Chih-Ming Chen
|
A Ph.D. researcher with a deep passion for data analysis — research that isn't only theoretical but applicable to real-world problems. Builds recommendation products, competes in predictive-modeling challenges, and leads open-source machine-learning projects.
CHARACTER SHEET
Stats · Skills · Training Arc
SKILL STATS
PLAYER INFO
TRAINING ARC
QUEST LOG
Experience — 10+ years of completed missions
Miso Technologies
Algorithm Developer
- Maintain and improve the Personalization-as-a-Service system.
KKBox / KKLab / KKStream
Algorithm Developer
- Built a decade-spanning arsenal of recommendation algorithms for music & video streaming.
Reveal the full loot table
- Cross-domain recommendations
- Text-based item-to-item recommendations
- Streaming item recommendations
- Ticket recommendations
- Personalized song recommendations
- Song-to-song recommendations
- Image-to-song recommendation
- Heterogeneous query-based recommendations
- Approximate nearest-neighbor search
- Personalized daily music discovery
Living Analytics Research Centre (LARC)
Research Assistant
- Developed a text-aware recommendation model (published at CIKM'20).
CHANGING.AI
Algorithm Developer Intern
- Maintained and improved the Recommendation-as-a-Service system.
Tmall, Alibaba Group
Algorithm Developer Intern
- Investigated online-to-offline recommendation approaches.
- Implemented an O2O graph-embedding solution and a list-wise learning-to-rank embedding framework.
CITI, Academia Sinica
Research Assistant
- Investigated recommendation-related research.
- Investigated network-embedding modeling techniques.
CS Department, National Chengchi University
Teaching Assistant
- Python Programming (graduate level), 2017–2018.
- Web Search and Mining (graduate level), 2013–2019.
- Computer Programming course / lab class, 2011–2012.
ACHIEVEMENTS
Boss fights — data-science competitions & honors
PIXNET Visitor Log Data Challenge
1st place / 40+ teams
Moving-average features + factorization machine.
KDD Cup 2014 — DonorsChoose.org
2nd on private leaderboard
Pruned feature interactions in FMs + EDA-driven prediction smoothing.
KAMERA Data Challenge
2nd place / 100+ teams
Outlier smoothing + linear regression for medical-emergency data.
ACM RecSys Challenge 2020
4th place worldwide
Tweet-engagement prediction at Twitter scale.
WSDM Cup 2018 — KKBox Music Rec
7th / 1000+ teams
Single GBM with similarity features from his own proNet toolkit.
KDD Cup 2015 — MOOC Dropouts
9th prize / 800+ teams
Linear ensemble with novel backward-cumulative features.
ACM RecSys 2019 — Tutorial Talk
90-minute tutorial: SMORe
Modularize graph embedding for recommendation.
Tradeshift Text Classification (Kaggle)
20th / 300+ teams
Two-layer boosting prediction pipeline.
TIGP Travel Grant
Publication support
Taiwan International Graduate Program.
Departmental Scholarship, NCCU CS
Top-ranking students
Awarded for academic excellence.
INVENTORY
Open-source artifacts — forged & battle-tested
SMORe
Modularized recommendation from a graph perspective. Simulates many network-embedding models with high training speed, low memory, and accurate performance.
Inspect item →ICE — Item Concept Embedding
Embeds item concepts so similarity works across item types — a song can retrieve similar songs or similar concepts. Published at SIGIR 2017.
Inspect item →awesome-network-embedding
A curated paper list for network-embedding research — a community-loved map of the representation-learning realm.
Inspect item →competitive-recsys
A curated collection of recommender-system research — the field guide for anyone entering the RecSys arena.
Inspect item →CODEX
Publications — 14 scrolls inscribed in the archives
JOIN THE PARTY
Recruiting for an ML quest? Looking for a teammate who embeds all the world? Send a raven — or just an email.