Dataset for Measuring Idea Similarity in Multimodal Academic Papers

Accordingly, this study constructs a similarity sample dataset based on the aforementioned scenarios. Specifically, under the synonym substitution scenario, the overall layout and structural framework of the original diagram remain unchanged, with only the textual content being modified. Some similar samples are constructed through conceptual substitution while maintaining the general research logic.

Under the structural reorganization scenario, samples are generated primarily by altering the shapes, sequencing, or orientation of the flowchart elements while preserving the textual content, thereby retaining the overarching research approach.Based on the above methodologies for constructing similarity samples, this dataset treats the originally extracted research flowcharts as baseline samples and further incorporates similarity samples generated via operations such as synonym substitution and structural reorganization.