Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
What understood about that we've used LUIS through the year
Search
NAVITIME JAPAN
PRO
January 16, 2018
Technology
0
14
What understood about that we've used LUIS through the year
What understood that we've used LUIS through the year.
NAVITIME JAPAN
PRO
January 16, 2018
Tweet
Share
More Decks by NAVITIME JAPAN
See All by NAVITIME JAPAN
ユーザーのためなら 『デザイン』 以外にも手を伸ばせる
navitimejapan
PRO
1
630
フツーのIT女子が、 Engineering Managerになるまで
navitimejapan
PRO
3
63
不確実性に打ち勝つOKR戦略/How to manage uncertainty with OKR strategy
navitimejapan
PRO
4
2.8k
アジャイルを小さいままで 組織に広める 二周目 / Agile Transformation in NAVITIME JAPAN iteration 2
navitimejapan
PRO
4
1.1k
変更障害率0%よりも「継続的な学習と実験」を価値とする 〜障害を「起こってはならないもの」としていた組織がDirtの実施に至るまで〜 / DevOps Transformation in NAVITIME JAPAN
navitimejapan
PRO
6
4.5k
こうしてふりかえりは終わってしまった / A Demise of a retrospective
navitimejapan
PRO
42
25k
もーひとつの時間がない症候群 / Yet Another SOT Syndrome
navitimejapan
PRO
1
2k
シーズン2〜スクラムチームのバトンを渡す〜 / Season 2 -pass the button of a scrum team-
navitimejapan
PRO
2
2.7k
チームのパフォーマンスを引き出す、ワクワクするプロダクトゴール、OKR / Waku-waku Product Goal and OKR
navitimejapan
PRO
16
17k
Other Decks in Technology
See All in Technology
ハイパフォーマンスな組織をつくるための開発生産性の考え方 / developer-productivity-high-performer-link-and-motivation
lmi
3
270
中央集権体制からDataOpsへの転換 / centralized-to-dataops-transformation
pei0804
7
1.5k
Elementaryを用いたデータ品質の可視化とデータ基盤の運用改善
10xinc
6
1.6k
戦略的DDDを実践するための跳躍力 / OOC 2024
pictiny
6
4.2k
初心者が行く!サーバレスWebアプリ開発の道
nagaharutogawa
0
450
社内共通ルールを値オブジェクトにして社内ライブラリとして運用してみた話
leveragestech
2
1.2k
関数型DDDの理論と実践:「決定を遅らせる」を先につくり、 ビジネスの機動力と価値をあげる
knih
2
490
中学生でもわかる深層学習
e869120
10
2.4k
検証からプロダクトへ: シームレスなLLM開発の ためのしくみ作り
nunukim
1
210
KTC_DBRE.pdf
_awache
1
290
20240321_生成AI時代のDevOps
kzkmaeda
2
610
Evolutionary Optimization of Model Merging Recipes
fuyu_quant0
3
550
Featured
See All Featured
Ruby is Unlike a Banana
tanoku
95
10k
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
14
1.3k
Docker and Python
trallard
33
2.6k
Become a Pro
speakerdeck
PRO
8
4.4k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
657
120k
Agile that works and the tools we love
rasmusluckow
323
20k
Product Roadmaps are Hard
iamctodd
43
9.6k
KATA
mclloyd
14
11k
Side Projects
sachag
451
41k
Build your cross-platform service in a week with App Engine
jlugia
223
17k
Optimizing for Happiness
mojombo
369
69k
個人開発の失敗を避けるイケてる考え方 / tips for indie hackers
panda_program
58
14k
Transcript
What understood about that we’ve used LUIS through the year
Shinichi Tanabe January 12, 2018 Minami Aoyama Night #5
Speaker Shinichi Tanabe (田邊 晋一/たなべ しんいち) • NAVITIME JAPAN
Co., Ltd. ◦ Joined in 2008 ◦ Cogbot project ◦ Programmer
Products
None
Encounter
September 15, 2016
None
None
First impression
Easy to use, runs fast and smart.
Easy to use
Let’s go to the portal site! https://www.luis.ai
Step1. Create new app
None
Step2. Add intent
None
Step3. Add utterances
None
Step4. Add entities
None
None
Step5. Train
None
Step6. Test
None
None
None
None
None
Step7. Publish
None
None
That’s all!
Furthermore...
You can get a happy bonus.
Versioning
None
None
None
None
Runs fast and smart
Comparison between and LUIS Dialogflow
Test model
Test model Intent Places.FindPlace Utterances おいしいカレーが食べたいな どこか近くでおすすめのレストランを教えて Entities Cuisine カレー
PlaceType レストラン
Training speed
LUIS 2 - 4 sec Dialogflow 4 - 8 sec
The training speed result of test model
Precision and recall
Test utterance LUIS Dialogflow Intent Entity Intent Entity おいしいカレーが食べたいな 〇
〇 〇 〇 どこか近くでおすすめのレストラ ンを教えて 〇 〇 〇 〇 Precision result of test model
Test utterance LUIS Dialogflow Intent Entity Intent Entity おすすめのバーを教えて 〇
〇 × × おすすめのバー教えて 〇 × × × おいしいうどんが食べたい 〇 〇 〇 × おいしいうどん食べたい 〇 × 〇 × Recall result of test model
Yes, he was perfect!
Getting started
But, we had some questions.
Questions 1. How should we defines intents and entities? 2.
How do we know accuracy and precision? 3. When will he go GA(General Availability)?
1. How should we defines intents and entities?
Anti pattern Utterance : Intent ≒ 1 : 1
Use or copy pre-build model positively.
None
None
2. How do we know accuracy and precision?
Comprehensive test on model
Batch testing
Test result details in a visualized view.
Error matrix
True positive True negative Green zone indicates correct prediction
False negative False positive Red zone indicates incorrect prediction
3. When will he go GA?
LUIS is now GA!!
How to get along with LUIS
Points 1. Start small model which has few intents. 2.
Use or copy pre-build model positively. 3. Raise requests before do something about that yourself.
Thank you!