Question Answering over Electronic Devices: A New Benchmark Dataset and a Multi-Task Learning based QA Framework

This repo has the code for the paper "Question Answering over Electronic Devices: A New Benchmark Dataset and a Multi-Task Learning based QA Framework" accepted at EMNLP 2021 Findings. The blog on this paper can be found here, the poster here, and a corresponding presentation here.

Required dependencies -

Please run pip install -r requirements.txt (python3 required)

E-Manual pre-training corpus

Go to this link. A RoBERTa BASE Model pre-trained on the corpus can be found here, and a BERT BASE UNCASED Model pre-trained on the same here.

Codes

• Annotated Data and Amazon User Forum Data Samples are present in data (See README)
• Data Analysis is done in data_analysis (See README)
• Corpus extraction code is present in pre_training_corpus_extraction (See README)
• E-Manual Data Extraction code is present in EManual_data_extraction (See README)
• Code on pre-training is given in pre-training (See README)
• Code on unsupervised IR method and fine-tuning variants is given in fine_tuning_variants_scripts (See README)
• Code on multi-task learning is given in MTL_scripts (See README)
• Code on funtions for evaluation of MTL and fine-tuning variants is given in evaluation (See README)
  • For ROUGE-L Precision, Recall and F1-Score: https://pypi.org/project/py-rouge/
  • For S+WMS: https://github.com/eaclark07/sms

Baselines

1. Dense Passage Retrieval(DPR) - Used HuggingFace implementation (https://huggingface.co/transformers/model_doc/dpr.html)
2. Technical Answer Prediction (TAP) - took the help of code in https://github.com/IBM/techqa
3. MultiSpan - took the help of code in https://github.com/eladsegal/tag-based-multi-span-extraction

Citation

Please cite the work if you would like to use it.

@inproceedings{nandy-etal-2021-question-answering,
    title = "Question Answering over Electronic Devices: A New Benchmark Dataset and a Multi-Task Learning based {QA} Framework",
    author = "Nandy, Abhilash  and
      Sharma, Soumya  and
      Maddhashiya, Shubham  and
      Sachdeva, Kapil  and
      Goyal, Pawan  and
      Ganguly, NIloy",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
    month = nov,
    year = "2021",
    address = "Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.findings-emnlp.392",
    doi = "10.18653/v1/2021.findings-emnlp.392",
    pages = "4600--4609",
    abstract = "Answering questions asked from instructional corpora such as E-manuals, recipe books, etc., has been far less studied than open-domain factoid context-based question answering. This can be primarily attributed to the absence of standard benchmark datasets. In this paper, we meticulously create a large amount of data connected with E-manuals and develop a suitable algorithm to exploit it. We collect E-Manual Corpus, a huge corpus of 307,957 E-manuals, and pretrain RoBERTa on this large corpus. We create various benchmark QA datasets which include question answer pairs curated by experts based upon two E-manuals, real user questions from Community Question Answering Forum pertaining to E-manuals etc. We introduce EMQAP (E-Manual Question Answering Pipeline) that answers questions pertaining to electronics devices. Built upon the pretrained RoBERTa, it harbors a supervised multi-task learning framework which efficiently performs the dual tasks of identifying the section in the E-manual where the answer can be found and the exact answer span within that section. For E-Manual annotated question-answer pairs, we show an improvement of about 40{\%} in ROUGE-L F1 scores over most competitive baseline. We perform a detailed ablation study and establish the versatility of EMQAP across different circumstances. The code and datasets are shared at https://github.com/abhi1nandy2/EMNLP-2021-Findings, and the corresponding project website is https://sites.google.com/view/emanualqa/home.",
}