์ด ๋‹จ๊ณ„์—์„œ๋Š” ์–ธ์–ด ๊ทœ์น™(Linguistic Rule)์„ ์‚ฌ์šฉํ•˜์—ฌ ์‚ฌ์ „ ์• ๋…ธํ…Œ์ด์…˜์„ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ์‚ฌ์ „ ์• ๋…ธํ…Œ์ด์…˜์„ ์ค‘๋ณตํ•˜์—ฌ ์ˆ˜ํ–‰ํ•˜๋Š” ๊ฒฝ์šฐ ์ด์ „ ๋‹จ๊ณ„์—์„œ ์ˆ˜ํ–‰ํ•œ ์‚ฌ์ „ ์• ๋…ธํ…Œ์ด์…˜์€ ์‚ฌ๋ผ์ง‘๋‹ˆ๋‹ค. ์ด๋ฅผ ์›ํ•˜์ง€ ์•Š๋Š” ๊ฒฝ์šฐ ์ด๋ฒˆ ๋‹จ๊ณ„๋ฅผ ์Šคํ‚ตํ•˜์‹ญ์‹œ์˜ค.

์ „์ฒด ์‹ค์Šต์€ ๋‹ค์Œ์˜ ์ˆœ์„œ๋กœ ์ง„ํ–‰๋ฉ๋‹ˆ๋‹ค.
โ€ข Exercise 01: ํ”„๋กœ์ ํŠธ ์ƒ์„ฑ, ํƒ€์ž… ์‹œ์Šคํ…œ ์ƒ์„ฑ ๋ฐ ์ˆ˜์ •
โ€ข Exercise 02: ๊ธฐ๊ณ„ ํ•™์Šต์šฉ ์ž๋ฃŒ(Training Corpus) ์—…๋กœ๋“œ
โ€ข Exercise 03: ๋”•์…”๋„ˆ๋ฆฌ ์• ๋…ธํ…Œํ‹ฐํ„ฐ(Dictionary Annotator) ์ƒ์„ฑ ๋ฐ ์‚ฌ์ „ ์• ๋…ธํ…Œ์ด์…˜(Pre-annotation)
โ€ข Exercise 04: ์–ธ์–ด์  ๊ทœ์น™(Linguistic Rule)์„ ์‚ฌ์šฉํ•œ ์‚ฌ์ „ ์• ๋…ธํ…Œ์ด์…˜(Pre-annotation)
โ€ข Exercise 05: ํœด๋จผ ์• ๋…ธํ…Œ์ด์…˜(Human annotation) ํƒœ์Šคํฌ ์ƒ์„ฑ
โ€ข Exercise 06: ํœด๋จผ ์• ๋…ธํ…Œ์ด์…˜(Human annotation)
โ€ข Exercise 07: ํŒ๊ฒฐ(Adjudication)

์–ธ์–ด ๊ทœ์น™(Linguistic Rule)์„ ์‚ฌ์šฉํ•˜๋ฉด ์‰ฝ๊ณ  ๋น ๋ฅด๊ฒŒ ์—”ํ‹ฐํ‹ฐ ์• ๋…ธํ…Œ์ด์…˜์„ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. Watson Knowledge Studio์—์„œ Rule ๊ธฐ๋ฐ˜์˜ ์• ๋…ธํ…Œ์ด์…˜์€ ๋…๋ฆฝ์ ์œผ๋กœ ๊ตฌ์„ฑ ๋ฐ ๋ฐฐํฌ๋  ์ˆ˜๋„ ์žˆ๊ณ  ๋จธ์‹ ๋Ÿฌ๋‹ ์• ๋…ธํ…Œ์ดํ„ฐ ๋นŒ๋“œ๋ฅผ ์œ„ํ•œ Pre-Annotator๋กœ์จ ์‚ฌ์šฉ๋  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค.

์–ธ์–ด ๊ทœ์น™ ๊ธฐ๋ฐ˜์˜ ์• ๋…ธํ…Œ์ด์…˜์„ ์‚ฌ์šฉํ•˜๋ฉด ์ข‹์€ ์ƒํ™ฉ์€ ์˜ˆ๋ฅผ๋“ค์–ด ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค.:
โ€ข ๊ด€๊ณ„ ๋ณด๋‹ค๋Š” ์—”ํ‹ฐํ‹ฐ ์ถ”์ถœ์— ๊ด€์‹ฌ์ด ์žˆ๋Š” ๊ฒฝ์šฐ
โ€ข ์—”ํ‹ฐํ‹ฐ ๋ฐ ๊ด€๊ณ„๊ฐ€ ๋น„๊ต์  ์‰ฝ๊ฒŒ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์–ด ์‰ฝ๊ฒŒ ์• ๋…ธํ…Œ์ด์…˜ ํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒฝ์šฐ
โ€ข ๋ฌธ์„œ์—์„œ ์—”ํ‹ฐํ‹ฐ๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ•์— ์žˆ์–ด ๋ณ€ํ™” ๋ฐ ๋‹ค์–‘์„ฑ์ด ์ ์€ ๊ฒฝ์šฐ
โ€ข ๋ฌธ์„œ์— ๊ณต๋™ ์ง€์‹œ์–ด(co-references)๊ฐ€ ์ƒ๋Œ€์ ์œผ๋กœ ์ ์€ ๊ฒฝ์šฐ

์œ„์™€ ๊ฐ™์€ ์ƒํ™ฉ์ด๋ผ ํ• ์ง€๋ผ๋„ ๋จธ์‹  ๋Ÿฌ๋‹ ์• ๋…ธํ…Œ์ดํ„ฐ๊ฐ€ ๋” ํšจ๊ณผ์ ์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฃฐ ๊ธฐ๋ฐ˜์˜ ์• ๋…ธํ…Œ์ดํ„ฐ๋Š” ์‚ฌ์ „-์• ๋…ธํ…Œ์ดํ„ฐ๋กœ์จ ๋” ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค. ํœด๋จผ-์• ๋…ธํ…Œ์ดํ„ฐ๊ฐ€ Ground-Truth๋ฅผ ๋งŒ๋“œ๋Š” ์—…๋ฌด๋ฅผ ํ›จ์”ฌ ์‰ฝ๊ฒŒ ๋งŒ๋“ค์–ด ์ค๋‹ˆ๋‹ค.

โ€ข Rules: These are linguistic rules that are built on individual words (which are called tokens) as well as dictionaries, regex rules, and other linguistic rules. This provides the ability to build rules on top of previously built rules (and in the future, machine-learning annotations).
โ€ข Dictionaries: These are the dictionaries that you created earlier, exposed here for use in the linguistic rules annotator.
โ€ข Regex: These are rules built using the Regular Expression language. Regex is a pattern matching expression language. It is especially useful for modeling specific token patterns such as credit card numbers, email address, postal codes, and license plates. You do not use Regex rules in this exercise, but they are easy to create and deploy. You can find more information on regex at this link: https://en.wikipedia.org/wiki/Regular_expression.

1) Document Annotation > Rules๋กœ ์ด๋™ํ•ฉ๋‹ˆ๋‹ค.

์–ธ์–ด ๊ทœ์น™ ๊ธฐ๋ฐ˜์œผ๋กœ ์‚ฌ์ „-์• ๋…ธํ…Œ์ด์…˜์„ ์ˆ˜ํ–‰ํ•˜๋ ค๋ฉด ๋จผ์ € ๋ฌธ์„œ๋ฅผ ์ƒ์„ฑํ•˜๊ฑฐ๋‚˜ ๊ฐ€์ ธ์™€์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด ์‹ค์Šต์—์„œ๋Š” ์ด ๊ทœ์น™์„ ๋จธ์‹ ๋Ÿฌ๋‹ ์• ๋…ธํ…Œ์ด์…˜์„ ๋นŒ๋“œํ•˜๊ธฐ ์œ„ํ•œ ์‚ฌ์ „-์• ๋…ธํ…Œ์ด์…˜ ์šฉ๋„๋กœ ์‚ฌ์šฉํ•˜๋ฏ€๋กœ ์ด ์ „ ๋‹จ๊ณ„์—์„œ ์—…๋กœ๋“œํ•œ ๋ฌธ์„œ๋ฅผ ๊ทธ๋Œ€๋กœ ์‚ฌ์šฉํ•˜๋ฉด ๋ฉ๋‹ˆ๋‹ค.

2) Documents ํƒญ ์šฐ์ธก์˜ + ์•„์ด์ฝ˜์„ ๋ˆŒ๋Ÿฌ ๋ฌธ์„œ๋ฅผ ๊ฐ€์ ธ์˜ต๋‹ˆ๋‹ค. Copy From All Set ํƒญ์—์„œ ์—…๋กœ๋“œํ•œ ๋ฌธ์„œ ์ค‘ ํ•˜๋‚˜๋ฅผ ์„ ํƒํ•˜๊ณ  ํ•˜๋‹จ์˜ Copy๋ฒ„ํŠผ์„ ๋ˆŒ๋Ÿฌ ๋ณต์‚ฌํ•ฉ๋‹ˆ๋‹ค.

๋‹ค์Œ๊ณผ ๊ฐ™์ด ๋ฌธ์„œ๊ฐ€ ๋ณต์‚ฌ๋ฉ๋‹ˆ๋‹ค.

3) ๋‹ค์Œ ๋‹จ๊ณ„๋กœ Class๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. Class๋Š” Entity Type๊ณผ ๋™๋“ฑํ•œ ์šฉ์–ด๋กœ Rule์—์„œ๋งŒ ์“ฐ์ด๋Š” ์šฉ์–ด์ž…๋‹ˆ๋‹ค. ๋‚˜์ค‘์— Class๋ฅผ ์ƒ์‘ํ•˜๋Š” Entity Type์— ๋งคํ•‘ํ•˜์—ฌ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์Œ์˜ ํด๋ž˜์Šค๋ฅผ ์ƒ์„ฑํ•˜์‹ญ์‹œ์˜ค.

– Model_Year_Class
– Manufacturer_Class
– Model_Class

Class ํƒญ์—์„œ + ๋ฒ„ํŠผ์„ ๋ˆ„๋ฆ…๋‹ˆ๋‹ค.

Class ์ด๋ฆ„์„ ์ž…๋ ฅํ•˜๊ณ  Add ๋ฒ„ํŠผ์„ ๋ˆ„๋ฆ…๋‹ˆ๋‹ค.

์œ„ ๋‹จ๊ณ„๋ฅผ ๋ฐ˜๋ณตํ•˜์—ฌ 3๊ฐœ์˜ ํด๋ž˜์Šค๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.

4) Dictionary์— Class๋ฅผ ๋งตํ•‘ํ•ฉ๋‹ˆ๋‹ค. Document Annotation > Dictionaries๋กœ ์ด๋™ํ•ฉ๋‹ˆ๋‹ค.

manufacturer_dict ๋ฅผ ํด๋ฆญํ•˜๋ฉด ๋‹ค์Œ์ฒ˜๋Ÿผ ํด๋ž˜์Šค๋ฅผ ๋งตํ•‘ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์…€๋ ‰ํŠธ ๋ฐ•์Šค์—์„œ Manufacturer_Class๋ฅผ ์„ ํƒํ•˜๊ณ  Save ๋ฒ„ํŠผ์„ ๋ˆ„๋ฆ…๋‹ˆ๋‹ค.

๋งˆ์ฐฎ๊ฐ€์ง€๋กœ ๋‹ค์Œ๊ณผ ๊ฐ™์ด model_dict ๋”•์…”๋„ˆ๋ฆฌ์— Model_Class๋ฅผ ๋งตํ•‘ํ•ฉ๋‹ˆ๋‹ค.

5) ๊ฐ€์šด๋ฐ Documents ํƒญ์— ์žˆ๋Š” ๋ฌธ์„œ๋ฅผ ํด๋ฆญํ•˜์‹ญ์‹œ์˜ค. Manufacturer ๋ฐ Model์ด ์• ๋…ธํ…Œ์ด์…˜ ๋˜์–ด ์žˆ๋Š” ๊ฒƒ์„ ํ™•์ธํ•˜์‹ญ์‹œ์˜ค.

6) ์ด์ œ Model_Year ๋ฅผ ์• ๋…ธํ…Œ์ด์…˜ ํ•˜๊ธฐ ์œ„ํ•œ ์–ธ์–ด ๊ทœ์น™์„ ์ƒ์„ฑํ•ด ๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค. ํ•œ ๋ฌธ์žฅ์—์„œ ๋…„๋„ ๋ฐ”๋กœ ๋’ค์— ์ œ์กฐ์‚ฌ ์ด๋ฆ„๊ณผ ๋ชจ๋ธ ์ด๋ฆ„์ด ์˜ค๋ฉด Model_Year๋ผ๊ณ  ์ •์˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์‹œ Document Annotation > Rules๋กœ ์ด๋™ํ•˜์‹ญ์‹œ์˜ค.

7) Document ํƒญ์—์„œ “2007”, “ํ˜„๋Œ€”, “์†Œ๋‚˜ํƒ€”๋ฅผ ํด๋ฆญํ•˜๋ฉด Rules ํƒญ์— ์žˆ๋Š” + ๋ฒ„ํŠผ์ด ํ™œ์„ฑํ™”๋ฉ๋‹ˆ๋‹ค.

+ ๋ฒ„ํŠผ์„ ํด๋ฆญํ•ฉ๋‹ˆ๋‹ค.

8) ๋‹ค์Œ๊ณผ ๊ฐ™์ด ๋ฃฐ ํŽธ์ง‘ ํ™”๋ฉด์ด ๋‚˜์˜ต๋‹ˆ๋‹ค. “2007” ์œ„์— ์žˆ๋Š” ๋ฐ•์Šค๋ฅผ ํด๋ฆญํ•˜์—ฌ Model_Year_Class๋ฅผ ๋งตํ•‘ํ•ฉ๋‹ˆ๋‹ค.

9) Rules ํƒญ์—์„œ Rule์˜ ์ด๋ฆ„์„ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ง€์ •ํ•ด์ค๋‹ˆ๋‹ค.

10) “2007”์„ ์„ ํƒํ•˜๊ณ  Open Properties ๋ฒ„ํŠผ์„ ์„ ํƒํ•˜๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์†์„ฑ ๋ณ€๊ฒฝ ํ™”๋ฉด์ด ๋‚˜ํƒ€๋‚ฉ๋‹ˆ๋‹ค.

11) ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์†์„ฑ์„ ๋ณ€๊ฒฝํ•ฉ๋‹ˆ๋‹ค.

12) Save ๋ฒ„ํŠผ์„ ๋ˆŒ๋Ÿฌ ์ €์žฅํ•ฉ๋‹ˆ๋‹ค. ์ƒ˜ํ”Œ ๋ฌธ์„œ์—์„œ Model_Year๊ฐ€ ์ œ๋Œ€๋กœ ์• ๋…ธํ…Œ์ด์…˜ ๋˜์—ˆ๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.

13) ์ด์ œ ์ƒ์„ฑํ•œ ๋ฃฐ๋กœ ๋ฌธ์„œ์— ์‚ฌ์ „-์• ๋…ธํ…Œ์ด์…˜ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•  ์ฐจ๋ก€์ž…๋‹ˆ๋‹ค. Model Management > Versions๋กœ ์ด๋™ํ•ฉ๋‹ˆ๋‹ค.

14) Rule-based model type mapping ํƒญ์œผ๋กœ ์ด๋™ํ•ฉ๋‹ˆ๋‹ค.

15) Model, Manufacturer, Model_Year ํƒ€์ž…์— ํด๋ž˜์Šค๋ฅผ ๋งตํ•‘ํ•ฉ๋‹ˆ๋‹ค. Actions ์ปฌ๋Ÿผ์— ์žˆ๋Š” Edit ๋ฒ„ํŠผ์„ ํด๋ฆญํ•˜๊ณ  ์›ํ•˜๋Š” ํด๋ž˜์Šค๋ฅผ ์„ ํƒํ•œ ํ›„ ์ €์žฅํ•˜๋ฉด ๋ฉ๋‹ˆ๋‹ค.

16) Rule-based ํƒญ์œผ๋กœ ์ด๋™ํ•ฉ๋‹ˆ๋‹ค. Run this model ๋ฒ„ํŠผ์„ ํด๋ฆญํ•ฉ๋‹ˆ๋‹ค.

17) ํŒ์—…์—์„œ ์‚ฌ์ „ ์• ๋…ธํ…Œ์ด์…˜์„ ์ˆ˜ํ–‰ํ•  ๋ฌธ์„œ ์…‹์„ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

18) Run ๋ฒ„ํŠผ์„ ํด๋ฆญํ•˜๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ˆ˜ํ–‰๋ฉ๋‹ˆ๋‹ค.

ํ† ๋ก  ์ฐธ๊ฐ€

์ด๋ฉ”์ผ์€ ๊ณต๊ฐœ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ํ•„์ˆ˜ ์ž…๋ ฅ์ฐฝ์€ * ๋กœ ํ‘œ์‹œ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.