This is the code for emnlp2022 paper A Generative Model for End-to-End Argument Mining with Reconstructed Positional Encoding and Constrained Pointer Mechanism
Task: argument component identification (ACI) and argumentative relation identification (ARI).
Install the package in the requirements.txt
The evaluation dataset includes Argument Annotated Essays and Consumer Debt Collection Practices (CDCP).
After enter the folder, you can run the code by directly using
python train.py --dataset essayThe performance is reported in average. You can check the performance in logs folder as follow:
...
...
{"metric": {"Seq2SeqSpanMetric_essay_add_none": {"triple_f": 44.56, "triple_rec": 44.42, "triple_pre": 44.7, "am_component_f": 70.82000000000001, "am_component_rec": 71.43, "am_component_pre": 70.22, "em": 0.3333, "invalid": 0.0048, "entity_info": {"<<P>>": {"acc": 74.1784, "recall": 71.9818, "f1": 73.0636}, "<<C>>": {"acc": 62.1053, "recall": 67.0455, "f1": 64.4809}, "<<MC>>": {"acc": 68.75, "recall": 77.6471, "f1": 72.9282}}, "entity_overall": {"acc": 70.2247191011236, "recall": 71.42857142857143, "f1": 70.8215297450425}, "invalid_len": 0.0048, "invalid_order": 0.0, "invalid_cross": 0.0, "invalid_cover": 0.0}, "step": 1748, "epoch": 46}}
{"metric": {"data_test": {"Seq2SeqSpanMetric_essay_add_none": {"triple_f": 50.339999999999996, "triple_rec": 49.44, "triple_pre": 51.29, "am_component_f": 76.09, "am_component_rec": 75.85, "am_component_pre": 76.33, "em": 0.3593, "invalid": 0.0, "entity_info": {"<<P>>": {"acc": 81.1297, "recall": 78.1211, "f1": 79.597}, "<<C>>": {"acc": 65.1235, "recall": 70.0997, "f1": 67.52}, "<<MC>>": {"acc": 75.6579, "recall": 75.1634, "f1": 75.4098}}, "entity_overall": {"acc": 76.33466135458168, "recall": 75.85114806017418, "f1": 76.09213661636221}, "invalid_len": 0.0, "invalid_order": 0.0, "invalid_cross": 0.0, "invalid_cover": 0.0}}}}
We update the inference code to load the saved model, make inference and show the inferece results. You can see more details in "infer.py" . The infer_res.ipynb is provided to load the pkl file and convert the sequence results into readable strings.