RP23 - Interactive retrieval methods for contradictory evidence

For novel therapies, there are often contradictory publications about their efficacy. Current retrieval methods offer little or no support for this situation. In this project, based on medical entity extraction methods, suitable literature will be searched for diagnosis-therapy pairs specified by the user and classified with respect to their respective evidence [1] and significance.

The main task is to extract the respective statements on efficacy from the publications found. These are then to be aggregated under consideration of the evidence of the respective publication [2]. For this purpose, novel methods for retrieval of contradictory information [3,4] can be applied.

Suitable classification schemes for describing the efficacy (and side effects) of a therapy should be defined and corresponding automatic classifiers implemented. Further, an appropriate user interface needs to be developed in collaboration with the intended end users, including defining interactive functionality. In particular, it is a question of how the experts can influence the arrangement of the documents found. In addition, the question of how to communicate the inherent uncertainty of system decisions will be explored. Also, the usefulness of summaries generated by large language models will be investigated. User studies will then be conducted to evaluate the new system in comparison with the status quo.

[1] Frihat, S. & Fuhr, N. (2023). Determining the Evidence Level of Studies Described in Medical Publications. (in preparation).

[2] Sabbah, F., & Fuhr, N. (2021). A Transparent Logical Framework for Aspect-Oriented Product Ranking Based on User Reviews. In Advances in Information Retrieval: 43rd European Conference on IR Research, ECIR 2021, Part I (pp. 558-571). Springer International Publishing.

[3] Schäfer, H. & Friedrich, C.M. (2019). UMLS mapping and Word embeddings for ICD code assignment using the MIMIC-III intensive care database. 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 6089-6092. 2019.

[4] Sabbah, F. (2023). Logic Based Models for Information Retrieval in Social Media Contexts. Dissertation, University of Duisburg-Essen.