Utilizing SBERT For Finding Similar Questions in Community Question Answering

Ha, Thi-Thanh; Nguyen, Van-Nha; Nguyen, Kiem-Hieu; Nguyen, Kim-Anh; Than, Quang-Khoat

doi:10.1109/kse53942.2021.9648830

“…STs have achieved state-of-the-art performance on downstream tasks such as semantic search and question answering (Santander-Cruz et al, 2022;Ha et al, 2021). Their analysis, however, has received considerably less attention than the analysis of the vanilla BERT model and its variants (Rogers et al, 2020;Conia and Navigli, 2022).…”

Section: Introductionmentioning

confidence: 99%

Representation biases in sentence transformers

Nikolaev,

Padó

2023

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics

3

0

View full text Add to dashboard Cite

Variants of the BERT architecture specialised for producing full-sentence representations often achieve better performance on downstream tasks than sentence embeddings extracted from vanilla BERT. However, there is still little understanding of what properties of inputs determine the properties of such representations. In this study, we construct several sets of sentences with pre-defined lexical and syntactic structures and show that SOTA sentence transformers have a strong nominal-participant-set bias: cosine similarities between pairs of sentences are more strongly determined by the overlap in the set of their noun participants than by having the same predicates, lengthy nominal modifiers, or adjuncts. At the same time, the precise syntactic-thematic functions of the participants are largely irrelevant.

show abstract

“…STs have achieved state-of-the-art performance on downstream tasks such as semantic search and question answering (Santander-Cruz et al, 2022;Ha et al, 2021). Their analysis, however, has received considerably less attention than the analysis of the vanilla BERT model and its variants (Rogers et al, 2020;Conia and Navigli, 2022).…”

Section: Introductionmentioning

confidence: 99%

Representation biases in sentence transformers

Nikolaev¹,

Padó²

2023

Preprint

0

View full text Add to dashboard Cite

Variants of the BERT architecture specialised for producing full-sentence representations often achieve better performance on downstream tasks than sentence embeddings extracted from vanilla BERT. However, there is still little understanding of what properties of inputs determine the properties of such representations. In this study, we construct several sets of sentences with pre-defined lexical and syntactic structures and show that SOTA sentence transformers have a strong nominal-participant-set bias: cosine similarities between pairs of sentences are more strongly determined by the overlap in the set of their noun participants than by having the same predicates, lengthy nominal modifiers, or adjuncts. At the same time, the precise syntactic-thematic functions of the participants are largely irrelevant.

show abstract