Splunk Search

Help with deduping similar values

tomapatan
Communicator

Hi Everyone,

I have a field called "User" that contains similar values and I was wondering how to remove or merge similar values?

For example: "Tony W" and "Anthony W" (both values of the same field) should be merged together.

I was looking at the fuzzy search and jellyfish apps on SplunkBase, but couldn't find a solution to the problem.

My search query:

(index="abc" Name=*) OR (index="xyz" department=* displayName=*)
| eval User=if(isnull(Name), upper(displayName), upper(Name))
| stats values(department) as department by User

Labels (1)
0 Karma
1 Solution

yuanliu
SplunkTrust
SplunkTrust

Unless you want to delve into the language processing wonderland, your best bet is to create a lookup and do it yourself.  Something like

shortenedspelled
tonyantony
nicknicolas
nicknikola
nicknocole
samsamuel
samsamantha
billwilliam
willwilliam

Let's call this table nicknames.  Then,

 

(index="abc" Name=*) OR (index="xyz" department=* displayName=*)
| eval User=if(isnull(Name), lower(displayName), lower(Name)) ``` lower or upper depends on lookup table design ```
| eval User = split(User, "\s+")
| eval firstName = mvindex(User, 0), lastName = mvindex(User, -1), middleInit = if(mvcount(User) > 2, mvindex(User, 1, -2), null())
| lookup nicknames shortened AS firstName output spelled
| lookup nicknames spelled AS firstName output shortened
| lookup nicknames spelled output shortened AS shortened2 ``` handle bill and will ```
| eval firstName = mvdedup(mvappend(firstName, spelled, shortened, shortened2))
| eval firstName = upper(mvjoin(mvsort(firstName), "/")) ``` upper wouldn't be needed if lookup table is in upper ```
| stats values(department) as department values(User) by firstName middleInit lastName ``` values(User) retains original data ```

 

Note that I chose to list a bunch of nicknames with ambiguity.  The only sensible way to handle them is to admit the uncertainty and retain data.

View solution in original post

Tags (1)
0 Karma

yuanliu
SplunkTrust
SplunkTrust

Unless you want to delve into the language processing wonderland, your best bet is to create a lookup and do it yourself.  Something like

shortenedspelled
tonyantony
nicknicolas
nicknikola
nicknocole
samsamuel
samsamantha
billwilliam
willwilliam

Let's call this table nicknames.  Then,

 

(index="abc" Name=*) OR (index="xyz" department=* displayName=*)
| eval User=if(isnull(Name), lower(displayName), lower(Name)) ``` lower or upper depends on lookup table design ```
| eval User = split(User, "\s+")
| eval firstName = mvindex(User, 0), lastName = mvindex(User, -1), middleInit = if(mvcount(User) > 2, mvindex(User, 1, -2), null())
| lookup nicknames shortened AS firstName output spelled
| lookup nicknames spelled AS firstName output shortened
| lookup nicknames spelled output shortened AS shortened2 ``` handle bill and will ```
| eval firstName = mvdedup(mvappend(firstName, spelled, shortened, shortened2))
| eval firstName = upper(mvjoin(mvsort(firstName), "/")) ``` upper wouldn't be needed if lookup table is in upper ```
| stats values(department) as department values(User) by firstName middleInit lastName ``` values(User) retains original data ```

 

Note that I chose to list a bunch of nicknames with ambiguity.  The only sensible way to handle them is to admit the uncertainty and retain data.

Tags (1)
0 Karma

tomapatan
Communicator

Hi yuanliu,

Appreciate the response, exactly what I needed.

Tags (1)
0 Karma
Get Updates on the Splunk Community!

How to Get Started with Splunk Data Management Pipeline Builders (Edge Processor & ...

If you want to gain full control over your growing data volumes, check out Splunk’s Data Management pipeline ...

Out of the Box to Up And Running - Streamlined Observability for Your Cloud ...

  Tech Talk Streamlined Observability for Your Cloud Environment Register    Out of the Box to Up And Running ...

Splunk Smartness with Brandon Sternfield | Episode 3

Hello and welcome to another episode of "Splunk Smartness," the interview series where we explore the power of ...