Information processing system, an information processing method and a computer readable storage medium

• US 10,354,010 B2
• Filed: 04/24/2015
• Issued: 07/16/2019
• Est. Priority Date: 04/24/2015
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Andrade Silva, Daniel Georg, Tamura, Akihiro, Tsuchida, Masaaki
• Primary Examiner(s): Desir, Pierre Louis
• Assistant Examiner(s): Kim, Jonathan C
• Application Number:
  US15/567,630
• Publication Number:
  US 20180137100A1
• Time in Patent Office: 1,544 Days
• Field of Search: None
• CPC Class Codes: G06F 16/3347 using vector based model
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

An information processing system to increase weights of words that are related to a text, but that do not explicitly occur in the text, in a weight vector representing the text, is provided. An adjusting system (100) includes a distance storing unit (110) and an adjusting unit (120). The distance storing unit (110) stores distances between any two terms of a plurality of terms. The distance between two terms becomes smaller as the two terms are semantically more similar. The adjusting unit (120) adjusts a weight of each term of the plurality of terms in a weight vector including weights of the plurality of terms and representing a text, on the basis of a distance between each term and other term in the weight vector and a weight of the other term.

Claims

1. An information processing system comprising:

a memory storing instructions; and

one or more processors configured to execute the instructions to;

store distances between any two terms of a plurality of terms, the wherein a distance of the stored distances becomes smaller as two terms are semantically more similar and if the two terms tend to occur in texts that belong to a same class;

adjust a weight of each term of the plurality of terms in a weight vector including weights of the plurality of terms and representing a text, on the basis of distances between each term and other terms in the weight vector and weights of the other terms; and

classify the text using the adjusted weight of each term of the plurality of terms in the weight vector,

wherein the weight of each term in the weight vector is adjusted by estimating a latent weight of each term of the plurality of terms in the weight vector by a MAP (maximum-a-posteriori) estimate with a posterior probability of the latent weight when observed weights of the plurality of terms in the weight vector are given, assuming the observed weights are generated from latent weights with Gaussian noise, and

wherein the posterior probability is approximated with a subset of the observed weights of the plurality of terms, wherein the subset corresponds to a subset of the plurality of terms in the weight vector selected based on the distances between each term and the other terms.

2. The information processing system according to claim 1, wherein

the one or more processors configured to further execute the instructions to:

calculate the distances between any two terms of the plurality of terms on the basis of a distance between feature vectors of the two terms, the feature vector including at least one of information about a local-word window context and information about a topical context.

3. The information processing system according to claim 2, wherein

the distances are corrected in such a way that the distance between two terms becomes smaller if the two terms tend to occur in texts that belong to the same class.

4. The information processing system according to claim 1, wherein

the weight of each term in the weight vector is adjusted by calculating a covariance matrix for the plurality of terms from the distances between each term and the other terms, and estimating the latent weight of each term of the plurality of terms from the weight vector on the basis of the MAP estimate using the calculated covariance matrix.

5. The information processing system according to claim 4, wherein

the latent weight of each term of the plurality of terms is estimated using the calculated covariance matrix restricted to the subset of the plurality of terms, wherein distances between each term and the subset of the plurality of terms are smaller than a distance between each term and a term not in the subset.

6. An information processing method comprising:

reading out distances between each term of a plurality of terms in a weight vector and other terms in the weight vector from a distance storage which stores distances between any two terms of the plurality of terms, wherein a distance of the stored distances becomes smaller as two terms are semantically more similar and if the two terms tend to occur in texts that belong to a same class, the weight vector including weights of the plurality of terms and representing a text;

adjusting a weight of each term of the plurality of terms in the weight vector on the basis of the distances between each term and other terms in the weight vector and weights of the other terms; and

classifying the text using the adjusted weight of each term of the plurality of terms in the weight vector,

wherein the weight of each term in the weight vector is adjusted by estimating a latent weight of each term of the plurality of terms in the weight vector by a MAP (maximum-a-posteriori) estimate with a posterior probability of the latent weight when observed weights of the plurality of terms in the weight vector are given, assuming the observed weights are generated from latent weights with Gaussian noise, and

wherein the posterior probability is approximated with a subset of the observed weights of the plurality of terms, wherein the subset corresponds to a subset of the plurality of terms in the weight vector selected based on the distances between each term and the other terms.

7. The information processing method according to claim 6, further comprising calculating the distances between any two terms of the plurality of terms on the basis of a distance between feature vectors of the two terms, the feature vector including at least one of information about a local-word window context and information about a topical context.

8. The information processing method according to claim 7, wherein,

the calculating corrects the distances in such a way that the distance between two terms becomes smaller if the two terms tend to occur in texts that belong to the same class.

9. The information processing method according to claim 6,

wherein the adjusting adjusts the weight of each term in the weight vector by calculating a covariance matrix for the plurality of terms from the distances between each term and the other terms, and estimating the latent weight of each term of the plurality of terms from the weight vector on the basis of the MAP estimate using the calculated covariance matrix.

10. A non-transitory computer readable storage medium recording thereon a program, causing a computer to perform a method comprising:

reading out distances between each term of a plurality of terms in a weight vector and other term terms in the weight vector from a distance storage which stores distances between any two terms of the plurality of terms, wherein a distance of the stored distances becomes smaller as two terms are semantically more similar and if the two terms tend to occur in texts that belong to a same class, the weight vector including weights of the plurality of terms and representing a text;

adjusting a weight of each term of the plurality of terms in the weight vector on the basis of the distances between each term and other terms in the weight vector and weights of the other terms; and

classifying the text using the adjusted weight of each term of the plurality of terms in the weight vector,

wherein the weight of each term in the weight vector is adjusted by estimating a latent weight of each term of the plurality of terms in the weight vector by a MAP (maximum-a-posteriori) estimate with a posterior probability of the latent weight when observed weights of the plurality of terms in the weight vector are given, assuming the observed weights are generated from latent weights with Gaussian noise, and

wherein the posterior probability is approximated with a subset of the observed weights of the plurality of terms, wherein the subset corresponds to a subset of the plurality of terms in the weight vector selected based on the distances between each term and the other terms.