DIY 챗봇 - LangCon

Size: px

Start display at page:

Download "DIY 챗봇 - LangCon"

송희 저
5 years ago
Views:

1 without Chatbot Builder & Deep Learning

3 Chatbot Builder (=Dialogue Manager),. We need different chatbot builders for various chatbot services. Chatbot builders can t call some external APIs. Deep Learning (=DL) DL can t perfectly control NLU outputs(predictions). I don t know DL. DL is hard to learn.

4 Messaging Platform Slack, Facebook, Dialogue Manager (DM) Rule based (Scenario based) Natural Language Understanding (NLU) Intent Classification (IC) Example based (Search-engine based) Named Entity Recognition (NER) Dictionary based (Regular-expression based) Natural Language Generation (NLG) Template based

6 Messaging Platform / DM ( ) NLU NLG External API

7 Messaging Platform DM NLU userid: bage NLG External API

8 input: T. Messaging Platform DM NLU entity: intent: T check.remain userid: bage NLG External API

9 input: T. Messaging Platform DM NLU entity: intent: T check.remain userid: bage state: T NLG External API

10 input: T. Messaging Platform DM NLU entity: intent: T check.remain userid: bage state: T NLG template: [name] [point]. External API

11 input: T. Messaging Platform DM NLU entity: intent: T check.remain userid: bage state: T NLG template: [name] [point]. External API name: point: 1000

12 input: T. Messaging Platform DM NLU entity: intent: T check.remain userid: bage output: state: T NLG template: [name] [point]. External API name: point: 1000

13 input: T. Messaging Platform DM NLU entity: intent: T check.remain userid: bage output: state: T NLG template: [name] [point]. External API name: point: 1000

14 Dialogue Manager

15 Scenario Messaging Platform Constants userid: bage state: T External API T. Request Actions Execute point: 1000 Conditions point > 0 True False state: T state: T

16 Data Structure Scenario Linked List Constants Constants Global input State Actions External API value Actions Conditions for variables Variables from User, API, NLU, NLG Conditions equal, greater, little, exists, State State.

17 Request from a user Inform to a user Understand NLU Generate NLG Execute external API Request Inform Messaging Platform DM Understand NLU Generate NLG Execute External API

18 Natural Language Understanding

19 Train label Train Data Test label Test Data preprocessing preprocessing classifier train/test classifier model classifier predictions

20 Train label Train Data Test label Test Data preprocessing preprocessing generator train/test generator classifier regular expressions tokens pattern file search engine NER IC Entity Intent predictions

21 Train /t [ ] T- Test /t [ ] T ~~ t- t string to pattern tokenization t\s* \s* t,-,,, tokenization t,,,,, NER IC /t [ ] predictions

22 Intent Classification

23 T ~~ T t t

24 t word spacing word spacing / / _t,t,,,,,,,,, _ n-gram tokens, t,,,,, word tokens morph analyzing t,,,,,, morph tokens

25 Train Test [ ] T- [ ] T ~~ preprocessing preprocessing t- t tokenization tokenization / / / / _t,t-,-,,,, _,_,, _ t,-,,, t,,, search _t,t,,,,,,,,, _ t,,,,, t,,,,,, search engine IC predictions

26 Train Test label sentence label sentence preprocessing preprocessing sentence sentence tokenization tokenization n-gram tokens word tokens morph tokens search n-gram tokens word tokens morph tokens search engine IC predictions

27 Named Entity Recognition

28 t- boundary spacing t - word spacing t - symbols t [-]? whitespace t \s* [-]? \s* \s* prefix match / \s t \s* [-]? \s* \s* \st\s*[-]?\s* \s*

29 Train Test /t T- /t T ~~ preprocessing preprocessing t- _t _ string to pattern pattern match \st\s*[-]?\s* \s* pattern file NER /t predictions

30 Train Test label words(phrase) label sentence preprocessing preprocessing words sentence string to pattern pattern match regular expressions pattern file NER label predictions

32 t t _t t,t,,,, _ n-gram tokens _t,t,,,,,,,,,,, _ n-gram tokens v0.1 v0.2

33 t user input t character based 67% Similiarity alphabet based 85% _t,t,,,, _ =4/6 _t,t,,,,,,,,,,, _ =11/13 n-gram tokens n-gram tokens v0.1 v0.2

34 t- t- t- \st\s*[-]?\s*\s* 1 \st\s*[-]?\s*.? \s* \st\s*[-]?\s* \s* \st\s*[-]?\s*.? \s* \st\s*[-]?\s*\s*.? v0.1 v0.2

35 t- user input t- failed pattern matching success \st\s*[-]?\s* \s* 1 \st\s*[-]?\s*.? \s* v0.1 v0.2

$t- t- t- t- 1 \st\s*[-]?\s*.? \s* \st\s*[-]?$

36 t- t- t- t- 1 \st\s*[-]?\s*.? \s* \st\s*[-]?\s* [  ]? \s* v0.2 v0.21

Software Requirrment Analysis를 위한 정보 검색 기술의 응용

Software Requirrment Analysis를 위한 정보 검색 기술의 응용 EPG 정보 검색을 위한 예제 기반 자연어 대화 시스템 김석환 * 이청재 정상근 이근배 포항공과대학교 컴퓨터공학과 지능소프트웨어연구실 {megaup, lcj80, hugman, gblee}@postech.ac.kr An Example-Based Natural Language System for EPG Information Access Seokhwan Kim