Phrase extraction text analysis method and system

• US 10,325,021 B2
• Filed: 06/19/2017
• Issued: 06/18/2019
• Est. Priority Date: 06/19/2017
• Status: Active Grant

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: GM Global Technology Operations LLC (General Motors Company)
• Inventors: Peranandam, Prakash Mohan, De, Soumen
• Primary Examiner(s): Shin, Seong-Ah A
• Application Number:
  US15/627,380
• Publication Number:
  US 20180365222A1
• Time in Patent Office: 729 Days
• Field of Search: 704 9, 707758
• CPC Class Codes: G06F 40/247 Thesauruses; Synonyms
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G08G 1/20 Monitoring the location of vehicles belonging to a group, e.g. fleet of vehicles, countable or determined number of vehicles

Abstract

A system and method for extracting a relevant phrase from text. The system and method may build a plurality of n-gram phrases using a seed from a seed list as a start, a middle, or an end of each n-gram phrase. The seed list may be directed to a specific vehicle system and each seed may indicate a symptom, part, or action to extract relevant phrases from vehicle information verbatims. The plurality of n-gram phrases may be filtered to obtain one or more relevant phrases. The filtering process may include calculating an external relevance factor, an internal relevance factor, or a context pattern relevance factor.

Claims

1. A method for extracting a relevant phrase from text, comprising the steps of:

accessing a vehicle information verbatim from a database with a server having a processor that is at least partially configured as a special purpose text analyzer;

tokenizing the vehicle information verbatim;

building a plurality of n-gram phrases from the vehicle information verbatim with the server, wherein the plurality of n-gram phrases include a seed from a seed list as a start, a middle, or an end of each n-gram phrase of the plurality of n-gram phrases, wherein the seed list includes a plurality of seeds, each seed being directed to a vehicle-related component or a vehicle-related functionality; and

filtering the plurality of n-gram phrases with the server to obtain the relevant phrase or an irrelevant phrase, wherein the filtering includes calculating an external relevance factor, an internal relevance factor, and a context pattern relevance factor, and the filtering includes using a weak filtering rule set or a strong filtering rule set,

wherein the weak filtering rule set is used to conjunctively consider the external relevance factor, the internal relevance factor, and the context pattern relevance factor so that n-gram phrases are irrelevant if an irrelevance threshold is met for each of the external relevance factor, the internal relevance factor, and the context pattern relevance factor, and

wherein the strong filtering rule set is used to disjunctively consider the external relevance factor, the internal relevance factor, and the context pattern relevance factor so that n-gram phrases are irrelevant if an irrelevance threshold is met for one of the external relevance factor, the internal relevance factor, or the context pattern relevance factor.

2. The method of claim 1, wherein the seed list is directed to a specific vehicle system.

3. The method of claim 2, wherein the seeds identify symptoms, parts, or actions.

4. The method of claim 1, wherein a training phase and a testing phase are performed, and the external relevance factor accounts for a total number of n-gram phrases having a same length in the training phase, the total number of n-gram phrases having the same length in the testing phase, a frequency of a word sequence in the training phase, and the frequency of the word sequence in the testing phase.

5. The method of claim 1, wherein the internal relevance factor takes into account a term frequency-inverse document frequency (tf-idf) analysis of vehicle information in the database.

6. The method of claim 1, wherein the context pattern relevance factor is a binary value that depends on whether the n-gram phrase has a relevant part of speech (POS) tag sequence or an irrelevant POS tag sequence.

7. The method of claim 1, further comprising a step of categorizing terms in the relevant phrase in a domain according to a property of the terms and a relation between the terms in a data organizational system.

8. A method for extracting a relevant phrase from text, comprising the steps of:

accessing a verbatim from a database with a server having a processor that is at least partially configured as a special purpose text analyzer;

tokenizing the verbatim;

building a plurality of n-gram phrases with the server from the verbatim, wherein the plurality of n-gram phrases include a seed from a seed list as a start, a middle, or an end of each n-gram phrase of the plurality of n-gram phrases, wherein the seed list includes a plurality of seeds;

calculating an external relevance factor for each n-gram phrase of the plurality of n-gram phrases;

calculating an internal relevance factor for each n-gram phrase of the plurality of n-gram phrases;

calculating a context pattern relevance factor for each n-gram phrase of the plurality of n-gram phrases; and

using the external relevance factor, the internal relevance factor, the context pattern relevance factor, or a combination of one or more of the external relevance factor, the internal relevance factor, and the context pattern relevance factor to identify the relevant phrase or an irrelevant phrase from the plurality of n-gram phrases, wherein a weak filtering rule set is used to conjunctively consider the external relevance factor, the internal relevance factor, and the context pattern relevance factor so that n-gram phrases are irrelevant if an irrelevance threshold is met for each of the external relevance factor, the internal relevance factor, and the context pattern relevance factor.

9. The method of claim 8, wherein the verbatim is a vehicle information verbatim and the seed list is directed to a specific vehicle system.

11. The method of claim 9, wherein the seeds identify symptoms, parts, or actions.

10. A method for extracting a relevant phrase from text, comprising the steps of:

accessing a verbatim from a database with a server having a processor that is at least partially configured as a special purpose text analyzer;

tokenizing the verbatim;

building a plurality of n-gram phrases with the server from the verbatim, wherein the plurality of n-gram phrases include a seed from a seed list as a start, a middle, or an end of each n-gram phrase of the plurality of n-gram phrases, wherein the seed list includes a plurality of seeds;

calculating an external relevance factor for each n-gram phrase of the plurality of n-gram phrases;

calculating an internal relevance factor for each n-gram phrase of the plurality of n-gram phrases;

calculating a context pattern relevance factor for each n-gram phrase of the plurality of n-gram phrases; and

using the external relevance factor, the internal relevance factor, the context pattern relevance factor, or a combination of one or more of the external relevance factor, the internal relevance factor, and the context pattern relevance factor to identify the relevant phrase or an irrelevant phrase from the plurality of n-gram phrases, wherein a strong filtering rule set is used to disjunctively consider the external relevance factor, the internal relevance factor, and the context pattern relevance factor so that n-gram phrases are irrelevant if an irrelevance threshold is met for one of the external relevance factor, the internal relevance factor, or the context pattern relevance factor.

12. The method of claim 10, wherein the verbatim is a vehicle information verbatim and the seed list is directed to a specific vehicle system.

13. The method of claim 12, wherein the seeds identify symptoms, parts, or actions.