Conversational system for recognizing, understanding, and acting on multiple intents and hypotheses
A conversational system that recognizes, understands, and acts on multiple intents that may be explicit or implicit during conversations with humans. During a conversation, one or more utterances are received and processed through a plurality of machine learning algorithms to establish precise meanings, additional intentions, and alternative hypothesis. Using a combination of machine learning algorithms and datastores, conversations are interpreted as intended and may diverge where needed or desired, delivering a more useful, natural, and human-like dialogue between machines and people.
1. A system for recognizing, understanding, and acting on multiple intents, comprising:
a computing device comprising a memory and a processor;
a persistent global intent knowledge graph, wherein the persistent global intent knowledge graph is a directed computational graph of intent trees;
a slot grammar parser comprising a first plurality of programming instructions stored in the memory of, and operating on the processor of, the computing device, wherein the first plurality of programming instructions, when operating on the processor, causes the computing device to:
receive an utterance;
perform deep parsing on the utterance to generate a first parse tree, wherein the first parse tree comprises a logical analysis of the utterance; and
perform shallow parsing on the utterance to generate a second parse tree, wherein the second parse tree comprises a grammatical analysis of the utterance; and
a predicate argument structure builder comprising a second plurality of programming instructions stored in the memory of, and operating on the processor of, the computing device, wherein the second plurality of programming instructions, when operating on the processor, causes the computing device to:
collapse the first and second parse trees into a plurality of possible core semantic intents;
populate a node in the persistent global intent knowledge graph for each intent in the plurality of possible core semantic intents;
determine a most likely intent by ranking each intent in the plurality of possible core semantic intents and selecting the top ranked intent; and
generate a response using natural language processing and the top ranked intent.
2. A method for recognizing, understanding, and acting on multiple intents, comprising the steps of:
receiving an utterance;
performing deep parsing on the utterance to generate a first parse tree, wherein the first parse tree comprises a logical analysis of the utterance;
performing shallow parsing on the utterance to generate a second parse tree, wherein the second parse tree comprises a grammatical analysis of the utterance;
collapsing the first and second parse trees into a plurality of possible core semantic intents;
populating a node in a persistent global intent knowledge graph for each intent in the plurality of possible core semantic intents;
determining a most likely intent by ranking each intent in the plurality of possible core semantic intents and selecting the top ranked intent; and
generating a response using the top ranked intent.
3. The method of claim 2 , further comprising the step of creating, from external data sources and sensors, a plurality of relevance nodes in the knowledge graph.