Leap NLP/NLU

For quick access to information on the internet, data banks and private databases require the use of powerful Chat Bot AI engines.

For quick access to information on the internet, data banks and private databases require the use of powerful Chat Bot AI engines.

For quick access to information on the internet, data banks and private databases require the use of powerful Chat Bot AI engines.
No matter how fast and accurate thesearch is, the search engine must be able to deal with the complexities of certain languages such as Arabic.
Arabic is not only written from right to left but it also has a very complicated morphological structure and parsing system. In different Arab countries writers and scholars spell some Arabic words differently. Our NLP Engine is designed to deal in an intelligent way, with all the complexities of the Arabic language.
The support for Arabic NLU relies on the use of the Starspace general purpose neural embedding model. We use an intent classifier that trains word embeddings from scratch. It is inherently language independent and is not reliant on having good word embeddings for a certain language to have good results.
For Arabic sentiment analysis, the model is based on deep neural network. The model was trained during 200 epochs, with a dropout of 0.5, learning rate of 0.001 and 300 hidden layers. If the model doesn’t improve during 15 consecutive epochs, the training stops, and only the best-found model is saved.