Using natural language understanding (NLU) to analyze text

If you’re a product owner trying to uncover insights from Voice of the Customer data, there are a few things to consider before choosing the right Saas. Here’s an in-depth analysis of all the comparative features of HumanFirst and Monkeylearn.


MonkeyLearn is a text analysis software where users can create customized text classification and extraction analysis by training machine learning models such as sentiment analysis, topic detection, keyword extraction, and more.

HumanFirst is a data-centric productivity suite for natural language that allows its users to discover, label, organize, and optimize AI training datasets. HumanFirst users can scale their models to thousands of intents trained across thousands of utterances, with flexible hierarchies that can be reused across projects, teams, and workspaces.

Both are similar in concept and provide an analysis of text and utterance data, but differentiate depending on the type of analysis your team is looking to perform.

Pre-Trained Vs. Customized Models

Monkeylearn uses pre-trained models to search through and automatically categorize text data. Examples of pre-trained models include urgency detection and sentiment analysis, which can be run on Google sheets via API:

MonkeyLearn integrates with Google Sheets using pre-trained models that will automatically categorize your data.

HumanFirst, on the other hand, focuses on all cases where basic sentiment analysis is simply not actionable enough. With HumanFirst, you can train a unique NLU model, giving you the opportunity to discover/classify topics and intents to understand your corpus of data more deeply. Creating an NLU model is quite simple using HF’s full-text semantic search, model-assisted active learning labeling workflows, and interactive real-time clustering. What you’ll end up with is a flexible, modular hierarchy of parent and child intents made up of your utterance data:

With HumanFirst, intents are organized into modular hierarchies trained with your utterance data. ‌‌

The ability to dig into feedback more granularly is important for exposing issues and opportunities, whereas basic sentiment analysis might not be that actionable in terms of product roadmap. This kind of specific detection is key in understanding and nurturing precise, intricate, and highly developed datasets (more on that later).

Whether your team needs pre-trained or custom models depends on the specificity of your industry and needs, and is entirely up to your team to decide.

Level of Analysis

They are also different in the level of analysis. MonkeyLearn provides more of an overview of your data, which is less actionable by nature. With their tool, you can expose whether your product is being well received by your customers. You also have the ability to identify trends in your data. Their data visualization dashboard summarizes metrics like intent, sentiment over time, categories, and sentiment by category:

MonkeyLearn’s visual dashboard

HumanFirst tends to get more granular in its analysis, built to deliver more actionable insights.  You can accelerate the discovery of topics and intents in your data with real-time, configurable clustering on all of your unstructured data, with the ability to toggle a granularity bar based on how specific you want your clusters to be (which is just a group of semantically similar utterances), uncovering all customer intents. This approach is meant to capture the long-tail of requests:

HumanFirst’s features also reflect that it was designed to be used by enterprise-grade clients. It is customizable, flexible, and developer-friendly. Extensible data sources and reusable datasets are available, with workspace revisions, diffs, revert & cloning. HumanFirst also offers on-prem deployment (fully air-gapped possible too).

MonkeyLearn was constructed to provide an overview of intents and trends, whereas HumanFirst was created to capture the long-tail of intents and requests.

Data Sources

Your choice of tooling may differ depending on the type of data sources you are analyzing. MonkeyLearn is optimized to work with flat lists of single sentence utterances (like the Google Sheets analysis example above). If this is sufficient for your needs, MonkeyLearn can do the job.

If you wish to extend to other types of media, such as multi-turn dialog and long-form content (i.e: emails, tickets, etc.) where multiple labels are found, then you may want to go with HumanFirst.

In conclusion:

Use MonkeyLearn if you:

  • Don't want to train your own model
  • Want to do basic sentiment analysis
  • Only want to work with flat lists of single sentence utterances

Use HumanFirst if you:

  • Want to train a custom NLU model to search & explore your data
  • Want a deeper analysis to capture the long-tail of intents
  • You’re looking for an enterprise-grade solution
  • Want to extend your analysis to other types of media

All-in-all, HumanFirst is a much broader data-centric solution, aimed at curating and building training data for an NLU model from unstructured data, while MonkeyLearn is a text analysis tool to classify and capture sentiment and topic trends.