Traditional Chinese medicine (TCM) is known for its diverse components, multiple therapeutic targets, and intricate mechanisms that offer considerable advantages in disease treatment. Identifying drug targets stands at the core of TCM network pharmacology research. Over the years, various web tools targeting drug discovery with diverse features have been created to aid in target prediction, thereby significantly advancing the field of drug discovery. This study presents the Herb-Symptom-Target Network for predicting herb-target associations via symptom-related links. It aims to create concise feature vectors for herb-protein and herb-symptom interactions in a complex network. Through node resource allocation and gene expression data, protein network weights are assigned to identify protein complexes using core-attachment and second-order neighbors. To understand the molecular mechanisms of herbs based on their clinical effects, machine learning techniques are utilized to predict complex herbal protein interactions. This involves integrating heterogeneous information from herbal medicines, symptoms, corresponding targets, and their relationships. As a result, potential targets of herbs within protein complexes are identified. The prediction of herb targets involves extracting feature vectors for network edges, which are then refined using supervised learning approaches. Experimental results demonstrate the method's effectiveness and validity when compared to state-of-the-art herb-target prediction techniques. Additionally, several predicted herb-target interactions have been manually validated using independent literature, confirming the method's potential to successfully merge diverse information for predicting new herb-target interactions.