Rank-Level Fusion of Random Indexing, Word Embedding, and TF-IDF-Based Rankings for Clinical Document Retrieval

The amount of clinical data present with medical professionals is growing at a high rate. The data may be collected from clinical records, Internet, social media, and medical books to mention a few. This vast amount of voluminous structured and unstructured clinical data is often tedious to search and analyze. Clinical document retrieval is intended for quick access to the required clinical documents by a staff, patient, doctor, nurse, or any other person in authority. Hence, it becomes paramount to develop a system to search data/document in the medical repositories for efficient and quick analysis of a patient case that may require instant attention. The users of such a system may pose several queries and expect retrieval of relevant clinical documents. The research work proposed in this paper is based on fusion of document retrieval results applied on several novel techniques. Novel technique of rank-level fusion for retrieval is applied on top of other retrieval techniques to obtain a decision of the final rank. The contribution and novelty of the present work are two-fold: firstly, we propose to use two new techniques for clinical document retrieval viz. Random indexing-based retrieval and GLOVE representation-based retrieval. Secondly, we perform the proposed technique of rank-level fusion over the results obtained through various ranking techniques. The fused rank helps in decision making in the scenario that different ranks are produced by different algorithms used for retrieval. The results obtained using this novel approach show improvement over existing techniques.