All Collections
Getting started
Creating your own AI in 4 steps
Creating your own AI in 4 steps

A brief overview of the Labelf platform

Ted Tigerschiöld avatar
Written by Ted Tigerschiöld
Updated over a week ago

Labelf is an end-to-end platform for building and deploying custom AI models for text. Let us look into what we mean by that. The AI (Artificial Intelligence) models that you will build using Labelf are text classifiers. A text classifier is an AI model that can look at a text and classify it. The texts you classify can be emails, support tickets, reviews, social media mentions or even internal reports. Let us do this in four easy steps.

Step 1 - Identify and upload your text data

In order build an accurate AI model you need to start with texts for the model to train on. Generally speaking you want to find texts that are as similar as possible to the type of of texts that you re going to use the model on. For example if you are building a model for classifying incoming customer support tickets, it is best to train the model on your historical tickets. Here you can read more on what makes a good dataset

Once you have found your dataset you simply upload it to Labelf . After that you will be asked to choose the column in the spreadsheet where your texts are located.

Step 2 - Set up your model

Now it is time to set up your AI Model, you start by giving it a name. The next thing you do is to choose which data source to use to train your model on (more on what training is in the next step). You now choose the dataset you uploaded.

Now you will be asked if your data is already labeled for the type of model you are building. This is an important question as it decides if Labelf will use labels that are already in your data file or if you will create new labels and train the model manually. Note that you can use the labels you have in your file even if not all rows are labeled. If your data is already labeled you click "Train directly on labels from an existing column" and skip on to step 4. If your data is not labeled it is not time to set up your own custom labels.

You do this by deciding the labels you want to categorise your texts with (here you can read more about data labels). For a simple sentiment analysis this could be the two labels "Positive" and "Negative". We start by creating the label "Positive" and assign a colour to it. Next you will be asked to give some examples for the label you just created. This helps the model to quickly recognize which type of texts should be associated with the label. If you cannot come up with any examples you can just press skip.

Now you set up the last Label and once that is done we can move on to start training the model.

Step 3 - Train the Model

You can skip this step if your data is already labeled.

Now it is time to train the model so that is is able to classify your texts with the right labels. When you click "Start teaching" you will be presented with texts from your data source. Now you simply read the text and assign the correct label to it. If you realise that you have missed to add a data label you can simply do that by clicking "add a label". Training the model can feel a bit tedious after a while but it is a very important step in building your AI. Once you have labeled around 50-100 texts the model will start to present you with bulk suggestions labeled texts. Once these appear you now accept or correct the models suggestions of the labels.

Search in your data

A powerful way to speed up the teaching process is by using the search bar. Search enables you to quickly find texts in your data that are easy to label. This could be through a certain keyword that you know will appear in texts of a certain label.

Step 4 - Evaluate performance and Deploy model

Once you have labeled for a while it is time to have a look at how good the model is performing. You do this by navigating to the Metrics page. Here you get an overview of how good the model is at doing its job at the moment. If you are unsure on how to intepret the metrics here is a blog post describing some of the concepts. If you feel that the model is not really performing as well as you want it to do it is best to go back to the training step and assist the model by labeling more texts for it.

Once you are satisfied with the performance of the model it is time to deploy it. In the integrations section you can find all the available integrations you can connect your model to.

In the API section can generate an API key to access your model. More about API access can be found here

Did this answer your question?