topic Re: Sorting similar names for display in QlikView

Sorting similar names for display

Anonymous — Fri, 29 Apr 2016 17:01:57 GMT

Hello,

I have instances of IDs and Names where the name values are very similar - in some cases, they match exactly with one or more other instance (it may be 1 other, 2 other, unlimited other instances) - the IDs are unique. A sample of the data is below:

ID	NAME
123456	A- Name
123457	A Name
123458	A+ Name
111410	ABC Co - 2006-11
111411	ABC Co - 2006-12
111418	ABC Co - 2007-11
111420	ABC Co - 2007-11
198220	XYZ - City A
198227	XYZ - City A-
199200	MMM Co.
199210	MMM Co.
199220	MMM Co.

How can I extract these IDs and Names, group the Names together by their other instances and display them in a way that clearly shows the grouped instances together (perhaps in a pivot table)? For example, I would like to display the 3 instances of 'MMM Co.' grouped together so the user knows all 3 instances are together.

Many thanks

Re: Sorting similar names for display

settu_periasamy — Fri, 29 Apr 2016 18:19:04 GMT

May be like this?

T1:

LOAD * INLINE [

ID, NAME

123456, A- Name

123457, A Name

123458, A+ Name

111410, ABC Co - 2006-11

111411, ABC Co - 2006-12

111418, ABC Co - 2007-11

111420, ABC Co - 2007-11

198220, XYZ - City A

198227, XYZ - City A-

199200, MMM Co.

199210, MMM Co.

199220, MMM Co.

];

T2:

Left Join(T1)

LOAD NAME,if(Count(NAME)>1,'Yes','No') as Flag Resident T1 Group by NAME;

Re: Sorting similar names for display

sunny_talwar — Fri, 29 Apr 2016 22:32:06 GMT

MMM Co. was straight forward because it was an exact match? Are you also looking to combine XYZ - City A and XYZ - City A- Since they are almost very close?

Re: Sorting similar names for display

MarcoWedel — Mon, 02 May 2016 20:53:55 GMT

Hi,

one solution could be to calculate a group number field (e.g. using the Levenshtein algorithm as described here: Levenshtein Algorithm) and to use this field to group and/or colour code NAME values using a straight or pivot table:

table1:
LOAD * INLINE [
    ID, NAME
    123456, A- Name
    123457, A Name
    123458, A+ Name
    111410, ABC Co - 2006-11
    111411, ABC Co - 2006-12
    111418, ABC Co - 2007-11
    111420, ABC Co - 2007-11
    198220, XYZ - City A
    198227, XYZ - City A-
    199200, MMM Co.
    199210, MMM Co.
    199220, MMM Co.
];

table2:
LOAD Distinct NAME Resident table1;
Join
LOAD Distinct NAME as NAME2 Resident table1;
Join
LOAD NAME, NAME2, levenshtein(NAME, NAME2) as LevDistNAME Resident table2;
Right Join
LOAD Distinct LevDistNAME Resident table2 Where LevDistNAME<=2;

Join (table1)
LOAD NAME,
     AutoNumber(Concat(NAME2,'/')) as NameGroup
Resident table2
Group By NAME;

DROP Table table2;

hope this helps

regards

Marco

Re: Sorting similar names for display

Anonymous — Tue, 03 May 2016 08:03:49 GMT

Hi Marco,

I am VERY interested in your solution - never heard of the Levenshtein algorithm before - but before I try implementing your suggestion, the Name fields I provided are just examples - in reality, the Names won't be hardcoded in an Inline Load, they'll be retrieved from a QVD, so I'll have no idea of the actual names nor how many there are - will your suggestion still work in this case?

Thanks so much.

Re: Sorting similar names for display

Anonymous — Tue, 03 May 2016 08:07:26 GMT

Yes Sunny, I am looking to combine names that are identical and I am looking to combine names that are similar. All the matching logic has already been done in the underlying SQL code - the SQL resultset is exported to a QVD, and my QVW will retrieve the data from the QVD, and I now need to find a way to group these names together in a graphical way.

Thanks.

Re: Sorting similar names for display

MarcoWedel — Tue, 03 May 2016 22:24:02 GMT

Hi,

I used the inline load only to get some test data. You can replace it with a QVD load in a real application.

An issue might rather be the performance for large tables / many distinct Names.

You could try to replace the cartesian product (Join of tables without common fields) with a loop. The following example reduced rather the memory footprint of the load instead of the runtime though (again loading/generating some sample data / to replace with QVD load):

table1:
LOAD RecNo() as ID,
    KeepChar(Hash128(RecNo()),'ABCDEFGHIJKLMNOPQRSTUVWXYZ') as NAME
AutoGenerate 10000;

tabNames:
LOAD Distinct NAME as NAME2 Resident table1;

FOR vCnt = 1 to FieldValueCount('NAME2')
LET vName = FieldValue('NAME2',vCnt);

table2:
LOAD *
Where LevDistNAME<=2;
LOAD NAME2,
      '$(vName)' as NAME,
      levenshtein('$(vName)', NAME2) as LevDistNAME
Resident tabNames;
NEXT vCnt

DROP Table tabNames;

Join (table1)
LOAD NAME,
    AutoNumber(Concat(NAME2,'/')) as NameGroup
Resident table2
Group By NAME;

DROP Table table2;

hope this helps

regards

Marco