Using semantic frames for intent classification
The present disclosure relates to chatbot systems, and more particularly, to techniques for identifying an intent for an utterance based on semantic framing. For an input utterance, a semantic frame is generated. The semantic frame includes semantically relevant grammatical relations and corresponding words identified in the utterance. The semantically relevant grammatical relations define context and relationships of words in the utterance. The semantic frame is used to identify an intent for the utterance, based on an intent model. The intent model maps features to corresponding words for a given intent. The semantic frame is compared to a plurality of intent models, and a best-matching intent model is used to identify the intent for the utterance.
1. A method for identifying an intent for an utterance, the method comprising:
receiving, by a computer system, the utterance from a user;
generating, by the computer system, a semantic frame for the utterance, wherein the semantic frame comprises one or more structured grammatical relations that define semantic roles of a plurality of words in the utterance in relation to one another;
determining, by the computer system, a plurality of scores for a respective plurality of intents based on matching of a particular relation in the semantic frame to a plurality of features associated with the plurality of intents, wherein the plurality of scores are determined based on an intent model, and wherein the plurality of scores are determined based on weights assigned to each of the respective features in the intent model;
identifying, by the computing system based on the scores, a particular intent, of the plurality of intents, as corresponding to the utterance;
generating, by the computing system, a response based on the identified intent; and
transmitting, by the computing system, the response as output.
2. The method of claim 1 , wherein the intent model is a multi-key map of features to words.
3. The method of claim 1 , wherein generating the semantic frame comprises:
generating a plurality of grammatical relations by labeling words to define relationships; and
aggregating the plurality of grammatical relations.
4. The method of claim 1 , wherein the relations correspond to parts of speech defining relationships between words.
5. A non-transitory computer readable medium storing a plurality of instructions executable by one or more processors for identifying an intent for an utterance, wherein the plurality of instructions, when executed by the one or more processors, causes the one or more processors to perform processing comprising:
receiving an utterance from a user;
generating a semantic frame for the utterance, wherein the semantic frame comprises one or more structured grammatical relations that define semantic roles of a plurality of words in the utterance in relation to one another;
determining a plurality of scores for a respective plurality of intents based on matching of a particular relation in the semantic frame to a plurality of features associated with the plurality of intents, wherein the plurality of scores are determined based on an intent model, and wherein the plurality of scores are determined based on weights assigned to each of the respective features in the intent model;
identifying, based on the scores, a particular intent, of the plurality of intents, as corresponding to the utterance;
generating a response based on the identified intent; and
transmitting the response as output.
6. The non-transitory computer readable medium of claim 5 , wherein the intent model is a multi-key map of features to words.
7. The non-transitory computer readable medium of claim 5 , wherein generating the semantic frame comprises:
generating a plurality of grammatical relations by labeling words to define relationships; and
aggregating the plurality of grammatical relations.
8. The non-transitory computer readable medium of claim 5 , wherein the features correspond to parts of speech and the relations correspond to parts of speech defining relationships between words.
9. A method for training one or more intent models comprising:
receiving, by a computer system, a set of training utterances, each training utterance being labeled with a particular intent; and
for each training utterance:
extracting, by the computer system from the training utterance, one or more features, wherein at least one of the features is a semantically relevant grammatical relation, wherein each feature has a value represented by one or more words included in the training utterance, and wherein extracting the one or more features comprises parsing the training utterance and generating a semantic frame to identify relations between the one or more words in the training utterance; and
storing, by the computer system, the one or more features along with their corresponding values in association with the intent with which the training utterance is labeled,
wherein the one or more features and their corresponding values are stored to one or more intent models useable for determining an intent for an utterance.
10. The method of claim 9 , wherein the intent models are useable in a chatbot system.
11. The method of claim 9 , wherein the features and the values are stored as a multi-key model for the corresponding intent.
12. The method of claim 9 , wherein a subset of the features are relations corresponding to parts of speech defining relationships between the values.
13. The method of claim 9 , wherein the features correspond to parts of speech.