At Chatterbox Labs, we do not believe that products should be static and fixed - not only does online language change but so do market requirements. What if the data source changes; or you need a different segmentation; or you move into a new market with a different language? You need to be able to address these with new and updated products, you need to do it quickly and you don't want it to consume your team impacting your existing dev schedule.
We designed our NLP Framework to pull together all the building blocks required for building and training Machine Learning & NLP products for text. It is extensible and it is flexible. Our products are not build on cumbersome, static rules - we use statistical machine learning for short social texts.
The powerful framework sits within your organisation - instead of hiring an army of data scientists (who, incidentally, cost up to $250k per annum) to develop your products from scratch, the extensible framework allows you to shortcut this and exploit our years in academia.
We eat our own dog food: we used our NLP Framework to build our lead product: Path 2 Purchase. We also use the framework to customise Path 2 Purchase to generate industry specific classification - a process that takes only days and definitely does not require recoding the software.
So what does the framework do? It manages the entire lifecycle of building and training a text classification product, from training data management and in-house expert data annotation, through transforming a noisy multilingual text into an abstract machine representation with experimenting, testing & evaluation baked in, and finally to producing production quality language models ready to operate and classify text data in a real-time environment.
We also don't think that you should touch code if you don't have to. Build & customise products, experiment and evaluate using simple yaml configuration files (think of them as simple text files). If you do want to touch code, just touch the bit you need. It's all in a pipeline structure, so each step is autonomous. One step passes a context to the next which can be read from or modified. If you want to change a step or create a completely new step, just change the bit you need. It's just Python code, so any engineer can handle it.
There is no major deployment process either - we embrace open source & cross platform technology. Just pull our code, run the install script and check out the examples. You'll be up and running in no time.
For more information contact Andrew Watson, VP Strategic Alliances (email@example.com) or visit our website chatterbox.co.