Rem Copyright (c) 1998, 2008, Oracle and/or its affiliates. All rights reserved.
Rem
Rem NAME
Rem dr0cls.pkh - Classification Training Service
Rem
Rem DESCRIPTION
Rem
Rem RETURNS
Rem
Rem NOTES
Rem
Rem MODIFIED (MM/DD/YY)
Rem shorwitz 08/28/08 - Bug 7237890
Rem daliao 09/02/03 - bug-3096033, no default preference
Rem daliao 01/13/03 - add clustering
Rem daliao 09/24/02 - add another generic training API.
Rem daliao 01/31/02 - daliao_classification
Rem daliao 10/2/2001 - Creation
Rem
CREATE OR REPLACE PACKAGE ctx_cls AUTHID current_user AS
/*------------------------------- TYPE DEFINITIONS ----------------------------*/
/* in-memory table for document assignment */
TYPE doc_rec IS RECORD (
docid number, -- document ID to identify the document
clusterid number, -- the ID of the cluster the document is assigned to
score number -- the similarity score between document and cluster
);
TYPE doc_tab is TABLE OF doc_rec INDEX BY BINARY_INTEGER;
/* in-memory table for cluster information */
TYPE cluster_rec IS RECORD (
clusterid number, -- cluster ID to identify a cluster
descript varchar2(4000), -- a string to describe the cluster
label varchar2(200), -- a suggested label for the cluster
sze number, -- number of documents assigned to the cluster
quality_score number, -- the quality score of the cluster
parent number -- parent cluster id. negative means no parent
);
TYPE cluster_tab IS TABLE OF cluster_rec INDEX BY BINARY_INTEGER;
TYPE docid_tab IS TABLE OF number INDEX BY BINARY_INTEGER;
/*------------------------------- train for ctx-rules --------------------------*/
/*
NAME
train - automatically generate ctx-rules from training examples
DESCRIPTION
This procedure will generate the ctx-rules for a given set of training
examples. The training examples are contained in the following two tables
table1: doctab must have the following columns:
docid number primary key
text doc. column which can be indexed by context index
table2: category table must have the following columns:
docid CONSTRAINT fk_id REFERENCES doctab(id)
category_id number
the foreign key in category table is recommended by not required.
The rules will be written to the result table specified.
The query table must have the following columns:
category_id number (the category_id)
query varchar2(4000) (the rule)
confidence number (the confidence level
(percentage) that a document
is relevant if this rule is
satisfied )
The names of table and column are not necessary the same as above.
ARGUMENTS
index_name - the name of the text index
docid - the name of docid column in document table
cattab - the name of category table
catdocid - the name of docid column in categroy table
catid - the name of category ID column in category table
restab - the name of result table
rescatid - the name of category ID column in result table
resquery - the name of query column in result table
resconfid - the name of confidence column in result table
pref_name - the name of preference
*/
PROCEDURE train (
index_name in varchar2,
docid in varchar2,
cattab in varchar2,
catdocid in varchar2,
catid in varchar2,
restab in varchar2,
rescatid in varchar2,
resquery in varchar2,
resconfid in varchar2,
pref_name in varchar2 DEFAULT NULL
);
/*------------------------------- generic train API ---------------------------*/
/*
NAME
train - automatically generate predicative model from examples for
classification
DESCRIPTION
This procedure will generate the predicative model from a given set of
training examples. The training examples are contained in the following
two tables
table1: doctab must have the following columns:
docid number primary key
text doc. column which can be indexed by context index
table2: category table must have the following columns:
docid number
category_id number
The names of table and column are not necessary the same as above.
The predicative model (classifier) will be written to the result tables.
The result table is either created by users before calling this function or
created in this program with the specified table name and under the current
user (if the specified table does not exist).
If user create the result table (which can support table schema for
different users), the table should have the following three columns with the
exact column names:
cat_id number
type number(3) not null
rule clob
ARGUMENTS
index_name - the name of the text index
docid - the name of docid column in document table
cattab - the name of category table
catdocid - the name of docid column in categroy table
catid - the name of category ID column in category table
restab - the name of generated result table
pref_name - the name of preference
*/
PROCEDURE train (
index_name in varchar2,
docid in varchar2,
cattab in varchar2,
catdocid in varchar2,
catid in varchar2,
restab in varchar2,
pref_name in varchar2
);
/* ---------- clustering API for permanent table result------------------------*/
/*
NAME
clustering - clustering a collection
DESCRIPTION
This procedure will generate a set of sub-group (clusters) from a provided
collection of documents. The collection is given by a table which having a
context index built with or without population. The collection table at least
has the following two collums, whose name may not be exactly the same.
docid number primary key
text doc. column which can be indexed by context index
The output of clustering is represented by two tables:
table 1: document membership table having the following collums with the
exact same names
docid number -- document ID to identify a document
clusterid number -- the ID of the cluster the document is assigned to
score number -- the similarity score between document and cluster
table 2: cluster description table having the following collums with the
exact same names
clusterid number -- cluster ID to identify a cluster
descript varchar2(4000) -- a string to describe the cluster
label varchar2(200) -- a suggested label for the cluster
size number -- number of documents assigned to the cluster
quality_score number -- the quality score of the cluster
parent number -- parent cluster id. negative means no parent
The output tables can either be created by users before calling this
function or created in this program with the specified table name and
under the current user (if the specified table does not exist).
ARGUMENTS:
index_name - the name of the text index
docid - the name of docid column in document table
doctab_name - the name of document membership table
clstab_name - the name of cluster description table
pref_name - the name of the preference
*/
PROCEDURE clustering (
index_name in varchar2,
docid in varchar2,
doctab_name in varchar2,
clstab_name in varchar2,
pref_name in varchar2 DEFAULT NULL
);
/* ---------- clustering API for in-memory table result------------------------*/
/*
NAME
clustering - clustering a collection
DESCRIPTION
This procedure will generate a set of sub-group (clusters) from a provided
collection of documents. The collection is given by a table which having a
context index built with or without population. The collection table at least
has the following two collums, whose name may not be exactly the same.
docid number primary key
text doc. column which can be indexed by context index
The output of clustering is represented by two in-memory tables:
table 1: document membership table ctx_cls.doc_tab
table 2: cluster description table ctx_cls.cluster_tab
ARGUMENTS:
index_name - the name of the text index
docid - the name of docid column in document table
dids - docid list to be clustered
doctab_name - the name of document membership table
clstab_name - the name of cluster description table
pref_name - the name of the preference
*/
PROCEDURE clustering (
index_name in varchar2,
docid in varchar2,
dids in docid_tab,
doctab_name in out nocopy doc_tab,
clstab_name in out nocopy cluster_tab,
pref_name in varchar2 DEFAULT NULL
);
END ctx_cls;
/