Table 2
Comparison of named entity detection across different models.
| Named entity | DeepSeek-V3 | ERNIE-4.5 | ERNIE-4 | GPT-4o | Qwen2.5-32B | Qwen3-32B | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Acc. | Prec. | Cov. | Acc. | Prec. | Cov. | Acc. | Prec. | Cov. | Acc. | Prec. | Cov. | Acc. | Prec. | Cov. | Acc. | Prec. | Cov. | |
| ATel alert fields | ||||||||||||||||||
| Transient Name | 0.759 | 0.753 | 1.000 | 0.759 | 0.763 | 0.986 | 0.759 | 0.760 | 0.986 | 0.785 | 0.812 | 0.944 | 0.646 | 0.750 | 0.833 | 0.810 | 0.819 | 0.972 |
| Transient RA | 0.949 | 0.880 | 0.960 | 0.937 | 0.955 | 0.840 | 0.759 | 0.538 | 0.960 | 0.962 | 1.000 | 0.880 | 0.924 | 0.952 | 0.800 | 0.949 | 0.955 | 0.880 |
| Transient Dec | 0.886 | 0.895 | 0.720 | 0.873 | 0.842 | 0.720 | 0.696 | 0.444 | 0.840 | 0.899 | 0.895 | 0.760 | 0.873 | 0.941 | 0.640 | 0.886 | 1.000 | 0.640 |
| Transient Uncertainty | 0.924 | 0.143 | 1.000 | 0.937 | 0.167 | 1.000 | 0.924 | 0.143 | 1.000 | 0.924 | 0.250 | 1.000 | 0.911 | 0.125 | 1.000 | 0.924 | 0.143 | 1.000 |
| Telescope | 0.671 | 0.667 | 0.987 | 0.709 | 0.707 | 0.987 | 0.633 | 0.623 | 1.000 | 0.696 | 0.726 | 0.947 | 0.544 | 0.635 | 0.827 | 0.734 | 0.730 | 0.987 |
| Transient Type | 0.671 | 0.658 | 1.000 | 0.608 | 0.629 | 0.929 | 0.595 | 0.581 | 0.986 | 0.722 | 0.721 | 0.957 | 0.595 | 0.694 | 0.814 | 0.671 | 0.667 | 0.957 |
| Uncertainty Confidence | 0.987 | 0.750 | 1.000 | 0.962 | 0.500 | 0.333 | 0.987 | 1.000 | 0.667 | 0.975 | 0.667 | 0.667 | 0.975 | 0.667 | 0.667 | 0.975 | 0.667 | 0.667 |
| Average (ATel) | 0.835 | 0.678 | 0.952 | 0.826 | 0.652 | 0.828 | 0.765 | 0.584 | 0.920 | 0.852 | 0.724 | 0.879 | 0.781 | 0.681 | 0.797 | 0.850 | 0.712 | 0.872 |
| GCN circular fields | ||||||||||||||||||
| Source Type | 0.767 | 0.775 | 0.989 | 0.689 | 0.816 | 0.844 | 0.667 | 0.682 | 0.978 | 0.711 | 0.744 | 0.956 | 0.767 | 0.873 | 0.878 | 0.733 | 0.835 | 0.878 |
| Source Name | 0.878 | 0.940 | 0.933 | 0.856 | 0.939 | 0.911 | 0.889 | 0.930 | 0.956 | 0.878 | 0.888 | 0.989 | 0.900 | 0.976 | 0.922 | 0.889 | 0.952 | 0.933 |
| RA | 0.967 | 0.897 | 1.000 | 0.922 | 1.000 | 0.750 | 0.933 | 0.806 | 1.000 | 0.989 | 0.964 | 1.000 | 0.967 | 0.900 | 1.000 | 0.933 | 0.957 | 0.821 |
| Dec | 0.956 | 0.862 | 1.000 | 0.911 | 0.952 | 0.750 | 0.922 | 0.800 | 0.964 | 0.978 | 0.929 | 1.000 | 0.944 | 0.833 | 1.000 | 0.922 | 0.913 | 0.821 |
| Location Error | 0.800 | 0.553 | 0.964 | 0.800 | 0.568 | 0.929 | 0.767 | 0.500 | 1.000 | 0.833 | 0.605 | 1.000 | 0.767 | 0.524 | 0.964 | 0.767 | 0.529 | 0.821 |
| Telescope | 0.644 | 0.644 | 1.000 | 0.700 | 0.724 | 0.967 | 0.689 | 0.697 | 0.989 | 0.700 | 0.708 | 0.989 | 0.589 | 0.602 | 0.978 | 0.611 | 0.625 | 0.978 |
| Average (GCN-Circular) | 0.835 | 0.779 | 0.981 | 0.813 | 0.833 | 0.858 | 0.811 | 0.736 | 0.981 | 0.848 | 0.806 | 0.989 | 0.822 | 0.785 | 0.957 | 0.809 | 0.802 | 0.875 |
Notes. Acc. = Accuracy; Prec. = Precision; Cov. = Coverage. Accuracy refers to the fraction of exactly matched fields over all annotated ones. Precision represents the proportion of correct predictions among all predictions made by the model. Coverage indicates the fraction of required fields that the model successfully extracted.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.