This webinar talks about converting unstructured e-commerce text into structured format by leveraging multi-task deep learning.
The existing problem: e-sellers upload text information in unstructured or semi-structured format. E-commerce engines use naive text search techniques as a result of which search engine performance suffers because of less number irrelevant results. Can we convert unstructured text into a structured (key:value) format with high precision and high recall?
Challenges: Presence of multiple products in text description (e.g., blue top goes well with black jeans, which is the primary product?), semantic dis-ambiguity (e.g., 'Blue' as brand vs color, top as filler vs fashion category), context spread across long sentences (e.g., this is stunning red color top, ....., this sleeveless piece is a gem, ....) scale: more than 4 m e-fashion listings crawled, parsed and converted into structured format.
Let us dive into the solutions with this webinar.
- Motivate the problem by examples
- Key challenges and requirements to solve the problem
- Bidirectional LSTM based deep network
- Case Study and deployment challenges
obtained his PhD in Computer Science and Engineering from IIT Bombay (2012), and subsequently worked with IBM Research Labs as Research Scientist for 2.5 years. He co-founded, Huew, a content-driven shopping destination that is organizing rich multi-media content on the web, and connecting it to e-commerce websites. Vijay has over 10 years of experience in working with cutting-edge machine learning, deep learning and networking systems.