All articles

What’s New With Applica RTA v1.3

As we start this new fall season, it’s a great time to refocus on how organizations can become more efficient during this unprecedented year of disruptions. Applica is pleased to announce the release of Applica RTA v1.3, which has some distinct enhancements that will directly impact the quality of a user’s document processing.

The first update is our new Extraction feature, which includes an innovative type of AI-based information retrieval from unstructured and semi-structured documents. Now that Applica RTA Studio supports bot Classification and Extraction, the platform is able to support a wider array of business cases.

For reference, a typical user would continue to leverage both Classification and Extraction depending on their task(s) in Applica RTA Studio. Here is a breakdown of typical use cases:  

We love to keep things simple for our end users. To train a custom AI model based on your business data, all you need to do is prepare a small number of documents with a metadata file of correct values, click a single button, and let RTA Studio do the work.

Moreover, we’ve created algorithms that do automatic tagging of documents, allowing the training of AI models without the need of burdensome manual annotation.

This new Extraction feature is based on NER (named entity recognition) techniques. We support multiple data types (e.g.: various date formats, amounts, etc.) and plan to implement even more in the future. Applica RTA Studio is able to extract information in any format and normalize it to the desired option. Most importantly, instead of finding all texts (or rather entities) our AI model retrieves for the one that represents what are you are actually looking for.

Standard NER vs. Applica Extraction

Standard NER (named entity recognition)

Applica’s Extraction algorithm

Applica RTA will determine a single answer, instead of finding random dates (or things that look like dates) in the document.

As previously mentioned, the user can determine whether to use Classification or Extraction for a given task. Classification is a good choice when you don’t have information directly in the document and you want to classify by topics. For example, someone is making a complaint against an organization but they don’t explicitly state “complaint”, therefore you can still classify this by context and assign a class to each document (e.g.: complaint, billing issue, etc.).

Additionally, it depends upon the classes needed for your task(s). If you have more than 1,000 classes it requires a vast number of examples, as the machine learning needs 10 examples for each class to self-learn, and thus Extraction is the best option here. If you have 10-50 classes then Classification is a typically a better choice, though users can always experiment to see which versions gives the best results based on their dataset.

Future releases will also include enhancements such as:

  • Extraction with Context
  • Extraction with Role Classifier
  • Simple Entity Extraction

The second update for v1.3 is the RTABot import and export feature, of which there are three main use cases:

  1. Transfer of RTABots between separate client environments: typically to adhere to security or data governance policies.
  2. Import of RTABots created by Applica’s R&D team: an RTAbot trained and refined by our team of data scientists can be very effective for complex models, as they eliminate the need for self-service users to do multiple iterations themselves and provide guaranteed precision and efficiency.
  3. Backup: depending on your installation and needs, it might be a quicker and more convenient backup solution – every bot can be exported and downloaded onto your computer.   

RTABot import / export feature example

Should you have any questions, please feel free to contact me for further information.