Leveraging word embeddings and semantic enrichment for automatic clinical evidence grading

Clinical practice guidelines are supported by the best available evidence from biomedical publications to assist clinical decision making. The recent technological advances in natural language processing and text mining have the potential in reducing the labor cost and time consumption of creating and updating the guidelines, and improving the quality of clinical recommendations. In order to identify high-quality biomedical publications automatically, we proposed an approach to classify unstructured biomedical text documents into predefined clinical evidence levels based on the linguistic features and semantic enrichment. We investigated the feasibility of leveraging word embeddings for clinical evidence grading that is formulated as a text classification problem, and proposed some strategies for semantic enrichment by incorporating the domain knowledge extracted from the knowledge bases and semantic networks. Moreover, we evaluated the proposed method by applying it to the clinical guidelines of breast cancer. The preliminary results demonstrated that the proposed method performed better than the widely-used baseline methods, and appropriate semantic enrichment could further improve the performance for this challenging task.