Systems and methods for determining structured proceeding outcomes

Patent metadata

Field	Value
Publication number	US-12008668-B2
Application number	US-202318102760-A
Country	US
Kind code	B2
Filing date	Jan 29, 2023
Priority date	Jun 19, 2018
Publication date	Jun 11, 2024
Grant date	Jun 11, 2024

Abstract

The present disclosure relates to systems and methods for analyzing and extracting data related to a structured proceeding, and for identifying, based on the analysis, at least one outcome associated with the structured proceeding. Embodiments provide for receiving data associated with a structured proceeding involving at least one party, the data including at least one docket entry, and analyzing, by an outcome location detector, the data to identify one or more docket entries in the at least one docket entry that are likely to include evidence of an outcome. Embodiments further include analyzing, by an outcome detector, the one or more docket entries determined to be likely to include evidence of an outcome to determine outcomes. The outcomes include at least one of a final outcome and at least one party outcome. The final outcome is associated with the structured proceeding overall, and the at least one party outcome is associated with a party of the at least one party that may have been terminated early.

First claim

The invention claimed is:

1. A method, performed by a deep learning classifier, for identifying an outcome of a structured proceeding, the method comprising:
receiving, at a first level of a neural network of the deep learning classifier, a first word token corresponding to a first word of a first entry of the structured proceeding;
receiving, at the first level of the neural network, a second word token corresponding to a first word of a second entry of the structured proceeding, wherein the first level of the neural network comprises a first plurality of bidirectional gated recurrent units (GRUs) and a second plurality of GRUs;
outputting, by the first plurality of GRUs, a first encoded entry corresponding to the first entry of the structured proceeding;
outputting, by the second plurality of GRUs, a second encoded entry corresponding to the second entry of the structured proceeding;
receiving, at a second level of the neural network of the deep learning classifier, the first encoded entry and the second encoded entry, wherein:
the second level of the neural network includes a third plurality of GRUs,
a first GRU of the third plurality of GRUs is coupled to the first plurality of GRUs and is configured to receive the first encoded entry, and
a second GRU of the third plurality of GRUs is coupled to the second plurality of GRUs and is configured to receive the second encoded entry; and

outputting, by the second level of the neural network, an encoded structured proceeding corresponding to the structured proceeding;
receiving, by a third level of the neural network of the deep learning classifier, the encoded structured proceeding; and
outputting, by the third level of the neural network, a probability score associated with an outcome corresponding to the encoded structured proceeding.

2. The method of claim 1 , wherein:
the first plurality of GRUs comprises at least a first bidirectional gated recurrent unit (GRU) and a second GRU, and
the second plurality of GRUs comprises at least a third GRU and a fourth GRU.

3. The method of claim 2 , wherein:
a first GRU of the first plurality of GRUs is configured to receive the first word token,
a second GRU of the second GRU plurality of GRUs is configured to receive the second word token,
a third GRU of the third plurality of GRUs is configured to receive the first encoded entry, and
a fourth GRU of the third plurality of GRUs is configured to receive the second encoded entry.

4. The method of claim 1 , wherein the third level of the neural network comprises a softmax activation layer.

5. The method of claim 1 , wherein:
receiving, at the first level of the neural network of the deep learning classifier, the first word token comprises receiving, at the first level of the neural network, a first plurality of word tokens, wherein the first word token is included in the first plurality of word tokens; and
receiving, at the first level of the neural network, the second word token comprises receiving, at the first level of the neural network, a second plurality of word tokens, wherein the second word token is included in the second plurality of word tokens.

6. The method of claim 5 , wherein the first plurality of word tokens are provided to the first plurality of GRUs, and wherein the second plurality of word tokens are provided to the second plurality of GRUs.

7. The method of claim 5 , wherein the second level of the neural network comprises a third plurality of GRUs, each GRU of which is configured to receive an encoded entry of the structured proceeding.

8. The method of claim 1 , wherein:
the structured proceeding comprises a docket,
the first entry comprises a first docket entry,
and
the second entry comprises a second docket entry.

9. The method of claim 8 , wherein:
the first word of the first entry corresponds to a first word of the first docket entry, and
the first word of the second entry corresponds to a first word of the second docket entry.

10. An apparatus configured to identify an outcome of a structured proceeding, the apparatus comprising:
a processor;
a memory coupled to the processor, wherein the memory comprises a deep learning classifier, and wherein the deep learning classifier is configured to:
receive, at a first level of a neural network of the deep learning classifier, a first word token corresponding to a first word of a first entry of the structured proceeding;
receive, at the first level of the neural network, a second word token corresponding to a first word of a second entry of the structured proceeding, wherein the first level of the neural network comprises a first plurality of bidirectional gated recurrent units (GRUs) and a second plurality of GRUs;
output, by the first plurality of GRUs, a first encoded entry corresponding to the first entry of the structured proceeding;
output, by the second plurality of GRUs, a second encoded entry corresponding to the second entry of the structured proceeding;
receive, at a second level of the neural network of the deep learning classifier, the first encoded entry and the second encoded entry, wherein:
the second level of the neural network includes a third plurality of GRUs, a first GRU of the third plurality of GRUs is coupled to the first plurality of GRUs and is configured to receive the first encoded entry, and
a second GRU of the third plurality of GRUs is coupled to the second plurality of GRUs and is configured to receive the second encoded entry; and

output, by the second level of the neural network, an encoded structured proceeding corresponding to the structured proceeding;
receive, by a third level of the neural network of the deep learning classifier, the encoded structured proceeding; and
output, by the third level of the neural network, a probability score associated with an outcome corresponding to the encoded structured proceeding.

11. The apparatus of claim 10 , wherein:
the first plurality of GRUs comprises at least a first bidirectional gated recurrent unit (GRU) and a second GRU, and
the second plurality of GRUs comprises at least a third GRU and a fourth GRU.

12. The apparatus of claim 11 , wherein:
a first GRU of the first plurality of GRUs is configured to receive the first word token,
a second GRU of the second GRU plurality of GRUs is configured to receive the second word token,
a third GRU of the third plurality of GRUs is configured to receive the first encoded entry, and
a fourth GRU of the third plurality of GRUs is configured to receive the second encoded entry.

13. The apparatus of claim 10 , wherein the third level of the neural network comprises a softmax activation layer.

14. The apparatus of claim 10 , wherein the deep learning classifier configured to receive the first word token and the second word token further comprises the deep learning classifier configured to:
receive, at the first level of the neural network, a first plurality of word tokens, wherein the first word token is included in the first plurality of word tokens, and
receive, at the first level of the neural network, a second plurality of word tokens, wherein the second word token is included in the second plurality of word tokens.

15. The apparatus of claim 14 , wherein the first plurality of word tokens are provided to the first plurality of GRUs, and wherein the second plurality of word tokens are provided to the second plurality of GRUs.

16. The apparatus of claim 14 , wherein the second level of the neural network comprises a third plurality of GRUs, each GRU of which is configured to receive an encoded entr

Assignees

• Thomson Reuters Entpr Centre Gmbh

Inventors

• Vacek Thomas
• Song Dezhao
• Nugent Tim
• Cowling Conner
• Teo Ronald
• Schilder Frank

Classifications

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title