From b58e24cc3c328e0577e6a8a44889d3639c0b289e Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Tue, 13 Jun 2017 10:16:31 +0100 Subject: Tweak the ranking of PG user dir search --- synapse/storage/user_directory.py | 53 +++++++++++++++++++++++++++++---------- 1 file changed, 40 insertions(+), 13 deletions(-) (limited to 'synapse/storage') diff --git a/synapse/storage/user_directory.py b/synapse/storage/user_directory.py index 6a4bf63f0d..0b874e0227 100644 --- a/synapse/storage/user_directory.py +++ b/synapse/storage/user_directory.py @@ -391,11 +391,14 @@ class UserDirectoryStore(SQLBaseStore): ] } """ - - search_query = _parse_query(self.database_engine, search_term) - if isinstance(self.database_engine, PostgresEngine): + full_query, exact_query, prefix_query = _parse_query_postgres(search_term) + # We order by rank and then if they have profile info + # The ranking algorithm is hand tweaked for "best" results. Broadly + # the idea is we give a higher weight to exact matches. + # The array of numbers are the weights for the various part of the + # search: (domain, _, display name, localpart) sql = """ SELECT user_id, display_name, avatar_url FROM user_directory_search @@ -403,13 +406,27 @@ class UserDirectoryStore(SQLBaseStore): INNER JOIN users_in_pubic_room USING (user_id) WHERE vector @@ to_tsquery('english', ?) ORDER BY - ts_rank_cd(vector, to_tsquery('english', ?), 1) DESC, + 2 * ts_rank_cd( + '{0.1, 0.1, 0.9, 1.0}', + vector, + to_tsquery('english', ?), + 8 + ) + + ts_rank_cd( + '{0.1, 0.1, 0.9, 1.0}', + vector, + to_tsquery('english', ?), + 8 + ) + DESC, display_name IS NULL, avatar_url IS NULL LIMIT ? """ - args = (search_query, search_query, limit + 1,) + args = (full_query, exact_query, prefix_query, limit + 1,) elif isinstance(self.database_engine, Sqlite3Engine): + search_query = _parse_query_sqlite(search_term) + sql = """ SELECT user_id, display_name, avatar_url FROM user_directory_search @@ -439,7 +456,7 @@ class UserDirectoryStore(SQLBaseStore): }) -def _parse_query(database_engine, search_term): +def _parse_query_sqlite(search_term): """Takes a plain unicode string from the user and converts it into a form that can be passed to database. We use this so that we can add prefix matching, which isn't something @@ -451,11 +468,21 @@ def _parse_query(database_engine, search_term): # Pull out the individual words, discarding any non-word characters. results = re.findall(r"([\w\-]+)", search_term, re.UNICODE) + return " & ".join("(%s* | %s)" % (result, result,) for result in results) + + +def _parse_query_postgres(search_term): + """Takes a plain unicode string from the user and converts it into a form + that can be passed to database. + We use this so that we can add prefix matching, which isn't something + that is supported by default. + """ + + # Pull out the individual words, discarding any non-word characters. + results = re.findall(r"([\w\-]+)", search_term, re.UNICODE) + + both = " & ".join("(%s:* | %s)" % (result, result,) for result in results) + exact = " & ".join("%s" % (result,) for result in results) + prefix = " & ".join("%s:*" % (result,) for result in results) - if isinstance(database_engine, PostgresEngine): - return " & ".join("(%s:* | %s)" % (result, result,) for result in results) - elif isinstance(database_engine, Sqlite3Engine): - return " & ".join("(%s* | %s)" % (result, result,) for result in results) - else: - # This should be unreachable. - raise Exception("Unrecognized database engine") + return both, exact, prefix -- cgit 1.4.1