Query correction based on reattempts learning
Systems and methods are described to access a set of reattempt query pairs, where each respective pair comprises an initial query and a reattempt of the initial query, and is associated with an indication of whether a reply generated for output based on the respective query pair was acceptable. In response to determining that a second query received after a first query constitutes a reattempt of the first query, a query pair in the set of reattempt query pairs may be identified that matches at least one of the first query and the second query, and is associated with an indication that a reply generated for output based on the query pair was acceptable. A search may be performed based on the identified query pair in the set of reattempt query pairs, and a reply may be generated for output based on the performed search.
1 . A method comprising:
receiving a first input query and a second input query;
accessing a set of reattempt query pairs stored at a data structure, wherein each respective pair in the set:
comprises an initial query and a reattempt query; and
is associated with an indication of whether a reply generated for output based on the respective query pair was acceptable;
inputting, into a trained reattempt machine learning model, data indicative of the first input query and the second input query;
receiving, from the trained reattempt machine learning model, an output probability indicating the second input query constitutes a reattempt query of the first input query;
based at least in part on the output probability indicating that the second input query constitutes the reattempt query of the first input query, accessing the data structure to select a query pair from the set of reattempt query pairs based at least in part on receiving determinations that:
(a) the query pair is related to at least one of the first input query or the second input query, and
(b) the query pair is associated with an indication that a reply generated for output based on the query pair was acceptable;
performing a search based on the selected query pair in the set of the reattempt query pairs; and
generating for output a reply based on the performed search.
2 . The method of claim 1 , wherein performing the search comprises:
modifying the second input query based on at least one query of the selected query pair; and
performing the search using the modified query.
3 . The method of claim 1 , wherein selecting the query pair from the set of reattempt query pairs comprises:
comparing the first input query and the second input query to each query of the query pair; and
receiving a determination that at least one of the first input query or the second input query is an exact match of a query of the query pair.
4 . The method of claim 1 , wherein:
the reply generated for output based on the performed search comprises a selectable identifier for a content item;
the method further comprises, based on receiving a user selection of the selectable identifier, causing the content item to be generated for consumption by a user;
wherein the indication that the reply generated for output based on the performed search was acceptable is generated based on the content item being consumed by the user for a period of time greater than a threshold period of time.
5 . The method of claim 1 , wherein the set of reattempt query pairs is received from a plurality of users within a predetermined period of time from a current time, and the plurality of users belong to at least one of particular demographics or a particular geographic area.
6 . The method of claim 5 , wherein the set of reattempt query pairs is discarded after the predetermined period of time elapses.
7 . The method of claim 1 , wherein the trained reattempt machine learning model is obtained by training an untrained machine learning model using data associated with a plurality of reattempt query pairs.
8 . The method of claim 7 , wherein the data comprises at least one of phonetic word match information between the plurality of reattempt query pairs or parts of speech of terms in the plurality of reattempt query pairs.
9 . The method of claim 1 , wherein:
the set of reattempt query pairs stored at the data structure comprises a first query pair and a second query pair;
the selecting the query pair from the set of reattempt query pairs is based on receiving determinations of relationships between the first input query, the second input query, and queries from each of the first query pair and the second query pair comprising:
receiving a determination that the first input query matches a query from the first query pair;
receiving a determination that the second input query does not match the queries of the first query pair;
receiving a determination that neither of the first input query nor the second input query match the queries of the second query pair;
receiving a determination that a query from the first query pair that does not match the first input query nor the second input query matches a query from the second query pair; and
receiving a determination that the second query pair is associated with the indication that a reply generated for output based on the second query pair was acceptable; and
based on receiving the determinations of the relationships, the performing the search based on the identified query pair comprises performing the search using the second query pair.
10 . The method of claim 9 , wherein:
the first query pair includes a first initial query and a first reattempt query;
the second query pair includes a second initial query and a second reattempt query; and
the receiving determinations of identifying the relationships further comprises:
receiving a determination that the first input query matches the first reattempt query, and that the first input query does not match the first initial query, the second initial query, nor the second reattempt query;
receiving a determination that the second input query does not match the first initial query, the first reattempt query, the second initial query, nor the second reattempt query; and
receiving a determination that the first reattempt query of the first query pair matches the second initial query of the second query pair; and
based on receiving the determinations of the relationships, performing the search based on the identified query pair comprises performing the search using the second initial query of the second query pair.
11 . A system comprising:
a storage device; and
control circuitry configured to:
receive a first input query and a second input query;
access a set of reattempt query pairs stored at the storage device, wherein each respective pair in the set:
comprises an initial query and a reattempt query; and
is associated with an indication of whether a reply generated for output based on the respective query pair was acceptable;
input, into a trained reattempt machine learning model, data indicative of the first input query and the second input query;
receive, from the trained reattempt machine learning model, an output probability indicating the second input query constitutes a reattempt query of the first input query;
based at least in part on the output probability indicating that the second input query constitutes the reattempt query of the first input query, accessing the storage device to select a query pair from the set of reattempt query pairs, wherein:
(a) the query pair is related to at least one of the first input query or the second input query, and
(b) the query pair is associated with an indication that a reply generated for output based on the query pair was acceptable;
perform a search based on the query pair in the set of the reattempt query pairs; and
generate for output a reply based on the performed search.
12 . The system of claim 11 , wherein the control circuitry is configured to perform the search by:
modifying the second input query based on at least one query of the selected query pair; and
performing the search using the modified query.
13 . The system of claim 11 , wherein the control circuitry is configured to select the query pair from the set of reattempt query pairs by:
comparing the first input query and the second input query to each query of the query pair; and
receiving a determination that at least one of the first input query or the second input query is an exact match of a query of the query pair.
14 . The system of claim 11 , wherein:
the reply generated for output based on the performed search comprises a selectable identifier for a content item;
the control circuitry is configured to:
based on receiving a user selection of the selectable identifier, cause the content item to be generated for consumption by a user; and
generate the indication that the reply generated for output based on the performed search was acceptable based on the content item being consumed by the user for a period of time greater than a threshold period of time.
15 . The system of claim 11 , wherein the set of reattempt query pairs is received from a plurality of users within a predetermined period of time from a current time, and the plurality of users belong to at least one of particular demographics or a particular geographic area.
16 . The system of claim 15 , wherein the set of reattempt query pairs is discarded after the predetermined period of time elapses.
17 . The system of claim 11 , wherein the trained reattempt machine learning model is obtained by training an untrained machine learning model using data associated with a plurality of reattempt query pairs.
18 . The system of claim 17 , wherein the data comprises at least one of phonetic word match information between the plurality of reattempt query pairs or parts of speech of terms in the plurality of reattempt query pairs.
19 . The system of claim 11 , wherein the set of reattempt query pairs stored at the storage device comprises a first query pair and a second query pair;
the control circuitry is configured to select the query pair from the set of reattempt query pairs based on receiving determinations of relationships between the first input query, the second input query, and queries from each of the first query pair and the second query pair comprising:
receiving a determination that the first input query matches a query from the first query pair;
receiving a determination that the second input query does not match the queries of the first query pair;
receiving a determination that neither of the first input query nor the second input query match the queries of the second query pair;
receiving a determination that a query from the first query pair that does not match the first input query nor the second input query matches a query from the second query pair; and
receiving a determination that the second query pair is associated with the indication that a reply generated for output based on the second query pair was acceptable; and
based on receiving the determinations of the relationships, the performing the search based on the identified query pair comprises performing the search using the second query pair.
20 . The system of claim 19 , wherein:
the first query pair includes a first initial query and a first reattempt query;
the second query pair includes a second initial query and a second reattempt query; and
the control circuitry is configured to receive the determinations of the relationships further by:
receiving a determination that the first input query matches the first reattempt query, and that the first input query does not match the first initial query, the second initial query, nor the second reattempt query;
receiving a determination that the second input query does not match the first initial query, the first reattempt query, the second initial query, nor the second reattempt query; and
receiving a determination that the first reattempt query of the first query pair matches the second initial query of the second query pair; and
based on receiving the determinations of the relationships, performing the search based on the identified query pair comprises performing the search using the second initial query of the second query pair.