Information processing device, information processing method, and recording medium

Patent metadata

Field	Value
Publication number	US-11544455-B2
Application number	US-201716623033-A
Country	US
Kind code	B2
Filing date	Jun 21, 2017
Priority date	Jun 21, 2017
Publication date	Jan 3, 2023
Grant date	Jan 3, 2023

Abstract

An information processing device according to the present invention includes: a memory; and a processor coupled to the memory. The processor performs operations. The operations includes: generating, based on language data, a predicate argument structure including a predicate and an argument being an object of the predicate; generating first data indicating co-occurrence of the predicate and the argument in the predicate argument structure; decomposing the first data into a plurality of pieces of second data including fewer elements than elements included in the first data, and generating, based on the second data, third data including potential co-occurrence of the predicate and the argument; selecting the predicate argument structure by using the first data and the third data, and calculating, by using the third data, a score for a pair of the predicate argument structures including the selected predicate argument structure; and selecting the pair, based on the score.

First claim

The invention claimed is:

1. An information processing device comprising:
a memory; and
at least one processor coupled to the memory,
the processor performing operations, the operations comprising:
generating, based on language data, a predicate argument structure including a predicate and an argument being an object of the predicate;
generating first data indicating co-occurrence of the predicate and the argument in the predicate argument structure;
the first data into a plurality of pieces of second data including fewer elements than elements included in the first data, and generating, based on the second data, third data including potential co-occurrence of the predicate and the argument;
selecting the predicate argument structure by using the first data and the third data, and calculating, by using the third data, a score for a pair of the predicate argument structures including the selected predicate argument structure; and
selecting the pair, based on the score,

wherein the operations further comprise:
calculating, as the first data, an original co-occurrence tensor including the predicate and the argument as modes;
decomposing, as the second data, the original co-occurrence tensor into factor tensors of lower rank than a rank of the original co-occurrence tensor, and generating, as the third data, a restore tensor being a product of the factor tensors;
generating, as the original co-occurrence tensor, a plurality of the original co-occurrence tensors including the predicate and any one of the arguments as modes;
decomposing the original co-occurrence tensor into the factor tensors in such a way that at least some of the factor tensors are an identical tensor in all the original co-occurrence tensors; and
calculating, based on the restore tensor, a probability of appearance for each of the predicate and the argument, calculating a relationship between a plurality of arguments by using the factor tensors, and calculating the score by using the probability of appearance and the relationship.

2. The information processing device according to claim 1 , wherein the operations further comprise
generating, as the original co-occurrence tensor, the original co-occurrence tensor including all the predicates and all the arguments as modes, and
calculating the score by using a probability of appearance for each of the predicate and the argument in the restore tensor.

3. An information processing method comprising:
generating, based on language data, a predicate argument structure including a predicate and an argument being an object of the predicate;
generating first data indicating co-occurrence of the predicate and the argument in the predicate argument structure;
decomposing the first data into a plurality of pieces of second data including fewer elements than elements included in the first data;
generating, based on the second data, third data including potential co-occurrence of the predicate and the argument;
selecting the predicate argument structure by using the first data and the third data;
calculating, by using the third data, a score for a pair of the predicate argument structures including the selected predicate argument structure; and
selecting the pair, based on the score,
wherein the method further comprises:
calculating, as the first data, an original co-occurrence tensor including the predicate and the argument as modes;
decomposing, as the second data, the original co-occurrence tensor into factor tensors of lower rank than a rank of the original co-occurrence tensor, and generating, as the third data, a restore tensor being a product of the factor tensors;
generating, as the original co-occurrence tensor, a plurality of the original co-occurrence tensors including the predicate and any one of the arguments as modes;
decomposing the original co-occurrence tensor into the factor tensors in such a way that at least some of the factor tensors are an identical tensor in all the original co-occurrence tensors; and
calculating, based on the restore tensor, a probability of appearance for each of the predicate and the argument, calculating a relationship between a plurality of arguments by using the factor tensors, and calculating the score by using the probability of appearance and the relationship.

4. A non-transitory computer-readable recording medium embodying a program, the program causing a computer to perform a method, the method comprising:
generating, based on language data, a predicate argument structure including a predicate and an argument being an object of the predicate;
generating first data indicating co-occurrence of the predicate and the argument in the predicate argument structure;
decomposing the first data into a plurality of pieces of second data including fewer elements than elements included in the first data;
generating, based on the second data, third data including potential co-occurrence of the predicate and the argument;
selecting the predicate argument structure by using the first data and the third data;
calculating, by using the third data, a score for a pair of the predicate argument structures including the selected predicate argument structure; and
selecting the pair, based on the score,
wherein the method further comprises:
calculating, as the first data, an original co-occurrence tensor including the predicate and the argument as modes;
decomposing, as the second data, the original co-occurrence tensor into factor tensors of lower rank than a rank of the original co-occurrence tensor, and generating, as the third data, a restore tensor being a product of the factor tensors;
generating, as the original co-occurrence tensor, a plurality of the original co-occurrence tensors including the predicate and any one of the arguments as modes;
decomposing the original co-occurrence tensor into the factor tensors in such a way that at least some of the factor tensors are an identical tensor in all the original co-occurrence tensors; and
calculating, based on the restore tensor, a probability of appearance for each of the predicate and the argument, calculating a relationship between a plurality of arguments by using the factor tensors, and calculating the score by using the probability of appearance and the relationship.

Assignees

• Nec Corp

Inventors

• HIGASHIYAMA Shohei
• Okajima Yuzuru
• Sadamasa Kunihiko

Classifications

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06F40/289

  Phrasal analysis, e.g. finite state techniques or chunking · CPC title

• G06N5/042

  Backward inferencing · CPC title

• G06F40/20 Primary

  Natural language analysis (semantic analysis of natural language G06F40/30) · CPC title

• G06F40/253

  Grammatical analysis; Style critique · CPC title