Araştırma makale önerisi için derin öğrenmenin kullanımı

In the field of scientific writing, it is crucial to be able to identify and choose relevant publications. However, due to the increasing number of papers being published, researchers face difficulties in manually navigating through the vast amount of literature. In response to the challenge of navigating the growing volume of academic literature, automated paper recommendation systems have emerged to assist researchers in finding relevant publications. These systems also address the need to establish connections between research outcomes and academic journal subjects for article submissions. Traditional text representation models have limitations in capturing semantic relationships between words and addressing the diverse range of topics across different journals, which affects classification and recommendation results. In this study, a dataset sourced from Kaggle, comprising titles and abstracts of research articles, has been expanded to a total of 60,286 instances. This expansion includes 40,000 additional instances from the National Library of Medicine alongside the original 20,286 instances. These instances are categorized into four classes: physics, mathematics, statistics, and computer science. The research employs various feature extraction methods, including doc2vec, TF-IDF, and Bag of Words, and evaluates the performance of classical machine learning algorithms and deep learning models on the feature vectors. The resulting model functions as a recommendation system, offering content-based suggestions for authors seeking papers relevant to their research interests.