A Bootstrapping-based Method to Automatically Identify Data-usage Statements in Publications

Figures & Tables

Statements	Positive (☑) OR negative (☒)
In our experiments, the experimental subset contains 1,552 images selected from the GT database and the FERET databases.	☑ The name, source, and compositions of data
The large-scale database contains 93,638 images captured from 9,668 palms of 4,834 individuals, in which 4–10 images are collected for each palm.	☑ The source and compositions of data
Consequently, both of the two experimental subsets contain 1,200 samples for training and 1,200 samples for testing.	☑ Data compositions and application
In order to show the robustness over short noisy intervals and satisfy the two defined semantics R1 and R2, we generate two completely separated clusters, C1 and C2, using two disjoint interval sequences, Q1 and Q2, and add the synthetically generated short noisy intervals marked in red. Each group contains 10 subjects.	☒ Algorithm description
	☒ Experiment participants
The average training time of the repeated random sub-sampling validation is 1.83 × 30 = 54.9 s, and that of the CBE cross-validation is 1.84 × 5 = 9.2 s.	☒ Experiment process

Seed-selection strategy	Pattern	Seed number	Pattern number	Statement number
COM-SEED	Predicate + Object	14,000	670	29,722
	Subject + Predicate	5,105	596	11,869
GEN-SEED	Predicate + Object	18,235	404	35,711
	Subject + Predicate	5,530	334	11,247

Pattern	Sentences covered by this pattern and the extracted data_clue words
Consists of # samples	The breast cancer set consists of 569 samples with 357 benign and 212 malignant. Dataset 1 is referred to as Char250, which has 250 samples per category for lower and upper cases, respectively; dataset 2 is referred to as Char1000, which has 1,000 samples per category for lower and upper cases, respectively. (Please note this pattern occurs twice here.)
We perform experiments	To assess the ability of the proposed clustering algorithm for classifying the shape classes, we perform experiments on an increasing number of shapes in the two Aslan and Tari datasets. We perform our experiments on a real-estate system with real-life house dataset used in.