Methods and systems for determining characteristics of a dialog between a computer and a user
A computer-implemented method is disclosed for determining one or more characteristics of a dialog between a computer system and user. The method may comprise receiving a system utterance comprising one or more tokens defining one or more words generated by the computer system; receiving a user utterance comprising one or more tokens defining one or more words uttered by a user in response to the system utterance, the system utterance and the user utterance forming a dialog context; receiving one or more utterance candidates comprising one or more tokens; for each utterance candidate, generating an input sequence combining the one or more tokens of each of the system utterance, the user utterance, and the utterance candidate; and for each utterance candidate, evaluating the generated input sequence with a model to determine a probability that the utterance candidate is relevant to the dialog context.
1. A method comprising:
receiving, by a processing device, a system utterance and a user utterance;
generating, by the processing device, a dialog sequence by encoding the system utterance and the user utterance into a series of vectors;
identifying, by the processing device, a candidate topic represented as a candidate vector associated with a characteristic of the dialog sequence;
determining, by the processing device using a machine learning model, a relevance score indicating a probability that the candidate topic is relevant to the dialog sequence by comparing the candidate vector to the series of vectors; and
generating, by the processing device, data indicating a dialog context including the candidate topic based on the relevance score.
2. The method of claim 1 , further comprising replacing the candidate topic with a different candidate topic based on the relevance score.
3. The method of claim 1 , further comprising determining whether the relevance score meets a threshold of relevance to the dialog sequence.
4. The method of claim 3 , further comprising updating the dialog sequence to include the candidate topic if the relevance score meets the threshold of relevance to the dialog sequence.
5. The method of claim 1 , further comprising generating a response to the user utterance based on the candidate topic and the dialog sequence.
6. The method of claim 1 , wherein the characteristic of the dialog sequence is a state of a dialog between a system and a user.
7. The method of claim 1 , wherein each vector in the series of vectors is assigned to a word in the dialog sequence.
8. The method of claim 1 , further comprising selecting the candidate topic from a plurality of potential topics.
9. The method of claim 1 , wherein the probability that the candidate topic is relevant to the dialog sequence is calculated using a Bidirectional Encoder Representations from Transformers (BERT) model.
10. A system comprising:
a memory component; and
a processing device to perform operations comprising:
receiving a user utterance including user utterance tokens defining words generated by a user;
forming a dialog token sequence representing a dialog sequence by combining the user utterance tokens with system utterance tokens defining words of a system utterance generated by a computer system in response to the user utterance;
identifying a candidate topic represented as candidate topic tokens associated with a characteristic of the dialog sequence;
generating an input sequence by combining the dialog token sequence with the candidate topic tokens;
determining, using a machine learning model, a relevance score indicating a probability that the candidate topic is relevant to the dialog sequence by comparing the candidate topic tokens to the dialog token sequence; and
generating data indicating a dialog context including the candidate topic based on the relevance score.
11. The system of claim 10 , wherein each token in the dialog token sequence is assigned to a word in the dialog sequence.
12. The system of claim 10 , wherein the user utterance is a question and the system utterance is an answer to the question.
13. The system of claim 10 , wherein the user utterance tokens and the system utterance tokens are separated with a separator token representing a start of an utterance.
14. The system of claim 10 , further comprising determining whether the relevance score meets a threshold of relevance to the dialog sequence.
15. The system of claim 14 , further comprising updating the dialog sequence to include the candidate topic if the relevance score meets the threshold of relevance to the dialog sequence.
16. A non-transitory computer-readable storage medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:
receiving a system utterance and a user utterance;
generating a dialog sequence by encoding the system utterance and the user utterance into a series of vectors;
identifying a candidate topic represented as a candidate vector associated with a characteristic of the dialog sequence;
determining, using a machine learning model, a relevance score indicating a probability that the candidate topic is relevant to the dialog sequence by comparing the candidate vector to the series of vectors; and
generating data indicating a dialog context including the candidate topic based on the relevance score.
17. The non-transitory computer-readable storage medium of claim 16 , further comprising replacing the candidate topic with a different candidate topic based on the relevance score.
18. The non-transitory computer-readable storage medium of claim 16 , wherein the characteristic of the dialog sequence is a state of a dialog between a system and a user.
19. The non-transitory computer-readable storage medium of claim 16 , further comprising selecting the candidate topic from a plurality of potential topics.
20. The method of claim 1 , wherein the data indicating the dialog context includes a description of the dialog sequence.