08-13, 12:00–12:40 (Asia/Yerevan), 214W PAB
The classic sentiment analysis analyzes texts, images, emojis, etc to know what other people think of a product, service, company, or event. While sentiment analysis can be considered one of the accomplished tasks of Natural Language Processing tasks, more fine-grained types of it like Target Based Sentiment Analysis(TSA) or Aspect-based sentiment analysis(ABSA) are the not quite the same. In TSA we want to see the sentiment of a given text towards a particular entity(in my case person or organization). This task is one of the non-solved ones. With the T5 question answering transformer model it was possible to solve the task with results 20% higher than the current leaderboards.
When it comes to analyzing text and getting valuable insights from it, at one point pretty much all the companies have to deal with the sentiment analysis task. While presenting the task I would love to briefly introduce some linguistic differentiation of different types of sentiment analysis tasks with their current leaderboards, benchmarks, and popular datasets and solving approaches but would mainly talk about the Target-based approach its difficulties, and how current models can solve the task. So I would like to show how some tested models work and how the T5-small question-answering model was the best one. I would also like to talk about the data-driven approach which is becoming the new thing in ML. Well known Google Brain and AI Brain Andrew’s NG, lately has covered the benefits of a bigger investment in data preparation with his team proving that investing in improved existing data quality is effective as collecting the triple amount of the data investing in improved existing data quality is effective as collecting the triple amount of the data.
Previous knowledge expected
Liana Minasyan is an NLP researcher and Machine Learning engineer at Polixis. She is involved in various NLP projects, from classic problems to tackling currently unsolved ones. As part of her job, she is creating tools for transliteration, data mapping, target-based sentiment analysis, text summarization, etc. She was involved in sentiment and emotion recognition research for Krisp in collaboration with TUMO. As her capstone at AUA, with YerevaNN supervision, she has worked on the Western and Eastern Armenian Treebank project and was a contributor to Stanford's Universal Dependencies and their NLP toolkit.