Why data-driven methods will shape the future (...)
summary
Argues that replacing brittle synonyms, taxonomies and rules with trained models scales better and faster: implements language modeling for autocompletion (RNN/LSTM sequence models), neural NER (Flair-style embeddings), graph and network embeddings for query-product relations, and query-product embeddings fed into a learning-to-rank layer. Reports end-to-end deep models can improve relevance but suffer bias, convergence and cold‑product problems in practice; recommends hybrid pipelines, metriced by nDCG/A-B tests, noting the GAN portrait sold for >$400,000 as a data‑art example.
AI-written summary · about
chapters (20)
- 0:00Intro
- 2:45Search quality
- 4:10Relevant search
- 5:09Relevant search tweaking
- 8:13Complex tests
- 8:44Looking for opening markets
- 10:36What kind of data
- 11:40What is machine learning
- 13:01How can I use machine learning
- 14:32Queries products
- 15:56Traditional architectures
- 17:56Complete suggestions
- 22:14Vector representation
- 25:52Image Recognition
- 28:45Click Logs
- 30:32Paper
- 34:05Ranking
- 35:22Functions
- 36:30Conditional functions
- 37:35Conclusion
related talks
41mModels Schm-odels: Why you Should Care about Data-Centric AI
53mGenerative AI: Overview and use cases
53mML BuzzWords demystified
48mThe Missing Step in Moving from Data to Models
44mContent Classification: NLP vs. AI in the Age of Data Vectorization
30mSemantic Search and Neural Information Retrieval
01h 16mGeneralization in natural language processing
34mGenerative AI Landscape Building a Modern Data Strategy