DEV Community

Cover image for Sequence labelling in Python (part 1)
Antonio Feregrino
Antonio Feregrino

Posted on

Sequence labelling in Python (part 1)

Why?

I was looking for a cool project to practice sequence labelling with Python so... there is this Mexican website called VuelaX, in it, flight offers are shown. Most of the offers follow a simple pattern: Destination - Origin - Price - Extras, while extracting this may seem easy for a regular expression, it is not as there are many patterns. It would be tough for us to cover them all.

I know it is not ideal to work in a foreign language, but bear with me, as the same techniques could be applied in your language of choice.

The idea is to create a tagger that will be able to extract this information. However, one first tag is to identify the information that we want to extract. Following the pattern described above:

  • o: Origin
  • d: Destination
  • s: Separator token
  • p: Price
  • f: Flag
  • n: Irrelevant token
Text d o p n
¡CUN a Holanda $8,885! Sin escala EE.UU CUN Holanda 8,885 Sin escala EE.UU
¡CDMX a Noruega $10,061! (Y agrega 9 noches de hotel por $7,890!) CDMX Noruega 10,061 Y agrega 9 noches de hotel por $7,890!
¡Todo México a Pisa, Toscana Italia $12,915! Sin escala EE.UU (Y por $3,975 agrega 13 noches hotel) México Pisa, Toscana Italia 12,915 Sin escala EE.UU (Y por $3,975 agrega 13 noches hotel)

CRFs in Python

If you are familiar with data science, you know this is known as a sequence labelling problem. While there are various ways to approach it, in this post, I will show you one that uses a statistical model known as Conditional Random Fields. Having said that, I will not delve too much into details, so if you want to learn more about CRFs you are on your own; I will show you a practical way to use it with a Python implementation.

Getting some data

To start, I scraped the offer titles data from the page mentioned above. I will not detail how I did it since it is pretty straightforward to find a tutorial on web scraping on the web. If you don't feel like spending some time scraping a website, I collected some data in a CSV file that you can access now here.

This tutorial will be divided into other 4 parts:

Hopefully, you will follow along and will ask some questions if you have by leaving a comment here or contacting me on twitter via @io_exception.

Top comments (1)

Collapse
 
rodyoukai profile image
Rodrigo Cuéllar Hidalgo

Is there a little error in your labeled data example, for example in the first text CUN is the Origin and Holanda is the destination, this happen in al rows...