Скачать или смотреть Charles-Albert Lehalle: "An Attempt to Understand Natural Language Processing"

Full Title: "An Attempt to Understand Natural Language Processing And Illustration On A Financial Dataset"

Speaker: Charles-Albert Lehalle (Capital Fund Management) - Joint work with Mengda Li
Currently Head of Data Analytics at Capital Fund Management (CFM, Paris) and visiting researcher at Imperial College (London), Charles-Albert Lehalle studied machine learning for stochastic control during his PhD 20 years ago.

Abstract:
I will present a theoretical analysis of word2vec language models and explain how the resulting understanding can generalize to more nonlinear ones (like BERT).

This analysis relies on trying to exhibit a generative model allowing to explain the asymptotic meaning of the loss functions used by these kinds of models. In particular, it allows to produce synthetic languages having a controlled number of "synonyms" and try to learn them with standard algorithms. I will then show how learning the language of financial news reflects (or not) the celebrated Loughran-McDonald sentiment lexicon. This on-going work is conducted with Mengda Li (ENS Paris Saclay).

Lightning Talk: "Can We Predict Stock Price Movements Using NLP Techniques?"
48:44    • Charles-Albert Lehalle: "An Attempt to Und...  

CFEM alumna Silvia Ruiz will discuss her team capstone project, which was titled, “How to Predict Stock Movements Using NLP Techniques.” By utilizing NLP techniques, the Cornell CFEM team, sponsored by Rebellion Research, explored whether investing signals can be extracted financial data. The team analyzed 10K and 10Q reports from S&P 500 companies using techniques such as FinBERT and word2vec.

Silvia Ruiz (MFE Cornell ’20, BS Mathematics Universidad Del Valle ’17) has experience working as a Data Scientist for Corporación Multi Inversiones and as a Risk Analytics Analyst for Morgan Stanley.

Github Repository: https://github.com/shadyfish03/Rebell...