saattrupdan/verdict-classifier-en

English Verdict Classifier

This model is a fine-tuned version of roberta-base on 2,500 deduplicated verdicts from Google Fact Check Tools API, translated into English with the Google Cloud Translation API. It achieves the following results on the evaluation set, being 1,000 such verdicts translated into English, but here including duplicates to represent the true distribution:

• Loss: 0.1290
• F1 Macro: 0.9171
• F1 Misinformation: 0.9896
• F1 Factual: 0.9890
• F1 Other: 0.7727
• Precision Macro: 0.8940
• Precision Misinformation: 0.9954
• Precision Factual: 0.9783
• Precision Other: 0.7083

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2500
• num_epochs: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	F1 Misinformation	F1 Factual	F1 Other	Precision Macro	Precision Misinformation	Precision Factual	Precision Other
1.1493	0.16	50	1.1040	0.0550	0.0	0.1650	0.0	0.0300	0.0	0.0899	0.0
1.0899	0.32	100	1.0765	0.0619	0.0203	0.1654	0.0	0.2301	0.6	0.0903	0.0
1.0136	0.48	150	1.0487	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.9868	0.64	200	1.0221	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.9599	0.8	250	0.9801	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.9554	0.96	300	0.9500	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.935	1.12	350	0.9071	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.948	1.28	400	0.8809	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.9344	1.44	450	0.8258	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.9182	1.6	500	0.7687	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.8942	1.76	550	0.5787	0.3102	0.9306	0.0	0.0	0.2900	0.8701	0.0	0.0
0.8932	1.92	600	0.4506	0.4043	0.9628	0.0	0.25	0.3777	0.9753	0.0	0.1579
0.7448	2.08	650	0.2884	0.5323	0.9650	0.3303	0.3017	0.7075	0.9810	0.9474	0.1942
0.6616	2.24	700	0.2162	0.8161	0.9710	0.9724	0.5051	0.7910	0.9824	0.9670	0.4237
0.575	2.4	750	0.1754	0.8305	0.9714	0.9780	0.5421	0.7961	0.9881	0.9674	0.4328
0.5246	2.56	800	0.1641	0.8102	0.9659	0.9175	0.5472	0.7614	0.9892	0.8558	0.4394
0.481	2.72	850	0.1399	0.8407	0.9756	0.9780	0.5686	0.8082	0.9894	0.9674	0.4677
0.4588	2.88	900	0.1212	0.8501	0.9786	0.9783	0.5934	0.8247	0.9871	0.9574	0.5294
0.4512	3.04	950	0.1388	0.8270	0.9702	0.9836	0.5273	0.7904	0.9893	0.9677	0.4143
0.3894	3.2	1000	0.1270	0.8411	0.9737	0.9836	0.5660	0.8043	0.9905	0.9677	0.4545
0.3772	3.36	1050	0.1267	0.8336	0.9732	0.9890	0.5385	0.8013	0.9882	0.9783	0.4375
0.3528	3.52	1100	0.1073	0.8546	0.9791	0.9890	0.5957	0.8284	0.9883	0.9783	0.5185
0.3694	3.68	1150	0.1120	0.8431	0.9786	0.9890	0.5618	0.8244	0.9849	0.9783	0.5102
0.3146	3.84	1200	0.1189	0.8325	0.9738	0.9836	0.54	0.8016	0.9870	0.9677	0.45
0.3038	4.01	1250	0.1041	0.8648	0.9815	0.9836	0.6292	0.8425	0.9884	0.9677	0.5714
0.2482	4.17	1300	0.1245	0.8588	0.9773	0.9836	0.6154	0.8202	0.9929	0.9677	0.5
0.2388	4.33	1350	0.1167	0.8701	0.9808	0.9836	0.6458	0.8377	0.9918	0.9677	0.5536
0.2593	4.49	1400	0.1215	0.8654	0.9790	0.9836	0.6337	0.8284	0.9929	0.9677	0.5246
0.239	4.65	1450	0.1057	0.8621	0.9803	0.9890	0.6170	0.8349	0.9895	0.9783	0.5370
0.2397	4.81	1500	0.1256	0.8544	0.9761	0.9890	0.5981	0.8162	0.9929	0.9783	0.4776
0.2238	4.97	1550	0.1189	0.8701	0.9802	0.9836	0.6465	0.8343	0.9929	0.9677	0.5424
0.1811	5.13	1600	0.1456	0.8438	0.9737	0.9836	0.5741	0.8051	0.9917	0.9677	0.4559
0.1615	5.29	1650	0.1076	0.8780	0.9838	0.9836	0.6667	0.8581	0.9895	0.9677	0.6170
0.1783	5.45	1700	0.1217	0.8869	0.9831	0.9836	0.6939	0.8497	0.9953	0.9677	0.5862
0.1615	5.61	1750	0.1305	0.8770	0.9808	0.9836	0.6667	0.8371	0.9953	0.9677	0.5484
0.155	5.77	1800	0.1218	0.8668	0.9821	0.9890	0.6292	0.8460	0.9884	0.9783	0.5714
0.167	5.93	1850	0.1091	0.8991	0.9873	0.9890	0.7209	0.8814	0.9919	0.9783	0.6739
0.1455	6.09	1900	0.1338	0.8535	0.9773	0.9890	0.5941	0.8202	0.9906	0.9783	0.4918
0.1301	6.25	1950	0.1321	0.8792	0.9820	0.9890	0.6667	0.8439	0.9941	0.9783	0.5593
0.1049	6.41	2000	0.1181	0.9031	0.9879	0.9834	0.7381	0.8911	0.9908	0.9780	0.7045
0.1403	6.57	2050	0.1432	0.8608	0.9779	0.9890	0.6154	0.8237	0.9929	0.9783	0.5
0.1178	6.73	2100	0.1443	0.8937	0.9844	0.9945	0.7021	0.8644	0.9930	0.9890	0.6111
0.1267	6.89	2150	0.1346	0.8494	0.9786	0.9890	0.5806	0.8249	0.9871	0.9783	0.5094
0.1043	7.05	2200	0.1494	0.8905	0.9832	0.9945	0.6939	0.8564	0.9941	0.9890	0.5862
0.0886	7.21	2250	0.1180	0.8946	0.9873	0.9890	0.7073	0.8861	0.9896	0.9783	0.6905
0.1183	7.37	2300	0.1777	0.8720	0.9790	0.9890	0.6481	0.8298	0.9964	0.9783	0.5147
0.0813	7.53	2350	0.1405	0.8912	0.9856	0.9836	0.7045	0.8685	0.9919	0.9677	0.6458
0.111	7.69	2400	0.1379	0.8874	0.9838	0.9836	0.6947	0.8540	0.9941	0.9677	0.6
0.1199	7.85	2450	0.1301	0.9080	0.9879	0.9890	0.7473	0.8801	0.9953	0.9783	0.6667
0.1054	8.01	2500	0.1478	0.8845	0.9838	0.9890	0.6809	0.8546	0.9930	0.9783	0.5926
0.105	8.17	2550	0.1333	0.9021	0.9879	0.9890	0.7294	0.8863	0.9919	0.9783	0.6889
0.09	8.33	2600	0.1555	0.8926	0.9855	0.9890	0.7033	0.8662	0.9930	0.9783	0.6275
0.0947	8.49	2650	0.1572	0.8831	0.9856	0.9890	0.6747	0.8726	0.9885	0.9783	0.6512
0.0784	8.65	2700	0.1477	0.8969	0.9873	0.9890	0.7143	0.8836	0.9908	0.9783	0.6818
0.0814	8.81	2750	0.1700	0.8932	0.9861	0.9890	0.7045	0.8720	0.9919	0.9783	0.6458
0.0962	8.97	2800	0.1290	0.9171	0.9896	0.9890	0.7727	0.8940	0.9954	0.9783	0.7083
0.0802	9.13	2850	0.1721	0.8796	0.9832	0.9890	0.6667	0.8517	0.9918	0.9783	0.5849
0.0844	9.29	2900	0.1516	0.9023	0.9867	0.9890	0.7312	0.8717	0.9953	0.9783	0.6415
0.0511	9.45	2950	0.1544	0.9062	0.9879	0.9890	0.7416	0.8820	0.9942	0.9783	0.6735
0.0751	9.61	3000	0.1748	0.8884	0.9832	0.9945	0.6875	0.8571	0.9930	0.9890	0.5893
0.0707	9.77	3050	0.1743	0.8721	0.9802	0.9890	0.6471	0.8349	0.9941	0.9783	0.5323
0.0951	9.93	3100	0.1660	0.8899	0.9850	0.9890	0.6957	0.8622	0.9930	0.9783	0.6154
0.0576	10.1	3150	0.2029	0.8613	0.9766	0.9890	0.6182	0.8197	0.9952	0.9783	0.4857
0.0727	10.26	3200	0.1709	0.8920	0.9849	0.9890	0.7021	0.8612	0.9942	0.9783	0.6111
0.0654	10.42	3250	0.1599	0.8999	0.9861	0.9945	0.7191	0.8780	0.9919	0.9890	0.6531
0.0553	10.58	3300	0.2091	0.8920	0.9849	0.9890	0.7021	0.8612	0.9942	0.9783	0.6111

Framework versions

• Transformers 4.11.3
• Pytorch 1.9.0+cu102
• Datasets 1.9.0
• Tokenizers 0.10.2