docs/content/guides/text-summarization.md

363 lines
14 KiB
Markdown

---
title: Build a text summarization app
linkTitle: Text summarization
keywords: nlp, natural language processing, text summarization, python, bert extractive summarizer
description: Learn how to build and run a text summarization application using Python, Bert Extractive Summarizer, and Docker.
summary: |
This guide shows how to containerize text summarization models with Docker.
tags: [ai]
languages: [python]
aliases:
- /guides/use-case/nlp/text-summarization/
params:
time: 20 minutes
---
## Overview
In this guide, you'll learn how to build and run a text summarization
application. You'll build the application using Python with the Bert Extractive
Summarizer, and then set up the environment and run the application using
Docker.
The sample text summarization application uses the Bert Extractive Summarizer.
This tool utilizes the HuggingFace Pytorch transformers library to run
extractive summarizations. This works by first embedding the sentences, then
running a clustering algorithm, finding the sentences that are closest to the
cluster's centroids.
## Prerequisites
- You have installed the latest version of [Docker Desktop](/get-started/get-docker.md). Docker adds new features regularly and some parts of this guide may work only with the latest version of Docker Desktop.
- You have a [Git client](https://git-scm.com/downloads). The examples in this section use a command-line based Git client, but you can use any client.
## Get the sample application
1. Open a terminal, and clone the sample application's repository using the
following command.
```console
$ git clone https://github.com/harsh4870/Docker-NLP.git
```
2. Verify that you cloned the repository.
You should see the following files in your `Docker-NLP` directory.
```text
01_sentiment_analysis.py
02_name_entity_recognition.py
03_text_classification.py
04_text_summarization.py
05_language_translation.py
entrypoint.sh
requirements.txt
Dockerfile
README.md
```
## Explore the application code
The source code for the text summarization application is in the `Docker-NLP/04_text_summarization.py` file. Open `04_text_summarization.py` in a text or code editor to explore its contents in the following steps.
1. Import the required libraries.
```python
from summarizer import Summarizer
```
This line of code imports the `Summarizer` class from the `summarizer`
package, essential for your text summarization application. The summarizer
module implements the Bert Extractive Summarizer, leveraging the HuggingFace
Pytorch transformers library, renowned in the NLP (Natural Language
Processing) domain. This library offers access to pre-trained models like
BERT, which revolutionized language understanding tasks, including text
summarization.
The BERT model, or Bidirectional Encoder Representations from Transformers,
excels in understanding context in language, using a mechanism known as
"attention" to determine the significance of words in a sentence. For
summarization, the model embeds sentences and then uses a clustering
algorithm to identify key sentences, those closest to the centroids of these
clusters, effectively capturing the main ideas of the text.
2. Specify the main execution block.
```python
if __name__ == "__main__":
```
This Python idiom ensures that the following code block runs only if this
script is the main program. It provides flexibility, allowing the script to
function both as a standalone program and as an imported module.
3. Create an infinite loop for continuous input.
```python
while True:
input_text = input("Enter the text for summarization (type 'exit' to end): ")
if input_text.lower() == 'exit':
print("Exiting...")
break
```
An infinite loop continuously prompts you for text
input, ensuring interactivity. The loop breaks when you type `exit`, allowing
you to control the application flow effectively.
4. Create an instance of Summarizer.
```python
bert_model = Summarizer()
```
Here, you create an instance of the Summarizer class named `bert_model`. This
instance is now ready to perform the summarization task using the BERT model,
simplifying the complex processes of embedding sentences and clustering into
an accessible interface.
5. Generate and print a summary.
```python
summary = bert_model(input_text)
print(summary)
```
Your input text is processed by the bert_model instance, which then returns a
summarized version. This demonstrates the power of Python's high-level
libraries in enabling complex operations with minimal code.
6. Create `requirements.txt`. The sample application already contains the
`requirements.txt` file to specify the necessary modules that the
application imports. Open `requirements.txt` in a code or text editor to
explore its contents.
```text
...
# 04 text_summarization
bert-extractive-summarizer==0.10.1
...
torch==2.1.2
```
The `bert-extractive-summarizer` and `torch` modules are required for the
text summarization application. The summarizer module generates a summary of
the input text. This requires PyTorch because the underlying BERT model,
which is used for generating the summary, is implemented in PyTorch.
## Explore the application environment
You'll use Docker to run the application in a container. Docker lets you
containerize the application, providing a consistent and isolated environment
for running it. This means the application will operate as intended within its
Docker container, regardless of the underlying system differences.
To run the application in a container, a Dockerfile is required. A Dockerfile is
a text document that contains all the commands you would call on the command
line to assemble an image. An image is a read-only template with instructions
for creating a Docker container.
The sample application already contains a `Dockerfile`. Open the `Dockerfile` in a code or text editor to explore its contents.
The following steps explain each part of the `Dockerfile`. For more details, see the [Dockerfile reference](/reference/dockerfile/).
1. Specify the base image.
```dockerfile
FROM python:3.8-slim
```
This command sets the foundation for the build. `python:3.8-slim` is a
lightweight version of the Python 3.8 image, optimized for size and speed.
Using this slim image reduces the overall size of your Docker image, leading
to quicker downloads and less surface area for security vulnerabilities. This
is particularly useful for a Python-based application where you might not
need the full standard Python image.
2. Set the working directory.
```dockerfile
WORKDIR /app
```
`WORKDIR` sets the current working directory within the Docker image. By
setting it to `/app`, you ensure that all subsequent commands in the
Dockerfile (like `COPY` and `RUN`) are executed in this directory. This also
helps in organizing your Docker image, as all application-related files are
contained in a specific directory.
3. Copy the requirements file into the image.
```dockerfile
COPY requirements.txt /app
```
The `COPY` command transfers the `requirements.txt` file from
your local machine into the Docker image. This file lists all Python
dependencies required by the application. Copying it into the container
lets the next command (`RUN pip install`) to install these dependencies
inside the image environment.
4. Install the Python dependencies in the image.
```dockerfile
RUN pip install --no-cache-dir -r requirements.txt
```
This line uses `pip`, Python's package installer, to install the packages
listed in `requirements.txt`. The `--no-cache-dir` option disables
the cache, which reduces the size of the Docker image by not storing the
unnecessary cache data.
5. Run additional commands.
```dockerfile
RUN python -m spacy download en_core_web_sm
```
This step is specific to NLP applications that require the spaCy library. It
downloads the `en_core_web_sm` model, which is a small English language model
for spaCy. While not needed for this app, it's included for compatibility
with other NLP applications that might use this Dockerfile.
6. Copy the application code into the image.
```dockerfile
COPY *.py /app
COPY entrypoint.sh /app
```
These commands copy your Python scripts and the `entrypoint.sh` script into the image's `/app` directory. This is crucial because the container needs these scripts to run the application. The `entrypoint.sh` script is particularly important as it dictates how the application starts inside the container.
7. Set permissions for the `entrypoint.sh` script.
```dockerfile
RUN chmod +x /app/entrypoint.sh
```
This command modifies the file permissions of `entrypoint.sh`, making it
executable. This step is necessary to ensure that the Docker container can
run this script to start the application.
8. Set the entry point.
```dockerfile
ENTRYPOINT ["/app/entrypoint.sh"]
```
The `ENTRYPOINT` instruction configures the container to run `entrypoint.sh`
as its default executable. This means that when the container starts, it
automatically executes the script.
You can explore the `entrypoint.sh` script by opening it in a code or text
editor. As the sample contains several applications, the script lets you
specify which application to run when the container starts.
## Run the application
To run the application using Docker:
1. Build the image.
In a terminal, run the following command inside the directory of where the `Dockerfile` is located.
```console
$ docker build -t basic-nlp .
```
The following is a break down of the command:
- `docker build`: This is the primary command used to build a Docker image
from a Dockerfile and a context. The context is typically a set of files at
a specified location, often the directory containing the Dockerfile.
- `-t basic-nlp`: This is an option for tagging the image. The `-t` flag
stands for tag. It assigns a name to the image, which in this case is
`basic-nlp`. Tags are a convenient way to reference images later,
especially when pushing them to a registry or running containers.
- `.`: This is the last part of the command and specifies the build context.
The period (`.`) denotes the current directory. Docker will look for a
Dockerfile in this directory. The build context (the current directory, in
this case) is sent to the Docker daemon to enable the build. It includes
all the files and subdirectories in the specified directory.
For more details, see the [docker build CLI reference](/reference/cli/docker/buildx/build/).
Docker outputs several logs to your console as it builds the image. You'll
see it download and install the dependencies. Depending on your network
connection, this may take several minutes. Docker does have a caching
feature, so subsequent builds can be faster. The console will
return to the prompt when it's complete.
2. Run the image as a container.
In a terminal, run the following command.
```console
$ docker run -it basic-nlp 04_text_summarization.py
```
The following is a break down of the command:
- `docker run`: This is the primary command used to run a new container from
a Docker image.
- `-it`: This is a combination of two options:
- `-i` or `--interactive`: This keeps the standard input (STDIN) open even
if not attached. It lets the container remain running in the
foreground and be interactive.
- `-t` or `--tty`: This allocates a pseudo-TTY, essentially simulating a
terminal, like a command prompt or a shell. It's what lets you
interact with the application inside the container.
- `basic-nlp`: This specifies the name of the Docker image to use for
creating the container. In this case, it's the image named `basic-nlp` that
you created with the `docker build` command.
- `04_text_summarization.py`: This is the script you want to run inside the
Docker container. It gets passed to the `entrypoint.sh` script, which runs
it when the container starts.
For more details, see the [docker run CLI reference](/reference/cli/docker/container/run/).
> [!NOTE]
>
> For Windows users, you may get an error when running the container. Verify
> that the line endings in the `entrypoint.sh` are `LF` (`\n`) and not `CRLF` (`\r\n`),
> then rebuild the image. For more details, see [Avoid unexpected syntax errors, use Unix style line endings for files in containers](/desktop/troubleshoot-and-support/troubleshoot/topics/#Unexpected-syntax-errors-use-Unix-style-line endings-for-files-in-containers).
You will see the following in your console after the container starts.
```console
Enter the text for summarization (type 'exit' to end):
```
3. Test the application.
Enter some text to get the text summarization.
```console
Enter the text for summarization (type 'exit' to end): Artificial intelligence (AI) is a branch of computer science that aims to create machines capable of intelligent behavior. These machines are designed to mimic human cognitive functions such as learning, problem-solving, and decision-making. AI technologies can be classified into two main types: narrow or weak AI, which is designed for a particular task, and general or strong AI, which possesses the ability to understand, learn, and apply knowledge across various domains. One of the most popular approaches in AI is machine learning, where algorithms are trained on large datasets to recognize patterns and make predictions.
Artificial intelligence (AI) is a branch of computer science that aims to create machines capable of intelligent behavior. These machines are designed to mimic human cognitive functions such as learning, problem-solving, and decision-making.
```
## Summary
In this guide, you learned how to build and run a text summarization
application. You learned how to build the application using Python with Bert
Extractive Summarizer, and then set up the environment and run the application
using Docker.
Related information:
- [Docker CLI reference](/reference/cli/docker/)
- [Dockerfile reference](/reference/dockerfile/)
- [Bert Extractive Summarizer](https://github.com/dmmiller612/bert-extractive-summarizer)
- [PyTorch](https://pytorch.org/)
- [Python documentation](https://docs.python.org/3/)
## Next steps
Explore more [natural language processing guides](./_index.md).