SYSTEMS AND METHODS FOR AUTOMATIC IDENTIFICATION OF POTENTIAL MATERIAL FACTS IN DOCUMENTS

• US 20160140210A1
• Filed: 11/18/2015
• Published: 05/19/2016
• Est. Priority Date: 11/19/2014
• Status: Active Grant

• Current Assignee: LexisNexis Group Inc. (RELX PLC)
• Original Assignee: LexisNexis Group Inc. (RELX PLC)
• Inventors: Osgood, Gene, Pendyala, Mahesh, Myers, Jacob Aaron
• Granted Patent: US 10,331,782 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/93 Document management systems
• CPC Class Codes: G06F 40/205 Parsing
• CPC Class Codes: G06F 40/279 Recognition of textual entities
• CPC Class Codes: G06Q 50/18 Legal services

Abstract

Systems and methods to identify potential material fact sentences in electronic legal documents obtained from electronic repositories are disclosed. A system includes a processing device and a storage medium in communication with the processing device. The storage medium includes programming instructions that cause the processing device to obtain a document and parse text within the document to determine whether each paragraph in the document is a fact paragraph, a discussion paragraph, or an outcome paragraph based on at least one of a heading associated with the paragraph and features of the paragraph. The storage medium further includes programming instructions that cause the processing device to extract each sentence in the fact paragraph, direct a trained sentence classifier to determine whether each sentence is a potential material fact sentence or a non-material fact sentence based on features of the sentence, and identify potential material fact sentences.

Claims

1. A system to identify potential material fact sentences in electronic legal documents obtained from electronic repositories, the system comprising:

a processing device; and

a non-transitory, processor-readable storage medium in communication with the processing device, the non-transitory, processor-readable storage medium comprising one or more programming instructions that, when executed, cause the processing device to;

obtain an electronic legal document from a repository,

parse text within the electronic legal document to determine whether each one of one or more paragraphs in the legal document is a fact paragraph, a discussion paragraph, or an outcome paragraph based on at least one of a heading associated with the paragraph and one or more features of the paragraph, and

for each one of the one or more paragraphs that is a fact paragraph;

extract each one of one or more sentences in the fact paragraph,

direct a trained sentence classifier to determine whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence based on one or more features of the sentence, and

identify one or more potential material fact sentences from the one or more sentences based on the determination.

2. The system of claim 1, wherein the one or more features of the sentence is selected from a group consisting of a number of noun phrases, a number of verb phrases, a number of dates, a number of time stamps, a number of monetary values, a number of lower court actions, a number of present court actions, a number of plaintiff actions, a number of legal phrases, a number of legal concepts, a number of non-material fact words, and a number of non-material fact phrases.

3. The system of claim 1, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by running a natural language parser on each one of the one or more sentences to determine the one or more features of the sentence.

4. The system of claim 1, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated by a support vector machine algorithm from training data.

5. The system of claim 1, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated by a decision tree algorithm from training data.

6. The system of claim 1, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated by a naï

ve Bayes algorithm from training data.

7. The system of claim 1, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated from a stacking committee of classifiers algorithm from training data and data outputted from one or more base classifiers.

8. The system of claim 1, wherein the heading is a facts heading, a discussion heading, or a outcome heading.

9. The system of claim 1, wherein the one or more features of the paragraph is selected from a group consisting of a position of the paragraph, a number of cases, a number of statutes, a number of past tense verbs, a number of present court words, a number of lower court words, a number of legal phrases, a number of defendant words, a number of plaintiff words, a number of dates, a number of signal words, and a number of footnotes.

10. A method to identify potential material fact sentences in electronic legal documents obtained from electronic repositories, the method comprising:

obtaining, by a processing device, an electronic legal document from a repository;

parsing, by the processing device, text within the electronic legal document to determine whether each one of one or more paragraphs in the legal document is a fact paragraph, a discussion paragraph, or an outcome paragraph based on at least one of a heading associated with the paragraph and one or more features of the paragraph; and

for each one of the one or more paragraphs that is a fact paragraph;

extracting, by the processing device, each one of one or more sentences in the fact paragraph,

directing, by the processing device, a trained sentence classifier to determine whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence based on one or more features of the sentence, and

identifying, by the processing device, one or more potential material fact sentences from the one or more sentences based on the determination.

11. The method of claim 10, wherein the one or more features of the sentence is selected from a group consisting of a number of noun phrases, a number of verb phrases, a number of dates, a number of time stamps, a number of monetary values, a number of lower court actions, a number of present court actions, a number of plaintiff actions, a number of legal phrases, a number of legal concepts, a number of non-material fact words, and a number of non-material fact phrases.

12. The method of claim 10, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by running a natural language parser on each one of the one or more sentences to determine the one or more features of the sentence.

13. The method of claim 10, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated by a support vector machine algorithm from training data.

14. The method of claim 10, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated by a decision tree algorithm from training data.

15. The method of claim 10, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated by a naï

ve Bayes algorithm from training data.

16. The method of claim 10, wherein the trained sentence classifier determines whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence by scoring the one or more features based on a trained model generated from a stacking committee of classifiers algorithm from training data and data outputted from one or more base classifiers.

17. The method of claim 10, wherein the heading is a facts heading, a discussion heading, or a outcome heading.

18. The method of claim 10, wherein the one or more features of the paragraph is selected from a group consisting of a position of the paragraph, a number of cases, a number of statutes, a number of past tense verbs, a number of present court words, a number of lower court words, a number of legal phrases, a number of defendant words, a number of plaintiff words, a number of dates, a number of signal words, and a number of footnotes.

19. A method to identify potential material fact sentences in electronic legal documents obtained from electronic repositories, the method comprising:

obtaining, by a processing device, an electronic legal document from a repository;

parsing, by the processing device, text within the electronic legal document to determine whether each one of one or more paragraphs in the legal document is a fact paragraph, a discussion paragraph, or an outcome paragraph based on at least one of a heading associated with the paragraph and one or more features of the paragraph; and

for each one of the one or more paragraphs that is a fact paragraph;

extracting, by the processing device, each one of one or more sentences in the fact paragraph,

directing, by the processing device, a natural language parser to parse each one of the one or more sentences in the fact paragraph to determine a number of noun phrases and a number of verb phrases,

extracting, by the processing device, one or more features selected from a number of dates, a number of time stamps, a number of monetary values, a number of lower court actions, a number of present court actions, a number of plaintiff actions, a number of defendant actions, a number of legal phrases, a number of legal concepts, a number of non-material fact words, and a number of non-material fact phrases from each one of the one or more sentences,

scoring, by the processing device, each one of the one or more sentences based on the number of noun phrases, the number of verb phrases, and the one or more features, and

determining, by the processing device, whether each one of the one or more sentences is a potential material fact sentence or a non-material fact sentence based on the scoring.

20. The method of claim 19, wherein the scoring comprises scoring each one of the one or more sentences based on a trained model generated by one or more base classifiers from training data.