To Sum or Not to Sum

Why Summarize?

How does it work?

  • Taking key pieces of information from the sources and reorder them to present a short chunk of text that covers all the topics. This is known as Extractive Summarization.
  • The other & more “human” is to go through the data first & then writing a summary from scratch, rephrasing/paraphrasing as per requirement. This is the Abstractive Summarization method. Ideally these would seem better but they are more difficult to build & operate.

What is a good summary?

Where do we come in this?

Language Models

  1. Capturing the entire context of the data without leaving any information
  2. Creating a grammatically correct new summary out of the captured information

GPTs

  1. You’re 1st training a simple model on lots of text that essentially train it in the English language. It generates text with context to the input.
  2. Followed by this, we provide labeled data in context to the task at our hand.
  • Instead of P(Output | Input), it had a newer objective of P(Output | Input, Task). This is called task conditioning, where the task is expected to different output for the same input based on the kind of task. But instead of feeding the task at an architectural level, it was fed a natural language sequence.
  • Due to this task conditioning, the model was capable of zero-shot learning, where the nature of the task is understood by the model purely based on task instructions.

Architecture & Implementation

Performance

Use-Cases

Information Extraction via News Media

Social Media Posts/Tweets

Financial & Legal Documents

Conclusion

About Connexun

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Connexun | news api

Connexun is the ultimate AI news engine — turning unstructured news content into multi-purpose actionable data.