Toward Intelligent Music Service

Toward Intelligent Music Service

Johnson Wu
LINE Taiwan Data Team Data Scientist

Be4518b119b8eb017625e0ead20f8fe7?s=128

LINE DevDay 2019

November 20, 2019
Tweet

Transcript

  1. 2019 DevDay Toward Intelligent Music Service > Johnson Wu >

    LINE Taiwan Data Team Data Scientist
  2. Self-Introduction • Johnson Wu • 2019.03 Join as Data Scientist

    @ LINE TW data dev team
 • Providing data-driven solutions for local projects
 
 • LINE Fact-Checker, LINE MUSIC TW, User Tagging, General dictionary services …etc.

  3. Agenda > AI technology on LINE MUSIC TW > Module

    introduction > Future works
  4. AI Technology on LINE MUSIC TW How AI can be

    applied ? A basic start. > Name Entity Recognition > Link prediction > Word Embedding Knowledge Graph With User Query Extraction of Knowledge From Unstructured Data > Link prediction Auto-Complete Recommendation > SmartText AutoComplete > A trie-based structure model for popularity and correction on Chinese words.
  5. AutoComplete Start from a simple service, in a search bar!

    Auto-complete user query with popularity search query - the source to start knowledge extraction From 20M+ user queries, 4M+ music dataset E G L E S R T H A … Eagles ? Earth song ? Ea…
  6. Module Introduction How to do auto-complete with popularity? Trie-based search

    solution trained on 20M+ user query logs. auto-complete ranking #(Eagles) > #(ear) > #(earth)
 E G L E S R T H A … Search Count: 100 Search Count: 50 Search Count: 60 homophones are trained for spell correction as well ෢ӽఴʢwu yue tien)→ޒ݄ఱ AutoComplete on Recent popular queries First word Complete 31.84% acc Saving input % 34% Middle word Complete 89.93% acc
  7. AI Technology on LINE MUSIC TW How AI can be

    applied ? Could we understand user’s query? > Name Entity Recognition > Link prediction > Word Embedding Knowledge Graph With User Query Extraction of Knowledge From Unstructured Data > Link prediction Auto-Complete Recommendation > SmartText AutoComplete > A trie-based structure model for popularity and correction on Chinese words.
  8. Prediction of relationship, entities on the user queries or extracted

    triples. ᤚґྛ Womxnly ᤚґྛ Ugly Beauty ᤚґྛ Lady in Red ᤚґྛ Life Sucks Muse ೔ෆམ Ѩ৴ ۚଟ՜ ం౛Ḛ ໷൒ࡾߋ … Link Prediction (1/2) Auto-suggesting with knowledgeable results sing_song Jolin Tsai sing_song Ugly Beauty Lady in Red Life Sucks Wom xnly lyrics_written Ѩ ৴ eng_name ం ౛Ḛ composed ໷ ൒ࡾߋ compose ۚ ଟ՜ write_lyrics ᙽ ڰੈք ީௗ ਓ ੜ༗ݶ ެ࢘ same_album Mus e ೔ ෆམ eng_name
  9. Prediction of relationship, entities on the user queries or extracted

    triples. Link Prediction (2/2) Auto-suggesting with knowledgeable results Related Singers ? Jolin Tsai sing_song Ugly Beauty զ ኷ዑ∍ ײ ᧷㟬త ଘࡏ Jay Chou eng_name eng_name Knight sprint Pirat e Say love you sing_song: Songs ranked by popularity compose 㘸 ޷ෆᄠ ࠂ നᔅٿ ෆ ֘ Knight Spirit Pirate Say Love You Popcorn’s Flavor This Is Love 㘸޷ෆᄠ (by Jay Chou) ࠂനᔅٿ (by Jay Chouʣ ෆ֘ʢby Jay Chouʣ …
  10. Module Introduction How to do knowledge extraction? GraphSpace: a StarSpace-based

    solution of word embedding model on entity/relation prediction from a query/triple input. Trained on 4M entities and 36M relation triples. Example of word analogy: female_singer chinese_named_song #relation: sing_song
 ᤚґྛ ౗ት #sing_song ߐ↠ Ոޙ #sing_song ҆ࣨಸඒዳ 䆪ᄒ࣌ঘঁԦ #sing_song
  11. AI Technology on LINE MUSIC TW How AI can be

    applied ? > Name Entity Recognition > Link prediction > Word Embedding Knowledge Graph With User Query Extraction of Knowledge From Unstructured Data > Link prediction Auto-Complete Recommendation > SmartText AutoComplete > A trie-based structure model for popularity and correction on Chinese words.
  12. Extract name entities from unstructured articles; compliment related search. Name

    Entity Recognition Explore knowledge from our data! YOU & I Coco Lee Randy Merrill Lady Gaga 㐋 䆾 ෍ ፮֨؅ݭ ᒜᅶ Singer of the Song Detected Entities Related to This Song GJ Ḛ୎Յ Ұ ݸڊ੕త ஀ੜ Album Intro
  13. Module Introduction(3/3) How to do name entity recognition? A fine-tuned

    BERT with bi-lstm based decoder model. Costum tags like “Singer”, “Song ”, “Album”, “Composer”, “Lyric writer”, “Album producer” 
 Model NER F1-score on costum tags Bert+Bilstm 0.63 Bilstm 0.21 Original text NER results 48,190 album articles 54,762 NER valid detections
  14. Summary Takeaway 1. A convenient autocomplete function for search service

    2. Knowledge graph solution for more intelligent auto-suggest 3. A potential way to extract knowledge from unstructured data.
  15. Future Works Making the model better 1. Higher accuracy on

    NER recognition. 2. More contents to recognize from unstructured articles. A. Singer style B. Song music Style 3. Human Evaluation designs. 4. Better efficiency for application service.
  16. Thank You