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
打造新電商搜尋體驗- 搜尋意圖辨識
Search
LINE Developers Taiwan
PRO
August 28, 2025
0
21
打造新電商搜尋體驗- 搜尋意圖辨識
LINE Developers Taiwan
PRO
August 28, 2025
Tweet
Share
More Decks by LINE Developers Taiwan
See All by LINE Developers Taiwan
清大企業參訪- Ben
line_developers_tw
PRO
0
920
LLM 商品規格萃取大冒險- Vila
line_developers_tw
PRO
0
1k
Playwright/MCP/AI -Winter
line_developers_tw
PRO
0
1k
LINE EC Product Catalog Development- Rei
line_developers_tw
PRO
0
1k
LINE 與 AI 機器人技術應用現況
line_developers_tw
PRO
0
10
QA Testing
line_developers_tw
PRO
0
260
jcconf_datadev_prod
line_developers_tw
PRO
0
12
jcconf_SPM_prod
line_developers_tw
PRO
0
9
jcconf_LINEPay_prod
line_developers_tw
PRO
0
8
Featured
See All Featured
The Cult of Friendly URLs
andyhume
79
6.7k
Building a Scalable Design System with Sketch
lauravandoore
463
34k
Raft: Consensus for Rubyists
vanstee
140
7.2k
JavaScript: Past, Present, and Future - NDC Porto 2020
reverentgeek
52
5.7k
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
55
3.1k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
35
3.2k
Rebuilding a faster, lazier Slack
samanthasiow
84
9.3k
Writing Fast Ruby
sferik
630
62k
Rails Girls Zürich Keynote
gr2m
95
14k
Intergalactic Javascript Robots from Outer Space
tanoku
273
27k
Build The Right Thing And Hit Your Dates
maggiecrowley
38
2.9k
RailsConf 2023
tenderlove
30
1.3k
Transcript
None
打造新電商搜尋體驗 搜尋意圖辨識 Speaker: Dan Chen
01 02 03 04 如何做搜尋意圖 Prompt vs Finetune LLM 驗證
什麼是搜尋意圖 05 總結 CONTENT
Dan Chen LINE Taiwan EC Dev - Data Scientis All
about Recommendation ….
Dan Chen LINE Taiwan EC Dev - Data Scientis Work
Experience Side Project Intern Work - Consultant (Startup / traditional industries) - DS4G (domestic violence / social housing / elderly issues) - TWiDS / WiDS - Publish TF2.0 books
真實的需求與目的 01 什麼是搜尋意圖?
搜尋意圖是… 特點項目文字 特點項目 特點項目文字 特點項目 關鍵字語句背後 用戶的真實需求與目的 提升用戶體驗 提升平均訂單價值 縮短購買決策週期
特點項目文字 特點項目 顧客 油性肌膚 No Results 生日快樂 星座 *https://www.flaticon.com/
過去機器學習到LLM世代 02 如何做搜尋意圖
Unsupervised Supervised 雖然預測速度快, 但不夠精準 需要大量工人智慧
- Requirement 1. Low latency 2. High precision Dictionary-based
該選擇什麼策略 03 Prompt vs Finetune
Prompt Fine-tune - What is ? - Ref: AI Engineering
Published by O'Reilly Media, Inc. *https://www.flaticon.com/
Prompt Fine-tune - When use? - Ref: AI Engineering Published
by O'Reilly Media, Inc. *https://www.flaticon.com/
如何驗證LLM 04 LLM驗證
*https://www.flaticon.com/ 驗證方法 1. Weak Supervise 2. Cross validation with different
model 3. Self-consistency/Ensemble 4. Synthetic Data
*https://www.flaticon.com/ 開源Performance比較 Breeze Taiwan llama gpt-OSS
*https://www.flaticon.com/ 建立 Reliable Sample
05 Conclusion
None
None