| CPC G10L 15/222 (2013.01) [G06F 16/634 (2019.01); G10L 15/30 (2013.01); G10L 25/51 (2013.01); G10L 25/84 (2013.01)] | 20 Claims |
|
1. A method comprising:
receiving, by a device executing an assistant application, a first query from a user;
determining, using a machine learning model, that the first query is related to a particular entity in a first context, wherein the first query comprises a first request to be provided with an indication of a first characteristic of the particular entity and related to the first context;
causing the device to generate for output a first reply to the first query, wherein the first reply provides the indication of the first characteristic to the user;
detecting an interruption for the user based on determining that the interruption persists for at least a threshold period of time;
detecting an end of the interruption; and
in response to detecting the interruption and subsequently detecting the end of the interruption:
generating a predicted second query comprising a plurality of words, wherein the predicted second query comprises a second request to be provided with a second characteristic of the particular entity of the first query in a second context and the predicted second query is a prediction of input that would have been received from the user after the first reply had the interruption not occurred, wherein the second context is different from the first context and the first characteristic is different than the second characteristic, and wherein the generating of the predicted second query comprises:
accessing, at a database storing a plurality of historical queries, historical queries previously received from a plurality of other users different from the user;
determining, using the machine learning model, a particular historical query of the accessed historical queries having one or more attributes that relate to an attribute of the particular entity of the first query; and
generating the predicted second query by identifying the plurality of words to be included in the predicted second query based on a plurality of words included in the particular historical query accessed via the database and having been received from one of the plurality of other users different from the user; and
causing the device to generate for output a prompt to provide a second reply to the predicted second query, the second reply comprising an indication of the second characteristic.
|