Determining query complexity in video question answering
A method for determining a complexity of a natural language query includes converting a first natural language query into executable program code, the first natural language query being a query for a first video to be answered by one or more video question answering (VideoQA) models. The method also includes generating, via a complexity model, an abstract syntax tree (AST) based on the executable program code. The method further includes determining, via the complexity model, a complexity of the first natural language query based on quantity of subtrees, from a group of subtrees, that are present in the AST.
1 . A method for determining a complexity of a natural language query, comprising:
converting a first natural language query into executable program code, the first natural language query being a query for a first video to be answered by one or more video question answering (VideoQA) models;
generating, via a complexity model, an abstract syntax tree (AST) based on the executable program code;
determining, via the complexity model, a complexity of the first natural language query based on a quantity of subtrees, from a group of subtrees, that are present in the AST;
generating, via a large language model, a group of queries associated with a second video based on a prompt and a natural language summary of the second video; and
selecting, via the complexity model, one or more queries of the group of queries, a respective complexity of each of the one or more queries being greater than or equal to a complexity threshold.
2 . The method of claim 1 , wherein the group of subtrees are subtrees that are common among a group of natural language queries.
3 . The method of claim 1 , wherein each subtree in the group of subtrees decreases a probability of the one or more VideoQA models correctly answering the first natural language query.
4 . The method of claim 1 , wherein:
determining the complexity further comprises encoding, via one-hot encoding, the AST into a vector based on the quantity of subtrees, from a group of subtrees, that are present in the AST;
the complexity model determines the complexity based on the vector.
5 . The method of claim 1 , further comprising generating the natural language summary via an image captioning model.
6 . The method of claim 1 , wherein the complexity is based on a likelihood of each of the one or more VideoQA models failing to answer the query.
7 . An apparatus for determining a complexity of a natural language query, comprising:
one or more processors; and
one or more memories coupled with the one or more processors and storing processor-executable code that, when executed by the one or more processors, is configured to cause the apparatus to:
convert a first natural language query into executable program code, the first natural language query being a query for a first video to be answered by one or more video question answering (VideoQA) models;
generate, via a complexity model, an abstract syntax tree (AST) based on the executable program code; and
determine, via the complexity model, a complexity of the first natural language query based on a quantity of subtrees, from a group of subtrees, that are present in the AST;
generate, via a large language model, a group of queries associated with a second video based on a prompt and a natural language summary of the second video; and
select, via the complexity model, one or more queries of the group of queries, a respective complexity of each of the one or more queries being greater than or equal to a complexity threshold.
8 . The apparatus of claim 7 , wherein the group of subtrees are subtrees that are common among a group of natural language queries.
9 . The apparatus of claim 7 , wherein each subtree in the group of subtrees decreases a probability of the one or more VideoQA models correctly answering the first natural language query.
10 . The apparatus of claim 7 , wherein:
execution of the processor-executable code to determine the complexity further causes the apparatus to encode, via one-hot encoding, the AST into a vector based on the quantity of subtrees, from a group of subtrees, that are present in the AST;
the complexity model determines the complexity based on the vector.
11 . The apparatus of claim 7 , wherein execution of the processor-executable code further causes the apparatus to generate the natural language summary via an image captioning model.
12 . The apparatus of claim 7 , wherein the complexity is based on a likelihood of each of the one or more VideoQA models failing to answer the query.
13 . A non-transitory computer-readable medium having program code recorded thereon for determining a complexity of a natural language query, the program code executed by one or more processors and comprising:
program code to convert a first natural language query into executable program code, the first natural language query being a query for a first video to be answered by one or more video question answering (VideoQA) models;
program code to generate, via a complexity model, an abstract syntax tree (AST) based on the executable program code; and
program code to determine, via the complexity model, a complexity of the first natural language query based on a quantity of subtrees, from a group of subtrees, that are present in the AST;
program code to generate, via a large language model, a group of queries associated with a second video based on a prompt and a natural language summary of the second video; and
program code to select, via the complexity model, one or more queries of the group of queries, a respective complexity of each of the one or more queries being greater than or equal to a complexity threshold.
14 . The non-transitory computer-readable medium of claim 13 , wherein the group of subtrees are subtrees that are common among a group of natural language queries.
15 . The non-transitory computer-readable medium of claim 13 , wherein each subtree in the group of subtrees decreases a probability of the one or more VideoQA models correctly answering the first natural language query.
16 . The non-transitory computer-readable medium of claim 13 , wherein:
the program code to determine the complexity further comprises program code to encode, via one-hot encoding, the AST into a vector based on the quantity of subtrees, from a group of subtrees, that are present in the AST;
the complexity model determines the complexity based on the vector.
17 . The non-transitory computer-readable medium of claim 13 , wherein the program code further comprises program code to generate the natural language summary via an image captioning model.