Siamese Long Short-Term Memory for Detecting Conflict of Interest on Scientific Papers

Abstract

Scientific articles cited by other researchers have an impact on increasing author credibility. However, the citation process may be misused to unnaturally raise a bibliometric indicator value such as researcher’s h-index. Researchers may overly cites their own works, referred as self-citation, even though the topic of the references are not related to the current article. Further misconduct is excessive citations on the works of peoples related to the researcher which can be coercive or not, referred as conflict of interest (CoI). The proposed method uses a deep learning approach, Siamese Long ShortTerm Memory (LSTM), to recognize subject similarities between a scientific article and its references. Standard text similarity fails to do so because contextual relatedness of sentences in the articles need some learning process. Siamese-LSTM learns contextual relatedness of sentences in the article using two identical LSTM. Steps of the proposed method are (i) word-embedding to get weight values of terms but still considers their semantic relations, (ii) k-means clustering to generate training data for reducing time complexity in Siamese-LSTM learning of scientific articles, (iii) learns Siamese-LSTM weight from training data to identify contextual relatedness of sentences, (iv) calculate similarity of a scientific article with its references based on Siamese-LSTM. The empirical experiments are used to analyze similarity values and the possibility for conflict of interest in an article.