NCCS Nonprofit Program Classification

1
NCCS Nonprofit Program Classification

• Jan. 2004 Update

2
The Task

• More than 637,000 returns representing more than
  1 million program descriptions.
• Nearly 619,000 unique programs. (Same
  description appears year after year code only
  once)

3
The Challenge (1)

• Oh, look at the beautiful forest. I bet we can
  find a trail through to the other side. And
  they ventured forth into the forest. The
  underbrush grew dense and they were soon totally
  lost ... They could no longer see the forest from
  the trees.

4
The NPC System

• Parallels the NTEE with 26 major groups
• 13 common/general codes
• 53 beneficiary/population codes
• 671 program codes

5
Classification Strategies

• Word Weighting
• NCCS program (Foxpro)
• Existing text classification/artificial
  intelligence programs
• Challenges Lots of noise too many codes
• Manual Verification
• Start with auto-generated codes from word
  weighting
• Challenges Staff and outsourcer burnout and the
  limits of human cognition ? mixed quality

6
Classification Strategy 3 Rule-Building

• Started with words and concepts spreadsheet
• Now web-based interface, similar in function to
  commercial packages
• 847 rules for 658 NPC codes
• Plus international/ethnic group codes
• Plus higher-order rules for prioritizing codes,
  fine-tuning integrating beneficiary, general,
  and international codes

7
Example 1

• Adoption (3) 2 concepts (adopt stopwords)
• ADOPT adopt
• STOPWORDS vet//, pet//, pets//, cat//, ....
• Translation Adopt is in either the org. name
  or program description AND NONE of the STOPWORDS
  are in the org. name or prog. desc.
• Note Wildcard at end of word assumed unless
  //. Double-slash means that the word must
  match EXACTLY (vet but not veteran).

8
Example 2

• P04.02 Family Preservation (22) 3 concepts
• PRESERVATION interven, maintain, support
• FAMILY famil
• CHILD child, youth, adolescen, foster
• Translation At least one word from each of the 3
  concepts must be in the org. name or prog. desc.

9
The Challenge (2)

• NASA project leader Rob Manning noted "It
  looks pretty easy, but I want to remind you that
  we required eight thrusters to turn the vehicle,
  we had two cooling pumps that had to work, we had
  37 pyrotechnic devices that included . . . two
  thermal batteries, eight cable cutters, three gas
  generators, one mortar cannon and actually in
  this case, five or so solid rocket motors," he
  said. "We had four sensors, a star scanner, a sun
  sensor, a radar altimeter, two inertial
  measurement units . . . a descent camera, two
  radios, one computer and a lot of software and
  airbags." And, he added in amazement, "they
  worked! - W. Post 1/5/2003

10
Next Steps

• Almost all of the major systems are completed.
• Subsector Teams
• Classification and subsector expertise
• Refine subsectors, one by one
• Use NPC codes to assign and/or verify NTEEs
• Assign single primary NPC code whenever possible
• Use same process for directly assigning or
  verifying NTEEs