Method and apparatus for classifying data

Patent metadata

Field	Value
Publication number	US-12566820-B2
Application number	US-201817296638-A
Country	US
Kind code	B2
Filing date	Dec 3, 2018
Priority date	Dec 3, 2018
Publication date	Mar 3, 2026
Grant date	Mar 3, 2026

Abstract

A method and an apparatus are for classifying data. In an embodiment, the method includes: classifying at least two pieces of data, to obtain at least two data clusters; determining a bias degree of classification; re-classifying the at least two pieces of data by merging any several of the at least two data clusters; determining a bias degree of re-classification; and determining, by comparing the bias degree of first classification and the bias degree of re-classification, which classification is more accurate. By way of the method and apparatus, a related data cluster can be found from multiple data clusters for better data analysis.

First claim

What is claimed is:

1 . A method for classifying data, the method comprising:
first classifying at least two pieces of data to obtain at least two first data clusters; determining a first bias degree of the first classifying based on a plurality of first difference degrees, each of the plurality of first difference degrees being determined based on a difference between first output data and a respective first data cluster among the at least two first data clusters, the first output data being obtained by inputting the respective first data cluster into a corresponding first neural network model, and the corresponding first neural network model being among a plurality of first neural network models; re-classifying the at least two pieces of data by merging at least two of the at least two first data clusters to obtain a second data cluster; determining a second bias degree of the re-classifying based on a plurality of second difference degrees, each of the plurality of second difference degrees being determined based on a difference between second output data and a respective subset among subsets of data in the second data cluster, and the second output data being obtained by inputting the respective subset into a second neural network model; and determining which of the first classifying and the re-classifying is more accurate based on the first bias degree and the second bias degree.

2 . The method of claim 1 , wherein
each respective first neural network model among the plurality of first neural network models is obtained through training by using data in a corresponding first data cluster among the at least two first data clusters; an input layer and an output layer of each among the plurality of first neural network models have a same quantity of nodes; the second neural network model is obtained through training by using all data included in the second data clusters; an input layer and an output layer of the second neural network model have a same quantity of nodes; and the determining which of the first classifying and the re-classifying is more accurate includes
determining related data clusters among the at least two first data clusters, and
marking the related data clusters as related in response to the determining the related data clusters.

3 . The method of claim 2 , wherein the determining the related data clusters
is based on the second bias degree being less than the first bias degree.

4 . The method of claim 2 , further comprising:
merging data clusters marked as related among the at least two first data clusters into a third data cluster.

5 . The method of claim 2 , wherein
each among the plurality of first neural network models and the second neural network model are neural network models of a same type.

6 . The method of claim 1 , wherein
each respective first neural network model among the plurality of first neural network models is trained as an auto encoder for a corresponding first data cluster among the at least two first data clusters; and the second neural network model is trained as an auto encoder for the second data cluster.

7 . The method of claim 1 , wherein
the first output data has a same number of attributes as each respective first data cluster among the at least two first data clusters; and the second output data has a same number of attributes as each respective subset among the subsets of data in the second data cluster.

8 . The method of claim 1 , wherein
an input layer and an output layer of each among the plurality of first neural network models have a same quantity of nodes; and an input layer and an output layer of the second neural network model have a same quantity of nodes.

9 . A computation device, comprising:
at least one memory storing an executable instruction; and at least one processor configured to execute the executable instruction to cause the computation device to
first classifying at least two pieces of data, to obtain at least two first data clusters,
determine a first bias degree of the first classification based on a plurality of first difference degrees, each of the plurality of first difference degrees being determined based on a difference between first output data and a respective first data cluster among the at least two first data clusters, the first output data being obtained by inputting the respective first data cluster into a corresponding first neural network model, and the corresponding first neural network model being among a plurality of first neural network models,
re-classify the at least two pieces of data by merging at least two of the at least two first data clusters to obtain a second data cluster,
determine a second bias degree of the re-classification based on a plurality of second difference degrees, each of the plurality of second difference degrees being determined based on a difference between second output data and a respective subset among subsets of data in the second data cluster, and the second output data being obtained by inputting the respective subset into a second neural network model, and
determine which of the first classification and the re-classification is more accurate based on the first bias degree and the second bias degree.

10 . The computation device of claim 9 , wherein
each respective first neural network model among the plurality of first neural network models is obtained through training by using data in a corresponding first data structure among the at least two first data clusters; an input layer and an output layer of each among the plurality of first neural network models have a same quantity of nodes; the second neural network model is obtained through training by using all data included in the second data clusters; an input layer and an output layer of the second neural network model have a same quantity of nodes; and the at least one processor is configured to execute the executable instruction to cause the computation device to determine which of the first classification and the re-classification is more accurate including
determining related data clusters among the at least two first data clusters, and
marking the related data clusters as related in response to the determining the related data clusters.

11 . The computation device of claim 10 , wherein the determining the related data clusters is based on;
the second bias degree being less than the first bias degree.

12 . The computation device of claim 10 , wherein the at least one processor is configured to execute the executable instruction to cause the computation device to merge data clusters marked as related among the at least two first data clusters into a third data cluster.

13 . The computation device of claim 10 , wherein each among the plurality of first neural network models and the second neural network model are neural network models of a same type.

14 . The computation device of claim 9 , wherein
each respective first neural network model among the plurality of first neural network models is trained as an auto encoder for a corresponding first data cluster among the at least two first data clusters; and the second neural network model is trained as an auto encoder for the second data cluster.

15 . The computation device of claim 9 , wherein
the first output data has a same number of attributes as each respective first data cluster among the at least two first data clusters; and the second output data has a same number of attributes as each respective subset among the subsets of data in the second data cluster.

16 . A non

Assignees

• Siemens Mobility GmbH

Inventors

• Li Cong Chao
• Loh Chang Wei
• Wu chao hua
• Liu xiao nan
• Hui Hao Tian
• Lu Lei

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06F18/25

  Fusion techniques · CPC title

• G06F18/23

  Clustering techniques · CPC title

• G06N3/08

  Learning methods · CPC title

• G06V10/751

  Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching · CPC title