A reviewer identification using machine learning methods

This article addresses the task of automatically assigning peer reviewers based on historical data from previously submitted and reviewed manuscripts. In conventional editorial practice, reviewer selection relies heavily on the subjective judgment of editors, which can lead to delays and inconsistencies in the quality of expert evaluation. The purpose of this study is to demonstrate that simple natural language processing (NLP) models can be used to automate this process in an efficient and transparent manner. The dataset used in this research consists of both published and rejected articles submitted to the Almaz-Antey Air and Space Defense Corporation Journal, enriched with information about the reviewers assigned to each manuscript. Methodologically, the approach relies on basic text preprocessing, including lemmatization, removal of stop words and punctuation, followed by vectorization using bag-of-words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF) models. Text similarity is calculated via cosine distance between vectorized representations. The core assumption is that a newly submitted manuscript is most similar to an already reviewed one and, therefore, can be assigned to the same reviewers. The results indicate that simple frequency-based models (BoW, TF-IDF) achieve higher accuracy in reviewer assignment (up to 99%) compared to neural network approaches such as Doc2Vec, especially when enhanced with a reviewer co-review graph. The proposed method remains interpretable, requires minimal computational resources, and is fully compatible with office-level computing environments. The model has been shown to perform reliably under class imbalance and is applicable even to relatively small datasets, starting from around 30 manuscripts. However, its generalization to multi-journal editorial systems would require local adaptation, and the task of predicting publication outcomes calls for significantly larger corpora and the use of deep learning architectures. This approach can be seamlessly integrated into digital editorial platforms, contributing to faster decision-making, increased transparency in peer review, and reduced workload for journal staff.