电信系统与管理

Applied NLP - Data, Models, Pipelines, Business Decisions: Lessons from the field

Abstract

Łukasz Augustyniak

Natural Language Processing (NLP) projects often fail in their conception (e.g., lack of data, wrong data, wrong annotations), in their delivery, or their business usefulness. It is a good idea to identify what is the proper research problem, what a client is looking for, what are our limitations, and finally, how you would solve the problem. The applied NLP is still rapidly evolving and changing area - I want to present the experience from a several NLP project from social media analysis, news data extraction, call center analytics, legal documents annotation and many more. What are the best practices for data acquisition, creating the first models, packing all into pipelines, and how we can serve these models? In addition, I show how we may communicate our work beyond the client deliverables to guide other applied NLP practitioners, guiding them towards success and pointing them away from many NLP pitfalls.