This week we received very special news: our CTO, Lorenz Richter, was honored with an award for one of the 5 best scientific articles (out of more than 5,000 submitted articles) at this year's very prestigious 'International Conference on Machine Learning'. We have been working for years to bring the latest findings from ML research into the enterprise and are glad that this award underlines our ambitions.
Our Q2 newsletter is otherwise thematically all about NLP, i.e. machine processing of natural language. Machine Learning has been playing an increasingly important role here for years, and we see this in our projects as well. Reason enough to illustrate this development with two current customer projects and our new Question Answering demo.
For a legal-tech company in the field of tenancy law, we successfully implemented the extraction and automatic verification of very differently formatted service charge statements, which tenants can upload as PDFs. In order to extract the correct information despite the variety of forms of the statements, various NLP and computer vision methods were used. More information can be found in our first customer project example.
If you want to register a trade, you have to specify the code of the economic sector to which your activity belongs. For example, if you want to "sell things on eBay", you have to enter the sector code "47.91.9 - Other mail order and internet retail". For those who do not know this by heart, our neural search helps to establish the semantic relationship between the search query and the sector code description. For this purpose, we trained and adapted a neural language model and integrated it into the website of the trade office in NRW. Read more in our second customer project example.
The NLP trend "neural language models" with the focus of a profitable use in the corporate environment with possible practical challenges was recently also the topic of our webinar, which took place in cooperation with AI-Spektrum. If you missed it, you can watch the recording of the webinar and the slides here.
You can try out a neural language model directly on our website in our Question Answering demo. Here, the model tries to answer questions about a text by returning the appropriate part of the text as an answer. There are sample texts and questions, but anyone can also enter their own texts and questions to test the pre-trained model's understanding of natural language.
One possible application of question answering technology, namely enhancing and improving search applications of any kind, is described in our blog article "Enhancing Search with Question Answering".
We are planning our next newsletter for the end of Q3. We look forward to your feedback and any topic requests for next time.
Best regards Philipp Jackmuth