Portuguese word embeddings for the oil and gas industry: Development and evaluation

Enriching Portuguese Word Embeddings with Visual Information

“…Table 5. 16 shows that while multimodality did not manage to improve the best F-score achieved by the best Text-Only model, Standardized NILCFT300, it did raise another model, Standardized NILCW2V100, to tie with this score. Otherwise, whatever increases in F-score as a result of the multimodal fusion that can be observed are minimal at best for this task.…”

“…BERTimbau [44], a Portuguese language BERT model, was recently developed and added to the Hugging Face 4 library. These models, and others, have been used to advance the state-of-the-art in several Portuguese language NLP tasks [40,26,16].…”

Enriching Portuguese Word Embeddings with Visual Information

“…Two new geosciences domain embeddings were developed during the course of this study as part of a collaboration with experts from Petrobras' CENPES research nucleus through the Geologia Digital project: PetroVec and PetroVec-Hybrid 3 . These models were thoroughly tested using both intrinsic and extrinsic tasks, and the results were compiled into an article published in the Computers in Industry journal 4 [16]. These are the current state-of-the-art models for the Portuguese language in the Geosciences domain.…”

“…The test corpus for the geosciences domain, henceforth called GeoSim, was developed as part of the Geologia Digital project, and was used to test the PetroVec word embeddings [16]. It was developed in collaboration with several industry experts, Geology students and a PhD in Geology.…”

“…Word embeddings can be used to parse the text input into the model, using the feature vectors in its tagging predictions. Two annotated corpora were used to evaluate the multimodal embeddings: HAREM [39] for generic domain embeddings; and GeoCorpus 3.0 [16] for geosciences domain embeddings.…”