(所有這一切都是在PostgreSQL 9.0.1在Ubuntu 10.04,FWIW完成。)PostgreSQL的全文檢索隨機丟棄詞素
我試着寫我自己的lexize功能的PostgreSQL專門爲名稱的基礎上,雙metaphone算法。爲了生成C lexize函數,我開始使用dict_xsyn示例,並基本上將雙元metaphone交換爲同義詞查找。
但也許20%的時間to_tsvector
被調用,它似乎是下降的詞位。作爲一個例子,我創建了一個包含US Census Bureau's list of names的表格。
db=# select * from names order by rank limit 8;
name | freq | cumfreq | rank
----------+-------+---------+------
SMITH | 1.006 | 1.006 | 1
JOHNSON | 0.81 | 1.816 | 2
WILLIAMS | 0.699 | 2.515 | 3
JONES | 0.621 | 3.136 | 4
BROWN | 0.621 | 3.757 | 5
DAVIS | 0.48 | 4.237 | 6
MILLER | 0.424 | 4.66 | 7
WILSON | 0.339 | 5 | 8
(8 rows)
我們可以再增加一個矢量列,並與我的變音字典to_tsvector
填充它:
db=# alter table names add column vec tsvector;
ALTER TABLE
db=# update names set vec=to_tsvector('public.names', name);
UPDATE 88799
db=# select * from names order by rank limit 8;
name | freq | cumfreq | rank | vec
----------+-------+---------+------+-------------------------------
SMITH | 1.006 | 1.006 | 1 |
JOHNSON | 0.81 | 1.816 | 2 | 'ANSN':1 'JNSN':1 'johnson':1
WILLIAMS | 0.699 | 2.515 | 3 |
JONES | 0.621 | 3.136 | 4 |
BROWN | 0.621 | 3.757 | 5 |
DAVIS | 0.48 | 4.237 | 6 |
MILLER | 0.424 | 4.66 | 7 | 'MLR':1 'miller':1
WILSON | 0.339 | 5 | 8 | 'ALSN':1 'FLSN':1 'wilson':1
(8 rows)
的VEC領域的一串只是空!事實上:
db=# select count(1) from names where vec = to_tsvector('');
count
-------
41101
(1 row)
我能做到這一點批量更新,然後反覆計數不匹配的領域,每一次得到不同的數,因爲不同的行中不具有真正的計算載體。
但我覺得我的lexize函數本身可以嗎?
db=# alter table names drop column vec; alter table names add column lexemes varchar[];
ALTER TABLE
ALTER TABLE
db=# update names set lexemes=ts_lexize('dmetaphonedict', name);
UPDATE 88799
db=# select * from names order by rank limit 10;
name | freq | cumfreq | rank | lexemes
----------+-------+---------+------+----------------------
SMITH | 1.006 | 1.006 | 1 | {smith,SM0,XMT}
JOHNSON | 0.81 | 1.816 | 2 | {johnson,JNSN,ANSN}
WILLIAMS | 0.699 | 2.515 | 3 | {williams,ALMS,FLMS}
JONES | 0.621 | 3.136 | 4 | {jones,JNS,ANS}
BROWN | 0.621 | 3.757 | 5 | {brown,PRN}
DAVIS | 0.48 | 4.237 | 6 | {davis,TFS}
MILLER | 0.424 | 4.66 | 7 | {miller,MLR}
WILSON | 0.339 | 5 | 8 | {wilson,ALSN,FLSN}
MOORE | 0.312 | 5.312 | 9 | {moore,MR}
TAYLOR | 0.311 | 5.623 | 10 | {taylor,TLR}
(10 rows)
db=# select count(1) from names where array_length(lexemes,1)=0;
count
-------
0
(1 row)
我可以一遍又一遍地做到這一點,每次得到0次不匹配。
我創建的相關文本搜索字典和配置有:
CREATE OR REPLACE FUNCTION ddmetaphonedict_init(internal)
RETURNS internal
AS '$libdir/dict_dmetaphone'
LANGUAGE C STRICT;
CREATE OR REPLACE FUNCTION ddmetaphonedict_lexize(internal, internal, internal, internal)
RETURNS internal
AS '$libdir/dict_dmetaphone'
LANGUAGE C STRICT;
CREATE TEXT SEARCH TEMPLATE dmetaphonedict_template (
LEXIZE = ddmetaphonedict_lexize,
INIT = ddmetaphonedict_init
);
CREATE TEXT SEARCH DICTIONARY dmetaphonedict (
TEMPLATE = dmetaphonedict_template
);
COMMENT ON TEXT SEARCH DICTIONARY dmetaphonedict IS 'dictionary for names, using dmetaphone';
create text search configuration names (copy=english);
alter text search configuration names alter mapping for asciiword, asciihword, word, hword with dmetaphonedict;
差不多直出dict_xsyn例子。
發生了什麼事?我能做些什麼來弄清楚這些東西在哪裏被丟棄?
添加:剛試了9.1。 to_tsvector產生反衝tsvector的速率大大降低(5(僅5,而不是5k)到80000以上的7000條記錄)。但他們仍然在那裏。