Data Mining - Pattern identification T Sql

Question

Hi All ,
I have a table that stores customer complaints on daily basis .
Example :
The table stores customer Information : Customer Name, Complaint Description , Priority .. i want to group the complaints that share the Nearly the same Complaint Description ..
CustomerName
Complaint Description 
Priority
Aroan
AMC server Down
High
James
MNC server Down
Critical
Ryun
node not responding al1232
low
Fred
node not responding 54313
Medium
Maroon 
Network down type rt:1234
High
Rish
Network Down Type rt :3828
Critical
So here even though Complaint Description for Aroan and James is not exactly same but the issue is same server down .Same for Ryun and Fred . is there a way to find the common patterns and present output somewhat like below :
Count 
Complaint Type 
2
Server Down 
2
Node Not Responding 
2
Network Down 
The Data base is OLTP one but not with frequent loads please help with an example how a query should look like for above requirement .
Thanks
Priya

Answer 1

There are 100 ways to answer this question, ranging from reasonably simple (my answer) to getting extremely sophisticated, but in the case of text matching like this, no matter how sophisticated you get, you'll never get a perfect answer (the closest thing
to perfect would be to force the user or help desk agent to pick from a list of complaints up front, before you get this far). 
If your scenario is as simple and predictable as your posted example, this should work fine.  For this answer, I mocked up the data via a SQL statement, but I would recommend creating a permanent table for the "Similarity List Table", that you can easily
add to (so if a new common complaint like 'I can't find the "any key"' needs to be added, you just add to the table, you don't need to modify the query). Note that this also uses "top 1" plus "order by rank" so that only one, the one that you designate
as the highest ranking will ever match.
With EXAMPLE_DATA (Name, Complaint, Severity) as (
Select 'Aroan', 'AMC server Down', 'High'
UNION ALL Select 'James', 'MNC server Down', 'Critical'
UNION ALL Select 'JRyun ', 'node not responding al1232', 'low'
UNION ALL Select 'JFred', 'node not responding 54313', 'Medium'
UNION ALL Select 'JMaroon', 'Network down type rt:1234', 'High'
UNION ALL Select 'JRish', 'Network Down Type rt :3828', 'Critical'
UNION ALL Select 'John', 'The darn network is down again rt:5646', 'High'
UNION ALL Select 'Tom', 'MNC server down, my node not responding abcd1234, and network down type jm:1234', 'Critical'
, Similarity_List_Table (rank, match_wildcard, Complaint_Category) as
Select 1, '%Server Down%', '0001 - Server Down'
UNION ALL Select 2, '%node not responding%', '0002 - Node not responding'
UNION ALL Select 3, '%network down type%', '0003 - Network down type'
Select *, Coalesce(complaint_category, Complaint) as filtrered_complain
from Example_Data
Outer Apply (Select top (1) complaint_category
from Similarity_list_table
where complaint like match_wildcard order by rank) as X
If your situation is more complex and unpredictable (which I am guessing it is), keep in mind what I said, no matter how complex you make something like this, it will still never be perfect.  Consider forcing the users up front to pick a category, and
even with that, expect an inordinate number of "Other" categories being chosen.  :-)