Text AI Updates Drive Faster Business Value

How can you save time in understanding the impact of language when working with text in ML models? With tens of thousands of Text AI projects, DataRobot has helped organizations unlock insights from text and generate predictions with text models—from assisting with customer support ticket triage to predicting real estate sale prices. Continuing to build on previously released Text AI capabilities,the DataRobot AI Platform introduces new features to help with language detection, blueprint optimization, and text prediction explanations that help customers quickly build and understand text driven models.

Enhanced Autopilot Language Detection and Automatic Hyperparameter Tuning

Language detection has been a staple of DataRobot when working with text, and now we’ve upgraded the capability. The turbocharged language detection feature now uses a deep learning algorithm to identify the language of text even more precisely. Not only that, but we’ve also added heuristics throughout the platform to optimize generated blueprints for the detected text. No need to spend weeks trying to fine tune models. DataRobot produces the most optimized blueprints and squeezes the greatest accuracy out of our extensive library of models.

The dataset below contains French Amazon^® product reviews where DataRobot correctly identified the language as French. Parameters were also automatically adjusted to optimize the blueprint for the French language.

Immediate Insights with Text Prediction Explanations 

DataRobot makes it faster to generate accurate text models and offers a large step forward in helping users understand the impact of the text on a model’s predictions by introducing text prediction explanations.

With prediction explanations, a user can identify the impact of a feature on a model’s predictions—both in terms of whether it is a negative or positive impact and the relative strength. However, this is not necessarily sufficient when it comes to text features. Text and human language is extremely complex, fluid, and inconsistent with contextual nuances, ambiguity, and many more complications that are involved in understanding text. 

Because language is so complex, it is critically important to be able to explain how a machine learning model interprets text to humans. With this new capability, users can better understand and trust the model’s results. Now users can validate the importance the model places on words, including both negative and positive impacts. Also, users can understand a model’s shortcomings when working with specific words in the broader context. An example of this would be a model that predicts hiring candidacy success. If text prediction explanations identify a specific name as extremely impactful, it may be a sign that the name is skewing the results of the model and should actually be removed as a datapoint to remove bias. Additionally, identifying impactful words can help users to zero in on important concepts that may affect the result of the specific problem they are attempting to solve.

Text prediction explanations save users time by surfacing a level of granularity that shows the importance of each word. Without this capability, users have to read the full text to achieve the same understanding, resulting in a massive loss in the time and value of using a machine learning model in the first place.

Continuing with the example of reviewing French Amazon reviews, DataRobot insights have identified both text features as having a relatively positive impact on predictions.

Clicking on the new orange pop up button will reveal text prediction explanations for the text feature that was selected.

Here’s what happens when a user opens text prediction explanations for the text feature.

Using this feature, users can now see the words that are most impactful to the model’s predictions. In this specific case, “Sony” is one of the words that’s highlighted as having relatively high impact. So, the Amazon seller of the product could use this insight to take a closer look at Sony products and how that relates to customer satisfaction.

Get Your Hands on These Text AI Upgrades Today

The DataRobot AI Platform customers can get started with these Text AI upgrades right away. The enhanced language detection and hyperparameter tuning is available in GA, and text prediction explanations are available in Public Preview with the July release.  

For more information, visit DataRobot documentation and schedule a demo.

About the author

Jon Chang

Senior Product Manager at DataRobot

Jon is a Senior Product Manager at DataRobot, focusing on product strategy in the deep learning space. Having spent a decade in product management, he has an absolute dedication to building great products, delivering value to customers, and a passion for everything AI. Prior to DataRobot, he provided digital product strategy consulting services, built fintech digital products, and was engaged in a weather analytics startup.

Meet Jon Chang