OpenTagging for attribute value extraction from text

A Pytorch implementation of "Scaling Up Open Tagging from Tens to Thousands: Comprehension Empowered Attribute Value Extraction from Product Title" (ACL 2019). [pdf]

Requirements:

• Pytorch>=1.9.0
• Python3.8
• transformers 4.8.1

Train and Evaluate

python src/main.py --do_train --do_eval --num_train_epochs 10

Examples of outputs

{
        "context": "luontnor professional men soccer boots brand long spikes soccer shoes outdoor lawn mens football boots training chuteiras",
        "attribute": "types of hobnail",
        "true value": "long spikes",
        "predict value": "long spikes"
},
{
        "context": "150x hd protable astronomical telescope tripod powerful terrestrial space monocular telescope moon watching binoculars",
        "attribute": "type",
        "true value": "monocular",
        "predict value": "monocular"
},

Model Serving

Model is served via a REST API with Flask,

python src/app.py

cURL request

curl -X POST http://0.0.0.0:8005/predict -H 'Content-Type: application/json' -d '{ "context": "Mens Womens Riding Cycling Socks Bicycle sports socks Breathable Anti-sweat Socks Basketball Football Socks", "attribute": "Hose Height"}'

Output

{
    "result":
            [{"value":"socks",
            "position":[7,7],
            "context":"[CLS] mens womens riding cycling socks bicycle sports socks breathable anti - sweat socks basketball football socks [SEP] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]",
            "tokens":["[CLS]","men","##s","women","##s","riding","cycling","socks","bicycle","sports","socks","breath","##able","anti","-","sweat","socks","basketball","football","socks","[SEP]","[PAD]","[PAD]","[PAD]","[PAD]","[PAD]","[PAD]","[PAD]","[PAD]","[PAD]","[PAD]","[PAD]"]
            }]
}