IR expert - Parameter to alter functionality ?

[robbe2]

Hi,

can anyone help me, I have a question about the IR Expert functionality of service center 5(.1).

Can the way the search algoritm works be altered ?
- by the use of a parameter ?
- by the use of a command line in the ini file ?

I know there exists a parameter that can be added to the ini file that can change the behaviour of the IR Expert results so that it doesn't compare the order of the words used in the description to sort the results found.

But, are there other parameters to "fine tune" IR expert.

Thanks in advance!!!

Robbe

 

[williamwhite]

Hello,

From the data administration guide. There is the info available:

Down here there is an extraction of this document pages 552-555 but you want to look From 549 to 561
the whole info is there


28-4 Data Administration Guide
Start-up Parameters
This section explains system-level start-up parameters governing the
operation of IR Expert. Parameters that are platform-dependent are clearly
marked.
ir_cluster_closeness
syntax: ir_cluster_closeness: n
IR Expert record similarity variance allowed, expressed as a percentage. As
the percentage increases, clusters become larger and more loosely related.
Default is 20.
ir_cluster_symbol
syntax: ir_cluster_symbol: *
The symbol used to invoke clustering.
ir_language
syntax: ir_language:language
Currently, built-in support is available for language options english and
german. Default is english.
Refer to File Management, page 28-7 for more information.
ir_languagefiles_path
syntax: ir_languagefiles_pathathspec
Where the IR language files reside. Language files include the stop words,
stem dictionary, suffix dictionary, and normal dictionary files. Refer to File
Management, page 28-7 for more information.
UNIX - The path to the IR language files. If you omit this option, the language
files must reside in the directory from which ServiceCenter executes.
MVS - The prefix of the IR language datasets. This option is required for OS/
390.
ir_max_clusters
syntax: ir_max_clusters:n
Maximum number of clusters returned. Default is 100.
IR Expert Technical Reference 28-5
ir_max_deep_distance
syntax: ir_max_deep_distance:n
Maximum number of insertions, deletions, or substitutions allowed during
automatic spelling correction. Default is 2.
ir_max_relevant_answers
syntax: ir_max_relevant_answers:n
Maximum number of relevant records returned by an IR Expert query.
Default is 50.
ir_max_shallow_distance
syntax: ir_max_shallow_distance:n
Maximum number of different letters allowed during automatic spelling
correction. Default is 1.
ir_min_cluster_members
syntax: ir_min_cluster_members:n
Least number of records allowed in any one cluster. Default is 3.
ir_prefix
syntax: ir_prefixathspec
Where the IR database files reside.
UNIX -The path to the IR database files. If you omit this option, IR Expert
files *.dc and *.map must reside in the directory from which ServiceCenter
executes.
MVS -The prefix of the IR database datasets. This option is required for OS/
390.
ir_query_drop_off
syntax: ir_query_drop_off:n
Record energy drop off allowed, expressed as a percentage. As this percentage
increases, IR Expert returns more loosely-related records. Default is 50.
ir_volser
syntax: ir_volser:volser
UNIX -Not used.
MVS -The volume serial number of the IR database datasets. This option is
required for OS/390.
ir_boost_same_sequence
syntax: ir_boost_same_sequence
This must be in the INI file to enable the PHRASE support
ir_term_drop_off:22
syntax: ir_term_drop_off:22
This parameter sets the minimum percentage (IDF value) of documents a
term can appear in before it is ignored in the IR Search. The goal is eliminate
those terms which will not contribute qualified leads/documents of interest.
Terms used in a large amount of documents provide little value in attempting
to isolate a record that matches the query. This parameter allows a term to be
logically in the stop list file once the percentage of documents it appears in is
exceeded.
For example, if you have a file containing 1000 documents (TDOC) and a term
in your query statement appears in 10 of those documents (TFREQ), the IDF
for that term = natural log(TDOC/TFREQ)+1, which in this case is 4.6. By
default, any term with an IDF weight of less than 2.5 is ignored. A term which
had a frequency of 100 in 1000 has an IDF value of 2.3, and would be ignored.
The lower the IDF value, the more terms are let in. It is simpler to visualize
this as a percentage, i.e., ir_term_drop_off value. A minimum IDF of 2.5 is
equal to an ir_term_drop_off value of 22 (22%). 50% of the documents would
equal an IDF value of 1.7.
ir_max_shared:nnnnn
syntax: ir_max_shared:nnnnn
This parameters sets the maximum amount of shared memory that you want
IR Expert to use. Versions 2 and later do not need to be completely in shared
memory, but rather uses the shared memory as a CACHE of the most
frequently referenced data. Increased shared memory improves performance.
If this parameter is not specified, then 30% of shared memory is used by IR
Expert.
The system determines if version 1.4 or a later version should be used by
checking the type field in the datadict entry for the file. If the type field
contains IR 2.0 then the IR Expert version 2 version is used. (The syntax must
be IR in caps, a space and then 2.0.) If this data is not contained in the type
field, the IR Expert 1.4 is used.

 

[robbe2]

William,

Thanks very much, that was what I was searching for !!!

Regards,

Robbe