Skip to main content

From Text to Numbers: Exploring Sentence Embeddings w/ SentenceTransformers

Level:
intermediate
Duration:
30 minutes

Abstract

In the realm of natural language processing, the concept of embedding plays a pivotal role in understanding textual data. In this talk, we’ll explore how text is turned into numbers for computers to understand, a process known as embedding. Particularly, we’ll focus on sentence embeddings.

Using easy-to-follow examples, we’ll break down the different parts of the SentenceTransformers, which is a Python framework for state-of-the-art sentence, text and image embeddings.

We’ll discuss why making these embeddings accurately is important. We’ll talk about how to measure the quality of embeddings, whether you have (some) labels or not. We’ll specifically concentrate on the application scenario within conversational data.

We’ll also talk about the best-performing models for making these embeddings and how to fine-tune them with your own data. This talk aims to make these concepts easy to understand, so everyone can use them effectively.


The speaker

Adam Zíka

Adam Zíka

At Salted CX, my role as a Machine Learning engineer revolves around specializing in Natural Language Processing. Specifically, I utilize Transformer models to gain insights into the operations of contact centers, enhancing visibility and understanding. I hold an Engineering Doctoral (EngD) degree in Data Science from the Technical University of Eindhoven.