Extensive experimental data on Traditional Chinese Medicine are available in literature and databases. However, many studies focus on specific diseases or pathways with small sample sizes. As a result, the fundamental pharmacological basis underlying TCM herb properties remains insufficiently elucidated. Based on the concept of the multi-component, multi-target, multi-pathway network of TCM, a data-driven strategy was developed for the profiling of TCM herb properties through network pharmacology and deep learning, facilitating the exploration of the scientific evidence underlying TCM herb properties. Large-scale ingredient and target data of TCM herbs were curated from the HERB2.0 database. KEGG pathway enrichment was conducted for each herb with relative frequency profiling of distinct property groups. Deep learning models were developed and optimized for classification with visual explanation. As a result, high-relative frequency pathways were highly concentrated in five systems (endocrine, immune, nervous, signal transduction, cell growth and death) of KEGG. Herbs with distinct properties exhibited a V-shaped trend (Hot>Warm>Neutral<Cool<Cold) in terms of the abundance of ingredients, targets and high-frequency pathways. The HeteroGAT model improved classification accuracy and provided visual explanations at the ingredient–target–pathway level. We demonstrated a viable strategy to profile TCM property classification from a holistic perspective on ingredients, targets, and pathways, which could help elucidate the scientific basis of TCM properties. However, further advances in model refinement and data matrices are required to enhance the effectiveness of this strategy.