Optimizing Journal Selection in Computer Science: A Hybrid Framework Integrating TF-IDF, GloVe, and Sentence-BERT

In the vast landscape of modern science, researchers in fields like computer science face a persistent and practical dilemma: where should they send their work? The academic world organizes its publications into a hierarchy, ranking journals from the most prestigious to the least, often using a system of four tiers labeled Q1 through Q4. These rankings are not arbitrary; they are calculated based on how often articles are cited and the reputation of the journals that publish them. For a scientist, choosing the right journal is critical. A paper placed in a top-tier venue can launch a career, while the same work in a lower-ranked journal might go unnoticed. However, navigating this system is difficult. The databases containing these rankings are massive, complex, and often require manual searching through thousands of entries to find a match for a specific research topic. The sheer volume of data makes it hard for authors to see which publication venue truly fits their unique contribution.

A researcher from Tay Do University in Vietnam has tackled this challenge by building a smart, automated tool designed to guide scientists to the right journal. Instead of relying on a human to manually sift through lists or using a single, rigid method to match text, they created a hybrid system that combines three different ways of understanding language. The first method, known as TF-IDF, acts like a traditional librarian who counts how often specific words appear in a document to determine its importance. The second method, GloVe, looks at the meaning behind words by analyzing how they are used together in vast amounts of text, understanding that words appearing in similar contexts often share similar meanings. The third and most advanced method, Sentence-BERT, converts entire sentences into spatial vectors to enable contextual similarity comparison, allowing the system to compare the research abstract against journal descriptions rather than just matching individual keywords. By weaving these three approaches together, the researcher built a program that can take a scientist's abstract, analyze its core message, and instantly suggest the most suitable journals from a database of over 4,000 entries.

The researcher tested this system using real data from major academic databases like Scopus and SCImago, focusing specifically on the fields of machine learning and deep learning. They fed the system thousands of journal titles and categories, cleaning the data to ensure accuracy, and then asked the program to match new research abstracts against this library. The results were revealing. When the system analyzed a sample abstract about using deep learning for medical image analysis, the different methods within the tool produced different suggestions. The traditional word-counting method pointed toward an e-learning journal, while the word-association method suggested a journal on safety science. However, the most sophisticated method, which utilized vector-based contextual comparison, identified a specialized workshop on visual computing for biomedicine as a highly relevant match with a specific similarity score. This demonstrated that combining multiple word embedding methods and modern language models can improve journal recommendation efficiency, providing a basis for quantitative comparison between traditional algorithms and advanced deep learning models.

The study also confirmed the distinct roles that different tiers of journals play in the scientific ecosystem. The data showed that research focused on deep learning is heavily concentrated in the top-tier Q1 journals, which are known for their high impact and prestige. These venues are where the most influential theoretical work tends to appear. However, the researcher found that Q2, Q3, and Q4 journals are far from obsolete; they serve as essential pillars for applied research and practical applications. These lower-ranked journals provide a necessary space for scientists to share specific, real-world implementations and help emerging authors establish their presence in the field. The automated system successfully highlighted this balance, offering recommendations that respected both the prestige of a journal and the specific nature of the research being submitted.

By integrating these advanced language models with standard statistical data, the researcher created a tool that does more than just list options; it provides a strategic path for publication. The system takes a researcher's abstract, processes it through its hybrid engine, and returns a ranked list of the most relevant journals, complete with their prestige scores and tier rankings. This approach saves researchers significant time and removes the guesswork from the submission process. The findings suggest that such a framework can help scientists, particularly those in developing research communities, navigate the global academic landscape more effectively. It ensures that high-quality work finds a home in the right venue, whether that is a top-tier journal for groundbreaking theory or a specialized outlet for practical application, ultimately helping to improve the visibility and impact of scientific discoveries.