Natural products have long been the most important source of ingredients in the discovery of new drugs. Moreover, since the Nagoya Protocol, finding alternative herbs with similar efficacy has become a very important issue in traditional medicine. Although random selection is a common method of finding ethno-medicinal herbs with similar efficacy, it proved to be less effective; therefore, this study aims to develop comprehensive a similarity based framework for targeted selection of herbs with similar efficacy adpting non-curated and curated modern scientific biomedical knowledge. The main goal of the framework is to rank candidate herbs on the basis of similarities that are calculated against target herbs (i.e., to tell which candidate herb is more similar to the target herb in terms of efficacy). In the preliminary study, we proposed the method adopting similarity based on medical subject headings (MeSH) between articles in MEDLINE which is the largest non-curated biomedical database. In order to evaluate the proposed method, we built up three kinds of validation datasets which contain lists of original herbs and corresponding herbs or plants with similar efficacy. The average area under curve (AUC) of the proposed method was found to be about 200- 2500% larger than the random selection method. It was also found that the AUC of the proposed method either remained the same or increased slightly in all three validation datasets as the search range was increased. However it has some limitations. As a result, to overcome the limitation and improving the performance of preliminary study, we proposed a novel framework for the targeted selection of herbs with similar efficacy by exploiting drug repositioning technique and curated modern scientific biomedical knowledge, with the goal of improving the possibility of inferring the traditional empirical ethno-pharmacological knowledge. In the main study, we proposed and evaluated a framework that is comprised of the following four layers: links, extract, similarity, and model. In the framework, multiple databases are linked to build an herb-compound-protein-disease network which was composed of one tripartite network and two bipartite networks allowing comprehensive and detailed information to be extracted. Further, various similarity scores between herbs are calculated, and then ranking models are trained and tested on the basis of theses similarity features. The proposed framework has been found to be feasible in terms of link loss. The ranking model showed improved performance by about 180?480%. While building the ranking model, we identified the compound information as being the most important knowledge source and structural similarity as the most useful measure. In other words, compound knowledge and structural similarity will do important roles in the following researches of finding herbs with similar efficacy. Finally, this study sheds light on biomedical applications; decrease cost in the discovery of alternative herbs through narrowing down the experiment object and reduce risk in new drug development by effective selection of herbs with similar therapeutic efficacy.