial_tokens(self, name):
+        """ Normalize the given term, split it into partial words and return
+            then token list for them.
+        """
+        norm_name = self.name_processor.get_search_normalized(name)
+
+        tokens = []
+        need_lookup = []
+        for partial in norm_name.split():
+            token = self._cache.partials.get(partial)
+            if token:
+                tokens.append(token)
+            else:
+                need_lookup.append(partial)
+
+        if need_lookup:
+            with self.conn.cursor() as cur:
+                cur.execute("""SELECT word, getorcreate_partial_word(word)
+                               FROM unnest(%s) word""",
+                            (need_lookup, ))
+
+                for partial, token in cur:
+                    tokens.append(token)
+                    self._cache.partials[partial] = token
+
+        return tokens
 
     def _compute_name_tokens(self, names):
         """ Computes the full name and partial name tokens for the given
@@ -551,30 +570,25 @@ class _TokenInfo:
         self.data['hnr'] = ';'.join(hnrs)
 
 
-    def add_street(self, fulls, _):
+    def add_street(self, tokens):
         """ Add addr:street match terms.
         """
-        if fulls:
+        if tokens:
-            self.data['street'] = self._mk_array(fulls)
+            self.data['street'] = self._mk_array(tokens)
 
 
-    def add_place(self, fulls, partials):
+    def add_place(self, tokens):
         """ Add addr:place search and match terms.
         """
-        if fulls:
+        if tokens:
-            self.data['place_search'] = self._mk_array(itertools.chain(fulls, partials))
+            self.data['place'] = self._mk_array(tokens)
-            self.data['place_match'] = self._mk_array(fulls)
 
 
     def add_address_terms(self, terms):
         """ Add additional address terms.
         """
-        tokens = {}
+        tokens = {key: self._mk_array(partials)
-
+                  for key, partials in terms if partials}
-        for key, fulls, partials in terms:
-            if fulls:
-                tokens[key] = [self._mk_array(itertools.chain(fulls, partials)),
-                               self._mk_array(fulls)]
 
         if tokens:
             self.data['addr'] = tokens
@@ -588,6 +602,7 @@ class _TokenCache:
     """
     def __init__(self):
         self.names = {}
+        self.partials = {}
         self.postcodes = set()
         self.housenumbers = {}
 

test/bdd/db/import/search_name.feature

@@ -125,9 +125,6 @@ Feature: Creation of search terms
         Then placex contains
          | object | parent_place_id |
          | N1     | N2              |
-        Then search_name contains
-         | object | name_vector | nameaddress_vector |
-         | N1     | #Walltown   | Strange, Town |
         When sending search query "23 Rose Street"
         Then exactly 1 results are returned
         And results contain
@@ -156,9 +153,6 @@ Feature: Creation of search terms
          | W1  | highway | residential | Rose Street  | :w-north |
          | N2  | place   | city        | Strange Town | :p-N1 |
         When importing
-        Then search_name contains
-         | object | name_vector      | nameaddress_vector      |
-         | N1     | #Walltown, #Blue house | Walltown, Strange, Town |
         When sending search query "23 Walltown, Strange Town"
         Then results contain
          | osm | display_name |
@@ -190,9 +184,6 @@ Feature: Creation of search terms
          | W1  | highway | residential | Rose Street  | :w-north |
          | N2  | place   | city        | Strange Town | :p-N1 |
         When importing
-        Then search_name contains
-         | object | name_vector      | nameaddress_vector      |
-         | N1     | #Moon sun, #Blue house | Moon, Sun, Strange, Town |
         When sending search query "23 Moon Sun, Strange Town"
         Then results contain
          | osm | display_name |
@@ -212,9 +203,6 @@ Feature: Creation of search terms
          | W1  | highway | residential | Rose Street  | Walltown  | :w-north |
          | N2  | place   | suburb      | Strange Town | Walltown  | :p-N1 |
         When importing
-        Then search_name contains
-         | object | name_vector | nameaddress_vector |
-         | N1     | #Walltown   | Strange, Town      |
         When sending search query "23 Rose Street, Walltown"
         Then exactly 1 result is returned
         And results contain
@@ -303,9 +291,6 @@ Feature: Creation of search terms
          | W1  | highway | residential | Rose Street  | :w-north |
          | N2  | place   | suburb      | Strange Town | :p-N1 |
         When importing
-        Then search_name contains
-         | object | name_vector | nameaddress_vector |
-         | N1     | #Green Moss | Walltown |
         When sending search query "Green Moss, Rose Street, Walltown"
         Then exactly 0 result is returned
         When sending search query "Green Moss, Walltown"