An intensive recent research work investigated the combined use of hand-curated knowledge resources and corpus-driven resources to learn effective text representations. The overall learning process could be run by online revising the learning objective or by offline refining an original learned representation. The differentiated impact of each of the learning approaches on the quality of the learned representations has not been studied so far in the literature. This article focuses on the design of comparable offline vs. online knowledge-enhanced document representation learning models and the comparison of their effectiveness using a set of standard IR and NLP downstream tasks. The results of quantitative and qualitative analyses show that (1) offline vs. online learning approaches have dissimilar result trends regarding the task as well as the dataset distribution counts with regard to domain application; (2) while considering external knowledge resources is undoubtedly beneficial, the way used to express relational constraints could affect semantic inference effectiveness. The findings of this work present opportunities for the design of future representation learning models, but also for providing insights about the evaluation of such models.

In order to measure the medical activity, hospitals are required to manually encode information concerning an inpatient episode using International Classification of Disease (ICD-10). This task is time consuming and requires substantial training for the staff. We propose to help by speeding up and facilitating the tedious task of coding patient information, specially while coding some secondary diagnoses that are not well described in the medical resources such as discharge letter and medical records. Our approach leverages data mining techniques in order to explore medical databases of previously encoded secondary diagnoses and use the stored structured information (age, gender, diagnoses count, medical procedures…) to build a decision tree that assigns the proper secondary diagnosis code into the corresponding inpatient episode or indicates the impatient episodes that contains implausible secondary diagnoses. The results suggest that better performance could be achieved by using low level of diagnoses granularity along with adding some filters to balance the repartition of the negative and positive examples in the training set. The obtained results show that there is big variation in the evaluation scores of the studied diagnoses, the highest score is 75% using F1 measurement and the lowest 25% using F1 measurement which indicates further enhancements are needed to achieve better performance regardless of the encoded diagnosis. However, the average accuracy of all the studied secondary diagnoses is around 80% which indicates better negative predictions therefore it could be useful in the prevention or the detection of wrong coding assignments of secondary diagnoses in the inpatient stay.