vlt5-keywords

Keywords generated with vlT5-base-keywords: encoder-decoder architecture, vlT5, keyword generation, scientific articles corpus

vlT5

The biggest advantage is the transferability of the vlT5 model, as it works well on all domains and types of text. The downside is that the text length and the number of keywords are similar to the training data: the text piece of an abstract length generates approximately 3 to 5 keywords. It works both extractive and abstractively. Longer pieces of text must be split into smaller chunks, and then propagated to the model.

Overview

• Language model: t5-base
• Language: pl, en (but works relatively well with others)
• Training data: POSMAC
• Online Demo: Visit our online demo for better results https://nlp-demo-1.voicelab.ai/
• Paper: Keyword Extraction from Short Texts with a Text-To-Text Transfer Transformer, ACIIDS 2022

Corpus

The model was trained on a POSMAC corpus. Polish Open Science Metadata Corpus (POSMAC) is a collection of 216,214 abstracts of scientific publications compiled in the CURLICAT project.

Tokenizer

As in the original plT5 implementation, the training dataset was tokenized into subwords using a sentencepiece unigram model with vocabulary size of 50k tokens.

Usage

from transformers import T5Tokenizer, T5ForConditionalGeneration

model = T5ForConditionalGeneration.from_pretrained("Voicelab/vlt5-base-keywords")
tokenizer = T5Tokenizer.from_pretrained("Voicelab/vlt5-base-keywords")

task_prefix = "Keywords: "
inputs = [
    "Christina Katrakis, who spoke to the BBC from Vorokhta in western Ukraine, relays the account of one family, who say Russian soldiers shot at their vehicles while they were leaving their village near Chernobyl in northern Ukraine. She says the cars had white flags and signs saying they were carrying children.",
    "Decays the learning rate of each parameter group by gamma every step_size epochs. Notice that such decay can happen simultaneously with other changes to the learning rate from outside this scheduler. When last_epoch=-1, sets initial lr as lr.",
    "Hello, I'd like to order a pizza with salami topping.",
]

for sample in inputs:
    input_sequences = [task_prefix + sample]
    input_ids = tokenizer(
        input_sequences, return_tensors="pt", truncation=True
    ).input_ids
    output = model.generate(input_ids, no_repeat_ngram_size=3, num_beams=4)
    predicted = tokenizer.decode(output[0], skip_special_tokens=True)
    print(sample, "\n --->", predicted)

Inference

Our results showed that the best generation results were achieved with no_repeat_ngram_size=3, num_beams=4

Results

|