Making computers understand humans
Avaamo’s extensions to Recurrent Neural Network algorithms are aimed at maximizing accuracy and recall for varying levels of complexity on the dataset. These extensions include multiple dimension reductions for untagged, unstructured data covering both text and speech. This improves detection of false positives as well as domain-based accuracy resolution
Avaamo’s NLU Engine applies syntactic, semantic, and stochastic processing to distill and discover the purpose behind the user’s message. To speed entity extraction and matching, Avaamo’s NLU engine comes pre-built with 1000’s of entities, but also enables developers to create their custom multi-hierarchical entity definitions with associated dependencies.
Avaamo NLU engine can not only detect the appropriate tone and sentiment of the user query but use additional dimensions including user conversation history, goals achieved, user feedback and accuracy of prediction to make a more contextual prediction on the user tone. Bot developers can also define appropriate actions based on the user tone- auto transferring to an agent when the use is frustrated or providing them with a relevant offer based on past interactions.
Avaamo’s NLU engine supports multiple languages by building segmentation support for different languages and allowing the bot developers to specify the training text and utterances in the appropriate language. In addition to support languages that are based on the Latin text (English, Spanish, Portuguese, French etc) our segmentation support extends to languages like Hindi, Simplified Chinese, Arabic, etc.
Avaamo’s NLU Engine evaluates all possible meanings and interpretations to determine what the user is asking.
Our recurrent neural networks and proprietary gating extensions have been developed and trained to understand syntax, tone, behavior and goals.
Based on all available information and supported by our cognitive platform our NLU engine executes judgement-intensive tasks.