Implementing Cosine Similarity Algorithm to Increase the Flexibility of Hematology Text Report Generation

Abstract

The previous hematology textual summary representation system, which applies template based method of Natural Language Generation to produce hematology laboratory test results in natural language representation, was at the cutting edge to generate more detailed hematology reports. The produced reports manage to provide texts which break down the critical components and abnormal components of blood found in conventional hematology test results. The produced reports in natural language representation aimed to help patients to easily define, spot and point out which blood components are acting up. Templates provide slots to generate every single sentence to be replaced by the data that we provide. However, the previous system is only able to produce fixed unflexible slots of blood components which are defined by the system, named T-Gen System. It nearly got off the ground as it is very unflexible because the produced templates cannot hold all of both critical and abnormal components found in a produced laboratory examination result. Therefore, this research project implements cosine similarity algorithm to expand template flexibility. Testing and evaluation were carried out manually by examining given components into the system which will be added consecutively. The testing shows that every blood component which was added consecutively succesfully appeared in the produced texts.