4_inference

Model inference

This folder contains all scripts to run models and create excel files for manual evaluation.

Question generation

To be able to use question generation, previous text extraction of a web page is required (see folder 0_webcrawler)

Check all possible arguments with:

python question_generator.py --help

The following commands are available. Generate questions from website (answer-agnostic, translation pipeline)

python question_generator.py --qg-strategy answer_agnostic --textblock-strategy all --index_name kmi_crawler_2364dfd1-a060-44d9-85a8-eb7e5ecc3fe1 --translation-pipeline True

Other inference commands

Generate questions from website (answer-agnostic):

python question_generator.py --qg-strategy answer_agnostic --textblock-strategy all --index_name kmi_crawler_2364dfd1-a060-44d9-85a8-eb7e5ecc3fe1

Generate questions from faq website (answer-agnostic):

python question_generator.py --qg-strategy answer_agnostic --textblock-strategy faq --index_name kmi_crawler_2364dfd1-a060-44d9-85a8-eb7e5ecc3fe1

Generate questions from website (answer-aware):

python question_generator.py --qg-strategy answer_aware --textblock-strategy all --index_name kmi_crawler_2364dfd1-a060-44d9-85a8-eb7e5ecc3fe1 --ner-library flair --ner-highlighting-strategy word

Paraphrasing

For paraphrasing there is a file in this directory named training_examples.txt which is used for inference. If you want to paraphrase other sentences create a new text file and provide its location via CLI.

Check all possible arguments with:

python paraphrase_generator.py --help

The following commands are available.

With transformer model:

python paraphrase_generator.py --strategy transformer_model --file training_examples.txt --out paraphrases1.xlsx

With backtranslation:

python paraphrase_generator.py --strategy backtranslation --file training_examples.txt --out paraphrases1.xlsx