A tool for aligning manuscripts with suitable journals

Chin, Hong An (2026) A tool for aligning manuscripts with suitable journals. Final Year Project, UTAR.

PDF Download (2348Kb)

Abstract

This project is a research-based project focused on machine learning and deep learning for the evaluation of machine learning algorithms used in aligning manuscripts with suitable journals. Traditional “Journal Finder” tools are limited to single publishers’ portfolios and rely on keyword-only matching (TF–IDF, BM25), which overlooks context and synonyms, resulting in poor recommendations and delayed publication. In this project, journal scope metadata from the Directory of Open Access Journals (DOAJ) has been assembled and normalised, covering 21,782 journals across 327 categories. Three retrieval models were implemented and compared—TF-IDF, Okapi BM25, and Sentence-BERT—with hyperparameters tuned using Bayesian optimisation (Optuna). Three distinct input strategies were tested—title only, title + scope, and a hybrid approach using title + scope + category—to determine the optimal method for content representation. For each pipeline, manuscripts and journal scopes were encoded into vector representations to retrieve the top candidate journals. Performance was rigorously evaluated on a held-out test set using HitRate@K, MAP@K, and NDCG@K, under both random and equal distribution setups. The results demonstrate that TF-IDF with the hybrid input strategy achieves a HitRate@1 of 88.84%, outperforming BM25 (83.47%) and Sentence-BERT (68.53%). This finding contradicted the initial hypothesis that transformer-based models would yield superior performance and was attributed to the standardised vocabulary and short document lengths in the journal metadata. The best-performing model was deployed in the FlareSearch web application as a practical journal recommendation tool.

Actions (login required)