Saturday 11:45–12:30 in Tower Suite 1

Beyond word2vec: GloVe, fastText, StarSpace.

Konstantinos Perifanos

Audience level:
Experienced

Description

Word embeddings is a very convenient and efficient way to extract semantic information from large collections of textual or textual-like data. We will be presenting an exploration and comparison of the performance of "traditional" embeddings techniques like word2vec and GloVe as well as fastText and StarSpace in NLP related problems such as metaphor and sarcasm detection

Abstract

Word embeddings, mappings from words to d-dimensional vectors, is a very convenient and efficient way to extract semantic information from large collections of textual or textual-like data.

In this talk we will have a closer look on the maths and the performance of different word embedding techniques, where the generated vector representations are used as features for NLP tasks such as metaphor and sarcasm detection as well as applications beyond natural language, eg product similarity from clickstream data, source code e.t.c

Subscribe to Receive PyData Updates

Subscribe