Device-agnostic framework to measure reliability during user interactions
Implementations relate to retrieving and processing metadata associated with a user query directed to an interactive assistant application. Implementations further relate to classifying the user query using labels assigned to invocation stage, input-receiving stage, response-receiving stage, and/or response-rendering stage of the user query that are determined based on processing the metadata associated with the user query. Whether the user query can be applied to evaluate a performance (e.g., surface reliability) of the interactive assistant application can be determined based on the classification of the user query.
1 . A method implemented using one or more processors, the method comprising:
identifying metadata associated with a user query that is directed to an interactive assistant application;
processing the metadata associated with the user query to determine a classification category to which the user query belongs,
wherein processing the metadata associated with the user query includes:
determining a respective label, from a plurality of predefined labels, for each of one or more stages of the interactive assistant application handling the user query, the one or more stages belonging to a plurality of predefined stages of the interactive assistant application, and
determining the classification category based on:
the respective label for each of the one or more stages of the interactive assistant application handling the user query, and
on a total period of time from invocation of the interactive assistant application to complete rendering of a response responsive to the user query satisfying a total threshold; and
generating, based on the classification category determined for the user query, an alert to one or more entities tasked with overseeing one of the plurality of predefined stages of the interactive assistant application handling the user query in response to the one of the plurality of predefined stages being determined to have a label indicating unsatisfactory handling of the user query during the one of the plurality of predefined stages.
2 . The method of claim 1 , wherein the plurality of predefined stages of the interactive assistant application include an invocation stage, an input-receiving stage, a response-generation stage, and a response-rendering stage, of the interactive assistant application handling the user query.
3 . The method of claim 1 , wherein determining the classification category based on the respective label for each of the one or more stages of the interactive assistant application handling the user query comprises: determining a first classification category for the user query based at least on each label for each of the plurality of predefined stages being the first label, the first classification category indicating a satisfactory overall surface performance of the interactive assistant application handling the user query.
4 . The method of claim 1 , wherein determining the classification category based on the respective label for each of the one or more stages of the interactive assistant application handling the user query comprises: determining a second classification category for the user query based on at least one second label being determined for at least one of the plurality of predefined stages, the second classification category indicating an unsatisfactory overall surface performance of the interactive assistant application handling the user query.
5 . The method of claim 1 , wherein the plurality of predefined labels further include a third label indicating that the respective stage of the interactive assistant application handling the user query renders the user query ineligible, and/or a fourth label indicating that the metadata associated with the user query misses information to classify the respective stage.
6 . The method of claim 5 , wherein determining the classification category based on the respective label for each of the one or more stages of the interactive assistant application handling the user query comprises:
determining an ineligible classification category for the user query based on at least one third label being determined for at least one of the plurality of predefined stages, the ineligible classification category indicating the user query is ineligible to evaluate a surface reliability of the interactive assistant application.
7 . The method of claim 5 , wherein determining the classification category based on the respective label for each of the one or more stages of the interactive assistant application handling the user query comprises:
determining a fourth classification category for the user query based on at least one fourth label being determined for at least one of the plurality of predefined stages, the fourth classification category indicating that information is missing from the metadata associated with the user query to determining the classification category for the user query.
8 . The method of claim 1 , further comprising:
receiving a user request to evaluate the user query, wherein identifying the metadata associated with the user query is in response to receiving the user request to evaluate the user query.
9 . The method of claim 1 , further comprising:
determining whether to apply the user query to evaluate a surface reliability of the interactive assistant application based on the classification category that the user query belongs to.