Generating natural answers with abstractive QA

Customer support chatbots aim to provide helpful, conversational answers, not just exact text snippets. To achieve this, they use abstractive question answering, which generates concise and fluent responses based on the context. Your task is to apply Hugging Face's "text2text-generation" pipeline with a model trained for abstractive QA to create natural answers from product information.

This exercise is part of the course

Natural Language Processing (NLP) in Python

Exercise instructions

Create a qa_pipeline using the "fangyuan/hotpotqa_abstractive" model with the "text2text-generation" task.
Use the provided context and question to generate an abstractive answer.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

from transformers import pipeline

# Create the abstractive question-answering pipeline
qa_pipeline = pipeline(
    task="____",
    model="____"
)

context = """This smartphone features a 6.5-inch OLED display, 128GB of storage, and a 48MP camera with night mode. It supports 5G connectivity and has a battery life of up to 24 hours."""

question = "What is the size of the smartphone's display?"

# Generate abstractive answer
result = qa_pipeline(f"____: {____} ____: {____}")
print(result)

Edit and Run Code

This exercise is part of the course

Natural Language Processing (NLP) in Python

IntermediateSkill Level

4.8+

Start Course for Free

Learn the essentials of text processing in Natural Language Processing (NLP). Master techniques such as tokenization, stop word and punctuation removal, and text normalization with lowercasing, stemming, and lemmatization to prepare text data for further analysis and insight extraction.

Exercise 1: Introduction to natural language processing Exercise 2: Sentence and word tokenization Exercise 3: NLP workflow Exercise 4: Stop words and punctuation handling Exercise 5: Removing stop words Exercise 6: Removing punctuation Exercise 7: Text normalization techniques Exercise 8: Lowercasing Exercise 9: Stemming Exercise 10: Lemmatization

Transform raw text into powerful numerical features. Create Bag-of-Words and TF-IDF representations to capture word importance across documents, then explore word embeddings like Word2Vec and GloVe to uncover deep semantic patterns. Visualize frequency, relevance, and similarity to bring your text data to life.

Exercise 1: Bag-of-Words representation Exercise 2: Building vocabulary from customer reviews Exercise 3: Transforming text to numbers with BoW Exercise 4: Frequency analysis of product reviews Exercise 5: Visualizing word frequencies Exercise 6: TF-IDF vectorization Exercise 7: TF-IDF representation of product feedback Exercise 8: Comparing BoW and TF-IDF representations Exercise 9: Embeddings Exercise 10: Exploring word relationships with embeddings Exercise 11: Visualizing and comparing word embeddings

Harness the power of pre-trained models to perform advanced text classification tasks. Use Hugging Face pipelines for sentiment analysis, topic classification, and natural language inference. Evaluate semantic similarity and grammatical correctness with state-of-the-art models, all without building anything from scratch.

Exercise 1: Hugging Face pipelines for sentiment analysis Exercise 2: Analyzing the sentiment of a review Exercise 3: Batch classifying multiple reviews Exercise 4: Comparing models on labeled review data Exercise 5: Zero-shot classification and QNLI Exercise 6: Zero-shot classification of support tickets Exercise 7: Does the text answer the question?Exercise 8: Question similarity and grammatical correctness Exercise 9: Detecting duplicate questions Exercise 10: Checking grammatical correctness

Dive into the core of modern NLP applications with token classification and text generation techniques. Learn to extract meaningful entities and grammatical structures using NER and PoS tagging. Master both extractive and abstractive question answering, and explore advanced generation tasks including summarization, translation, and language modeling using Hugging Face pipelines.

Exercise 1: Token classification Exercise 2: Identifying named entities in news headlines Exercise 3: Part of Speech tagging for text analysis Exercise 4: Question answering Exercise 5: Answering questions from product descriptions Exercise 6: Generating natural answers with abstractive QA

Current Exercise

Exercise 7: Sequence generation tasks Exercise 8: Summarizing news articles for quick insights Exercise 9: Translating customer reviews to French Exercise 10: Building a search completion system Exercise 11: Congratulations