Untitled

Abstract

We present FLAIR, an NLP framework designed to facilitate training and distribution of state-of-the-art sequence labeling, text classification and language models. The core idea of the framework is to present a simple, unified interface for conceptually very different types of word and document embeddings. This effectively hides all embedding-specific engineering complexity and allows researchers to "mix and match" various embeddings with little effort. The framework also implements standard model training and hyperparameter selection routines, as well as a data fetching module that can download publicly available NLP datasets and convert them into data structures for quick set up of experiments. Finally, FLAIR also ships with a "model zoo" of pre-trained models to allow researchers to use state-of-the-art NLP models in their applications. This paper gives an overview of the framework and its functionality.

Crosslingual named entity recognition for clinical de-identification applied to a COVID-19 Italian data set

“…Different types of embeddings have been selected and mixed, based on BPEmb subword embeddings [44] and Flair contextual string embeddings [54] , for which a detailed analysis is provided below.

Fig.

…”

“…Contextual language models, such as Embeddings from Language Models (ELMo) [51] and Flair [52] , proved to be superior to static models such as Word2Vec [14] and Global Vectors for Word Representation (GloVe) [53] thanks to the ability to analyze the context, and this further improved performance in cross-lingual scenarios. Flair embeddings [54] , [55] , which constitute a character-based contextual language model on which the Flair NLP framework [52] is based, prompted [56] to test a novel methodology for cross-lingual transfer learning for Japanese NER, based on a Bi-LSTM architecture and embeddings at both word and character level as input.…”

Intent-Driven Similarity in E-Commerce Listings

“…More recent work has focused on generating such embedding directly by an unsupervised manner, like with Skip-Thoughts [19], Quick-Thoughts [20] and sent2vec, [26] or by a supervised manner, like with InferSent [9]. More recent work in that field focuses on generating a contextualized embedding, like with ELMo [28], Flair [2] and, perhaps most notably, BERT [10].…”

“…For modeling we use the contextual string embeddings for sequence labeling approach via the Flair framework [2], which achieved state-of-the-art results in multiple sequence labeling tasks, such as Named Entity Recognition (NER) and part-of-speech (PoS) tagging. This framework allows to stack multiple word embeddings from various pre-trained models, which are then passed into a vanilla bidirectional Long short-term memory (BiLSTM) recurrent neural network and a subsequent conditional random field (CRF) decoding layer [2,12,21].…”

“…To train the L2Q model, we use a combination of the characterbased Flair embedding (forward and backward) stacked with GloVe embedding [27] following BiLSTM-CRF with 256 hidden states as suggested and pre-implemented by the Flair framework [2]. Note that the character-level embedding can be advantageous in our case given the sellers' unique and highly diverse jargon.…”