One of the most formidable problems in scientific article recommendation systems is recommending articles to cold-started users who have a small interaction history. Most traditional approaches use content-based filtering or collaborative filtering; the same thing may not really capture user preferences and article relationships. In this paper, we introduce a hybrid model combining community detection with topic analysis to recommend articles for cold-start users. The approach unifies both methods to discover hidden user groups and extract serious topics from articles. Clusters of domain-related articles form the community detection module, while the topic analysis extracts topics from these articles and converts them into dense vectors. To make recommendations, the final hybrid model combines community and topic similarities from individual models. Experiments on a large dataset show that our method is effective and outperforms traditional methods regarding precision, recall, and F1 in terms of dataset scale. Our model performs exceptionally well for new users, resulting in an improvement in recommendation accuracy. The introduced hybrid model may lead to a potential solution for the cold-start problem in scientific article recommendation systems, which aim at improving accuracy and collections of recommendations, especially with the limited interaction history. This abstract provides a quick summary of the paper, including the problem statement, proposed solution, and core research contributions.