In 2018, we have been working on advancing our NLP (Natural Language Processing) and Sentiment Analysis to address issues that businesses have been trying to overcome, such as getting in touch with customer behavior, gathering market intelligence and advancing customer experience with Chatbots.
What differentiates us in our ML (Machine Learning) and NLP development endeavors is that we cover languages that are not world-spread. This includes the Balkan languages (Slavic languages) and Scandinavian languages as well.
For English, French, German or Spanish, for example, it is much easier. This is especially the case with English which, as a lingua franca, has many available resources and hence it is easier to work with. On the other hand, there is almost nothing for Macedonian, and that is a challenge!
As part of data processing, what makes NLP more complex is the fact that languages are constantly evolving and it is difficult to keep pace. The way languages work and how they are structured is much different from the ways machines are structured.
At the moment, machines understand numbers and the aim of the NLP frameworks is to teach them how to understand language. Or, in other words, to translate a human language into a computer language.
Currently, we have developed a Media Intelligence Platform that works on Sentiment Analysis, text similarity, summarization and generating highlights from a certain text. This framework is applicable to all languages. The challenge, however, is that NLP can make perfect sense to machines but not as much to humans. It is delicate to validate the testing results because you always need to have a human factor to tell whether they are valid or not.
The accuracy, of course, depends on the training data set and may be conducted on tweets, articles, film reviews, academic papers and so on. The same applies to the extent of accuracy when it conducts a summary. At the moment, we have over 90% accuracy in our endeavors.
As for the implications, it is a very useful tool for people that need to conduct summaries of a given text, classify documents and so forth. Who can benefit? Everyone that conducts text-processing, news clustering, article categorization, text similarity detection, news credibility etc. It can extract important keywords, tags, thematic nuances – all the important elements of text analysis.
It is also beneficial in Chatbots development. Conversational Chatbots are created with the help of NLP and NLU. The first phase is the NLU (natural-language understanding) followed by the second, more advanced phase, text generation, i.e. the machine to give feedback on an allocated input. The last phase is to create a virtual assistant, or a conversational bot, with whom you can interact. Such bots are immensely NLP-powered.
And since it is language agnostic, the framework can also work on translation.
In 2019, we will continue to work on ML and NLP-development and create more advanced solutions that will have a bigger impact on business development and growth.