Work

Miscitations in scientific papers: dataset and detection

Year: 2024
Type: conference-paper
Abstract: Citations play an important role in scientific research.However, numerous inaccurate citations are found within scientific papers.These erroneous citations, also called miscitations, may result in the misinterpretation of research, distortion of the original author's intended meaning, and potentially even more serious consequences.The objective of our study is to detect automatically the erroneous citations, by assessing whether the citation contexts accurately reflects the content of the cited papers.We do not assess the correctness or formatting of the references within the reference list.We built a balanced "proof of concept" dataset, including both reliable and erroneous citations sourced from published open-access scientific papers, with the aim of investigating the feasibility of automatically detecting erroneous citations.In addition to the dataset, our study also introduces two methods for evaluating the reliability of citations using Natural Language Processing (NLP) techniques.The first method involves using cosine similarity metric.The second method implements a paraphrase classifier to distinguish between reliable and erroneous citations.Both methods take as input BERT embeddings and output the reliability of citations : reliable or erroneous.Based on our experimental results, the best performance is obtained with cosine similarity.
Source: HAL
Authors: Liu, Qinyue, Barhoumi, Amira, Labbé, Cyril
Language: English
Open Access status: green
OpenAlex ID: W4396667367
FWCI: 3.568
Cites: 25
Cited by: 3
Related to: 10