Knowledge Editing for Multi-Hop Question Answering Using Semantic Analysis

Authors: Dominic Simon, Rickard Ewetz

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the effectiveness of CHECK against five state-of-the-art frameworks on four datasets and achieve an average 22.8% improved MQA accuracy.
Researcher Affiliation Academia Dominic Simon , Rickard Ewetz University of Florida EMAIL
Pseudocode No The paper describes the methodology of the CHECK framework in Section 4, detailing steps like type extraction, question decomposition, and subquestion resolution, and includes flow diagrams (Figure 1, 2, 3). However, it does not present any explicit pseudocode blocks or algorithm listings.
Open Source Code Yes The code for CHECK is available at https://github.com/dominic-simon/CHECK.
Open Datasets Yes We use the MQu AKE [Zhong et al., 2023] dataset to evaluate the editors.
Dataset Splits No The paper describes the composition of the MQuAKE dataset and its subsets (e.g., "The counterfactual subset contains 3000 edit cases... The temporal subset is composed of 1868 edit cases...") but does not explicitly provide information about how these datasets were split into training, validation, or test sets for the experiments conducted in this paper.
Hardware Specification Yes All experiments were conducted on 1 NVIDIA A100 GPU and 8 CPU cores.
Software Dependencies No The paper mentions several models and frameworks (e.g., ReFinED entity linking model, Contriever dense retrieval model, GPT-J, Vicuna-7B, Falcon-7B) but does not provide specific version numbers for any software libraries, programming languages, or environments used to implement the methodology.
Experiment Setup Yes CHECK used a cosine similarity threshold of 0.8 and was limited to a maximum of 50 new tokens per model call. We use a temperature scale of 0.0 to 1.0 on increments of 0.1.