Granted Patent B2

US 12,488,183 · App. 18/080,387 · Granted Dec 2, 2025

Inventors: Miruna Jayakrishnasamy (Vellore, IN); Prakash Ranganathan (Tamilnadu, IN)

Assignee: Verizon Patent and Licensing Inc.

G06F40/289

Abstract

Claims (101)

1 . A method, comprising:

receiving, at a device, an input text;

identifying, from the input text, first and second independent contexts without a common bounding word;

for each of the independent contexts, generating top k phrases by:

obtaining a plurality of initial phrases based on one or more linguistic features,

filtering the plurality of initial phrases to generate a plurality of filtered candidate phrases,

computing phrase features for each of the plurality of filtered candidate phrases, and

identifying the top k phrases from the plurality of filtered candidate phrases based on their respective phrase features;

selecting, for each of the independent contexts, a most insightful phrase from the top k phrases for the independent context based on an artificial neural network that integrates diverse phrase selection models established based on different types of phrase features; and

facilitating an understanding of the input text based on the selected one or more most insightful phrases.

2 . The method of claim 1 , wherein identifying the independent contexts comprises:

determining one or more topics included in the input text;

identifying, with respect to each of the one or more topics, an individual context associated with the topic to generate corresponding contexts associated with the one or more topics;

determining whether each of the corresponding contexts is an independent context based on whether the corresponding context depend on any other of the corresponding contexts or share a bounding word with any other of the corresponding contexts.

3 . The method of claim 1 , wherein the one or more linguistic features include x-gram extracted from the input text with x being an integer, wherein a specified x-gram with a smaller x is used to capture shorter phrases and a specified x-gram with a larger x is used to capture longer phrases with more details.

4 . The method of claim 2 , wherein filtering the plurality of initial phrases comprises:

identifying a first list of entities for each of the plurality of initial phrases;

obtaining a second list of association entities with respect to the first list of entities for each of the plurality of initial phrases;

receiving information associated with the independent context on a topic and a tense associated the independent context;

removing any of the plurality of initial phrases based on some pre-defined criteria defined based on the first list of entity, the second list of association entities associated with each of the plurality of initial phrases, as well as the topic and the tense associated with the independent context; and

generating the plurality of filtered candidate phrases based on remaining initial phrases.

5 . The method of claim 1 , wherein identifying the top k phrases comprises:

computing phrase features for each of the plurality of filtered candidate phrases;

ranking the plurality of filtered candidate phrases by sorting the plurality of filtered candidate phrases according to a sorting criterion to produce a ranked list of filtered candidate phrases; and

selecting the first k filtered candidate phrases from the ranked list of filtered candidate phrases as the top k phrases for the independent context.

6 . The method of claim 5 , wherein

the phrase features include a length, a sentiment score, and a part-of-speech (POS) weight assigned to the filtered candidate phrase; and

the sorting criterion specifies how to sequence the plurality of filtered candidate phrases according to an ascending or a descending order of each of the phrase features computed for the plurality of filtered candidate phrases.

7 . The method of claim 1 , wherein selecting a most insightful phrase from the top k phrases for each independent context comprises:

receiving the top k phrases selected for the independent context;

selecting a first most insightful phrase from the top k phrases based on a first phrase selection model;

selecting a second most insightful phrase from the top k phrases based on a second phrase selection model;

integrating the first and the second most insightful phrases via a deep learned phrase selection model to generate the best insightful phrase, wherein

the first phrase selection model operates based on phrase vector representations for top k phrases, the second phrase selection model operates based on phrase features of the top k phrases.

8 . A machine readable non-transitory medium having information recorded thereon, wherein the information, when read by the machine, causes the machine to perform the following steps:

receiving an input text based on which at least one insightful phrase is to be extracted;

identifying, from the input text, first and second independent contexts without a common bounding word;

for each of the independent contexts, generating top k phrases by:

obtaining a plurality of initial phrases based on one or more linguistic features,

filtering the plurality of initial phrases to generate a plurality of filtered candidate phrases,

computing phrase features for each of the plurality of filtered candidate phrases, and

identifying the top k phrases from the plurality of filtered candidate phrases based on their respective phrase features; selecting, for each of the independent contexts, a most insightful phrase from the top k phrases for the independent context based on an artificial neural network that integrates diverse phrase selection models established based on different types of phrase features, and

facilitating an understanding of the input text based on the selected one or more most insightful phrases.

9 . The medium of claim 8 , wherein the step of identifying the independent contexts comprises:

determining one or more topics included in the input text;

identifying, with respect to each of the one or more topics, an individual context associated with the topic to generate corresponding contexts associated with the one or more topics;

determining whether each of the corresponding contexts is an independent context based on whether the corresponding context depend on any other of the corresponding contexts or share a bounding word with any other of the corresponding contexts.

10 . The medium of claim 8 , wherein the one or more linguistic features include x-gram extracted from the input text with x being an integer, wherein a specified x-gram with a smaller x is used to capture shorter phrases and a specified x-gram with a larger x is used to capture longer phrases with more details.

11 . The medium of claim 9 , wherein the step of filtering the plurality of initial phrases comprises:

identifying a first list of entities for each of the plurality of initial phrases;

obtaining a second list of association entities with respect to the first list of entities for each of the plurality of initial phrases;

receiving information associated with the independent context on a topic and a tense associated the independent context;

removing any of the plurality of initial phrases based on some pre-defined criteria defined based on the first list of entity, the second list of association entities associated with each of the plurality of initial phrases, as well as the topic and the tense associated with the independent context; and

generating the plurality of filtered candidate phrases based on remaining initial phrases.

12 . The medium of claim 8 , wherein the step of identifying the top k phrases comprises:

computing phrase features for each of the plurality of filtered candidate phrases;

ranking the plurality of filtered candidate phrases by sorting the plurality of filtered candidate phrases according to a sorting criterion to produce a ranked list of filtered candidate phrases; and

selecting the first k filtered candidate phrases from the ranked list of filtered candidate phrases as the top k phrases for the independent context.

13 . The medium of claim 12 , wherein

the phrase features include a length, a sentiment score, and a part-of-speech (POS) weight assigned to the filtered candidate phrase; and

the sorting criterion specifies how to sequence the plurality of filtered candidate phrases according to an ascending or a descending order of each of the phrase features computed for the plurality of filtered candidate phrases.

14 . The medium of claim 8 , wherein the step of selecting a most insightful phrase from the top k phrases for each independent context comprises:

receiving the top k phrases selected for the independent context;

selecting a first most insightful phrase from the top k phrases based on a first phrase selection model;

selecting a second most insightful phrase from the top k phrases based on a second phrase selection model;

integrating the first and the second most insightful phrases via a deep learned phrase selection model to generate the best insightful phrase, wherein

the first phrase selection model operates based on phrase vector representations for top k phrases, the second phrase selection model operates based on phrase features of the top k phrases.

15 . A system, comprising:

an input text preprocessor implemented by a processor and configured for receiving an input text based on which at least one insightful phrase is to be extracted;

an independent context identification mechanism implemented by a processor and configured for identifying, from the input text, first and second independent contexts without a common bounding word;

a phrase extraction engine implemented by a processor and configured for generating top k phrases, wherein for each of the independent contexts, the top k phrases are generated by:

obtaining a plurality of initial phrases based on one or more linguistic features,

filtering the plurality of initial phrases to generate a plurality of filtered candidate phrases,

computing phrase features for each of the plurality of filtered candidate phrases, and

identifying the top k phrases from the plurality of filtered candidate phrases based on their respective phrase features;

a model-based phrase selector implemented by a processor and configured for selecting, for each of the independent contexts, a most insightful phrase from the top k phrases for the independent context based on an artificial neural network that integrates diverse phrase selection models established based on different types of phrase features, and

a processor for understanding the input text based on the selected one or more most insightful phrases.

16 . The system of claim 15 , wherein the independent context identification mechanism comprises:

a multiple context detector implemented by a processor and configured for

determining one or more topics included in the input text, and

identifying, with respect to each of the one or more topics, an individual context associated with the topic to generate corresponding contexts associated with the one or more topics;

a multiple context dependency detector implemented by a processor and configured for determining whether each of the corresponding contexts is dependent on any other of the corresponding contexts; and

a context bounding determiner implemented by a processor and configured for determining whether each of the corresponding contexts shares a bounding word with any other of the corresponding contexts, wherein

a corresponding context is independent when it does not depend on any other of the corresponding contexts and does not share a bounding word with any other of the corresponding contexts.

17 . The system of claim 15 , wherein the phrase extraction engine comprises a candidate phrase filter implemented by a processor and configured for filtering the plurality of initial phrases by:

identifying a first list of entities for each of the plurality of initial phrases;

obtaining a second list of association entities with respect to the first list of entities for each of the plurality of initial phrases;

receiving information associated with the independent context on a topic and a tense associated the independent context;

removing any of the plurality of initial phrases based on some pre-defined criteria defined based on the first list of entity, the second list of association entities associated with each of the plurality of initial phrases, as well as the topic and the tense associated with the independent context; and

generating the plurality of filtered candidate phrases based on remaining initial phrases.

18 . The system of claim 15 , wherein the phrase extraction engine further comprises a top-k phrase selector implemented by a processor and configured for identifying the top k phrases by:

computing phrase features for each of the plurality of filtered candidate phrases;

ranking the plurality of filtered candidate phrases by sorting the plurality of filtered candidate phrases according to a sorting criterion to produce a ranked list of filtered candidate phrases; and

selecting the first k filtered candidate phrases from the ranked list of filtered candidate phrases as the top k phrases for the independent context.

19 . The system of claim 18 , wherein

the phrase features include a length, a sentiment score, and a part-of-speech (POS) weight assigned to the filtered candidate phrase; and

the sorting criterion specifies how to sequence the plurality of filtered candidate phrases according to an ascending or a descending order of each of the phrase features computed for the plurality of filtered candidate phrases.

20 . The system of claim 15 , wherein the artificial neural network includes:

a first phrase selection model implemented by a processor and configured for selecting a first most insightful phrase from the top k phrases associated with an independent context based on phrase vector representations of the top k phrases;

a second phrase selection model implemented by a processor and configured for selecting a second most insightful phrase from the top k phrases based on phrase features of the top k phrases; and

an integration layer implemented by a processor and configured for integrating the first and the second most insightful phrases to generate the best insightful phrase.

Assignments (1)

Continuity (1)

References Cited (10)