lebinlebin / ms_pointer_network

Re-implementation of multi-source pointer network.

ms_pointer_network

Re-implementation of multi-source pointer network (from the paper Multi-Source Pointer Network for Product Title Summarization).

Requirements

• python - 3.7
• pytorch - 1.4.0
• allennlp - 0.9.0

How to run

Train model

python3 train.py --config_path ./configs/ms_pointer.json --output_dir ./outputs/ms_pointer

Test model

python3 test.py --output_dir ./outputs/ms_pointer

数据格式

• tsv文件，第一列为source1的tokens，第二列为source2的tokens，第三列为target的tokens
• 每一列的tokens均使用空格分隔
• target中的词必须在两个source中出现至少一次

TODO

• 目前的实现共用一个词表，是否需要扩展到独立的词表，以解决oov的复制问题？