AutoNLU is a framework to train and use state-of-the-art Natural Language Understanding (NLU) solutions to practical text analysis problems in as few as 5 lines of code.
Simple API
Easy to use API that just works. Designed for developers, engineers and data scientists to achieve the most in a few simple lines of code. As much automation as possible and as flexible as needed.
Fast
Using clever batch preprocessing as well as optimized mixed-precision, AutoNLU is up to 5x faster than naive implementations of transformer models.
Always Up-to-Date in NLP
NLP is the fastest moving field in AI. With AutoNLU you don’t have to worry about the latest research to get state-of-the-art results - this is on us! You benefit from advances in the field simply by using the latest version of AutoNLU.
Extensively Tested
We use our extensive database of industry datasets to test AutoNLU and ensure it produces high-quality results for this broad set of cases.
Deep Learning First
AutoNLU makes state-of-the-art deep learning in NLP accessible for everyone without having to worry about the complexities of dealing with modern deep learning algorithms.
Interoperability
AutoNLU, DeepOpinion Studio and other DeepOpinion products are fully interoperable and models can be easily exchanged between the platforms.
C O D E E X A M P L E S
Create Practical NLU Solutions
in Few Lines of Code
Flexible
AutoNLU adapts to your training data and produces the best possible model to solve your problem automatically.
Powerful
AutoNLU solves e.g. all classification problems with the same five lines of code, even challenging “dual tasks” such as aspect-based sentiment or entity-intent detection.
Plug & Play
In two lines of code you can instantly use trained task models to predict your data and get results, e.g. by connecting to our Model Library.
Use Pre-Trained Task Model
Simply load pre-trained models for specific tasks and industries. E.g. start analyzing employee feedback from a survey on specific topics and sentiments with only two lines of code. Models for a variety of domains and tasks are available in our Model Library in DeepOpinion Studio.
From zero to solution in under 2 minutes
Use models that were extensively tested in industry production settings
Many models for specific tasks are available
Train a Single-Label Model
If you want to train a model that simply assigns exactly one label to each piece of text, use a list of strings as the target during training and AutoNLU will do the right thing. During prediction, AutoNLU will return one label for each piece of text you send in. This is great for e.g. spam detection, sentiment analysis or hate speech detection.
Easy to use API
Many languages supported
Train a Multi-Class Model
If you would like to train a model that assigns an arbitrary number of classes to each piece of text, you can give AutoNLU a list of classes for each piece of text as a training target. AutoNLU again knows how to best treat your data and internally selects the best model architecture to fit your needs. This is great for e.g. topic detection or customer support ticket tagging/triaging.
Adapts to you specific use case
Choose from a wide range of domain specific base models
Train a Class-Label Model for Aspect-Based Sentiment Analysis
If you would like to train a model that detects an arbitrary number of classes for each piece of text and assigns exactly one label to each of those classes, you can easily give a list of tuples for each piece of text as the training target. This is great for complex “dual” classification tasks such as aspect-based sentiment analysis or entity-intent detection.
Wide range of tasks are supported
Optimized training of the problem you want to solve
K E E P I N T O U C H
Stay informed about progress in AutoNLU.
We value privacy. We will only send you AutoNLU updates.
Select from a wide range of pre-trained industry models in our Model Library, and/or fit an existing model to your specific domain in one line of code.
Less Data Labeling
Reduce the manual data labeling effort by 75% and significantly improve model quality by using the Active Learning feature in AutoNLU.
Use Latest Models
Additional to our Model Library all base models available in the amazing HuggingFace transformers model library can be used seamlessly in AutoNLU. Sending 🖤 to 🤗.
Use Pre-Trained Industry Model as a Base Model
Train an aspect-based sentiment analysis model using a base model that was fine-tuned for a specific domain to achieve higher accuracy. In this case, we will use a language model that was fine-tuned for household electronics. Using AutoNLU, all you have to do is to specify a different available base model.
Language models available for many industry relevant domains
Switch from a general to a domain specific model by changing one parameter
Fine-Tune Models to your Industry
If no fine-tuned model for your specific domain is available you can easily fine-tune your own model for your domain, as long as you have a decent amount of text data from your domain. This data does not have to be labeled. A simple text file with e.g. customer reviews from you domain is sufficient.
Fine tune on your domain in one line of code
Only a standard text file from your domain is needed
Use your own domain model for many different tasks
Less Data Labeling and Improved Quality with Active Learning
AutoNLU supports a method called active learning. In active learning, a trained model is used to select examples from a corpus that will provide maximal information when added to the training set.
In practice, this means that less data has to be manually labeled to achieve high model accuracy.
Less manual data labeling effort required
The model tells you what it needs to know to learn
Use Models from Other Sources
Also models from other sources can be used seamlessly in AutoNLU, e.g. all base models available in HuggingFace transformers model library (sending 🖤 to 🤗). For example, training a less resource-hungry version of a restaurant ABSA model can be achieved by using DistilBERT as the base model before fine-tuning to your domain-specific data and training for your task.
Choose from the great variety of publicly available models and architectures
Tailor your model architecture to the requirements you have by changing one parameter
K E E P I N T O U C H
Stay informed about progress in AutoNLU.
We value privacy. We will only send you AutoNLU updates.
AutoNLU allows you to train and use state-of-the-art NLP models in as few as 5 lines of code.
Production-ready
AutoNLU is used as the machine learning core of DeepOpinion Studio and is in production for many industry-specific text analysis problems for leading companies in various industries.
Ultra-fast prototyping
Get things done fast and iterate quickly. Focus on your use case and leave training a state-of-the-art model to AutoNLU.
Progress Reporter Callback for Prediction
When integrating AutoNLU in your existing software infrastructure you need to be informed about the current performance of the model during training or the progress of the prediction process when batch processing. All this information is accessible through callback functions.
AutoNLU gives you information about what is happening with an easy API
Production environment in mind
Stop Callback to Control Training and Prediction
Of course, AutoNLU is more flexible than the previous examples showed and can be controlled by a variety of callbacks and arguments.
As an example, we show code that will stop training after three hours if the model has not converged up to this point. The best model up to this point will be returned after training has been stopped.
High automation, but still flexible
Don’t won’t waste training effort when stopping early
