Systems and methods for training a virtual assistant
A virtual assistant server determines a subset of test data corresponding to changes between a first version of training data of a virtual assistant and a second version of the training data of the virtual assistant. Subsequently, the virtual assistant server creates a test suite with the subset of test data and runs the test suite on a second language model of the virtual assistant created using the second version of the training data. Based on the running the test suite, the virtual assistant server generates one or more executable corrective actions to be implemented at the user device and provides the one or more executable corrective actions to the user device to implement to train the virtual assistant.
1 . A method for assisting a user accessing a user device to train a virtual assistant, the method comprising:
tracking, by a virtual assistant server, changes between a first version of training data and a second version of the training data of the virtual assistant, wherein at least one of the changes comprises addition, deletion, or modification of an utterance, a pattern, or a rule associated with one or more intents;
determining, by the virtual assistant server, a subset of test data corresponding to the tracked changes between the first version of the training data and the second version of the training data;
creating, by the virtual assistant server, a test suite with the subset of test data;
running, by the virtual assistant server, the test suite on a second language model of the virtual assistant created using the second version of the training data;
generating, by the virtual assistant server, based on the running the test suite, one or more executable corrective actions to be implemented at the user device;
providing, by the virtual assistant server, the one or more executable corrective actions to a display of the user device;
receiving, by the virtual assistant server, a selection of at least one of the one or more executable corrective actions;
editing, by the virtual assistant server, the second version of the training data based on the received selection; and
training, by the virtual assistant server, the virtual assistant with the edited second version of the training data.
2 . The method of claim 1 , wherein the virtual assistant server creates a first language model using the first version of the training data.
3 . The method of claim 1 , wherein the subset of test data comprises utterances in the test data which are structurally or semantically similar to the changes made to the first version of the training data.
4 . The method of claim 1 , wherein the test suite is created by using one or more expected intents or one or more expected entities of utterances in the subset of test data.
5 . The method of claim 1 , wherein the running the test suite determines test results comprising, for each test utterance of the subset of test data: one or more expected intents, one or more expected entities, one or more classified intents, or one or more classified entities.
6 . The method of claim 1 , wherein the one or more executable corrective actions are generated in natural language.
7 . A virtual assistant server comprising:
a processor; and
a memory coupled to the processor which is configured to be capable of executing programmed instructions stored in the memory to:
track changes between a first version of training data and a second version of the training data of a virtual assistant, wherein at least one of the changes comprises addition, deletion, or modification of an utterance, a pattern, or a rule associated with one or more intents;
determine a subset of test data corresponding to the tracked changes between the first version of the training data and the second version of the training data;
create a test suite with the subset of test data;
run the test suite on a second language model of the virtual assistant created using the second version of the training data;
generate based on the run the test suite, one or more executable corrective actions to be implemented at a user device;
provide the one or more executable corrective actions to a display of the user device;
receive a selection of at least one of the one or more executable corrective actions from the user device;
edit the second version of the training data based on the received selection; and
train the virtual assistant with the edited second version of the training data.
8 . The virtual assistant server of claim 7 , wherein the virtual assistant server creates a first language model using the first version of the training data.
9 . The virtual assistant server of claim 7 , wherein the subset of test data comprises utterances in the test data which are structurally or semantically similar to the changes made to the first version of the training data.
10 . The virtual assistant server of claim 7 , wherein the test suite is created by using one or more expected intents or one or more expected entities of utterances in the subset of test data.
11 . The virtual assistant server of claim 7 , wherein the running the test suite determines test results comprising, for each test utterance of the subset of test data: one or more expected intents, one or more expected entities, one or more classified intents, or one or more classified entities.
12 . The virtual assistant server of claim 7 , wherein the one or more executable corrective actions are generated in natural language.
13 . A non-transitory computer-readable medium having stored thereon instructions which when executed by a processor, causes the processor to:
track changes between a first version of training data and a second version of the training data of a virtual assistant, wherein at least one of the changes comprises addition, deletion, or modification of an utterance, a pattern, or a rule associated with one or more intents;
determine a subset of test data corresponding to the tracked changes between the first version of the training data and the second version of the training data;
create a test suite with the subset of test data;
run the test suite on a second language model of the virtual assistant created using the second version of the training data;
generate based on the run the test suite, one or more executable corrective actions to be implemented at a user device;
provide the one or more executable corrective actions to a display of the user device;
receive a selection of at least one of the one or more executable corrective actions from the user device;
edit the second version of the training data based on the received selection; and
train the virtual assistant with the edited second version of the training data.
14 . The non-transitory computer-readable medium of claim 13 , wherein the virtual assistant server creates a first language model using the first version of the training data.
15 . The non-transitory computer-readable medium of claim 13 , wherein the subset of test data comprises utterances in the test data which are structurally or semantically similar to the changes made to the first version of the training data.
16 . The non-transitory computer-readable medium of claim 13 , wherein the test suite is created by using one or more expected intents or one or more expected entities of utterances in the subset of test data.
17 . The non-transitory computer-readable medium of claim 13 , wherein the running the test suite determines test results comprising, for each test utterance of the subset of test data: one or more expected intents, one or more expected entities, one or more classified intents, or one or more classified entities.
18 . The non-transitory computer-readable medium of claim 13 , wherein the one or more executable corrective actions are generated in natural language.