Granted Patent B2

US 12,737,398 · App. 18/741,439 · Granted Sep 15, 2026

Inventors: Jaehyuk Kim (Seoul, KR); Dongjun Lee (Seoul, KR); Seokhyun Hwang (Gwangju-si, KR)

Assignee: Dunamu Inc.

G06F16/3334G06F16/338G06F40/295

Abstract

Claims (76)

1 . A method of generating keyword information performed by an electronic device, the method comprising:

identifying a text set including at least one text element;

using a named entity recognition model based on deep learning, identifying keywords in the at least one text element;

based on the text set, determining degrees of association of keyword pairs included in a keyword set including the keywords by:

identifying total frequency in which the keyword pairs are included in each of the at least one text element; and

based on the keyword set and the total frequency, determining a co-occurrence graph that includes nodes and edges connecting the nodes,

wherein each of the nodes corresponds to one of the keywords included in the keyword set,

wherein a weight of each of the edges includes at least one of a first sub weight and a second sub weight that are related to a first node and a second node that are connected to the each of the edges,

wherein the first sub weight is identified based on first total frequency in which a keyword corresponding to the first node and a keyword corresponding to the second node are included together in each of the at least one text element and second total frequency in which the keyword corresponding to the first node is included in each of the at least one text element, and

wherein the second sub weight is identified based on the first total frequency and third total frequency in which the keyword corresponding to the second node is included in each of the at least one text element;

obtaining information on a query word that is input by a user; and

based on at least one among the degrees of association, generating information on response words corresponding to the query word.

2 . The method of claim 1 , wherein the text set includes unstructured data related to finance, and

wherein the at least one text element is at least one sentence in the unstructured data related to finance.

3 . The method of claim 1 , wherein the co-occurrence graph includes nodes and edges connecting the nodes,

wherein each of the nodes corresponds to one of the keywords included in the keyword set, and

wherein a weight of each of the edges is identified based on total frequency in which two keywords corresponding to each of a first node and a second node that are connected to the each of the edges are included together in each of the at least one text element.

4 . The method of claim 1 , wherein the co-occurrence graph is a directed weighted co-occurrence graph.

5 . The method of claim 4 , wherein generating of the information on the response words comprises:

based on the directed weighted co-occurrence graph, identifying at least one keyword related to the query word among the keywords included in the keyword set; and

based on a degree of association of a keyword pair consisting of one of the at least one keyword and the query word, generating the information on the response words corresponding to the query word,

wherein the degree of association of a keyword pair consisting of one of the at least one keyword and the query word is determined through a predetermined calculation using at least one sub weight related to a node corresponding to the one of the at least one keyword and a node corresponding to the query word on the directed weighted co-occurrence graph.

6 . The method of claim 5 , wherein generating of the information on the response words comprises:

based on the degree of association of a keyword pair consisting of one of the at least one keyword and the query word, filtering some keywords with a low degree of association with the query word from the at least one keyword according to a set rule.

7 . The method of claim 1 , further comprising:

identifying a keyword of a word class that is set in the at least one text element using morpheme analyzing,

wherein the keyword set further includes the keyword of the word class that is set.

8 . The method of claim 1 , wherein identifying of the keywords comprises identifying a category of each of the keywords, and

wherein generating of the information on the response words comprises:

identifying at least one keyword related to the query word among the keywords included in the keyword set; and

filtering a keyword of a set category from the at least one keyword.

9 . The method of claim 1 , wherein the information on the response words includes at least one of information on a first text element in which the query word and the one of the response words are included together in the at least one text element and information on first text data including the first text element among text data included in the text set.

10 . The method of claim 1 , wherein generating of the information on the response words comprises:

based on information on a sort order of the response words, sorting and providing the information on the response words,

wherein the information on the sort order of the response words is a size order of a degree of association between the query word and the response words.

11 . The method of claim 1 , wherein generating of the information on the response words comprises:

in case that the query word includes a plurality of query words, identifying a plurality of response word sets corresponding to each of the plurality of query words; and

determining at least one keyword that is simultaneously included in the plurality of response word sets as the response words corresponding to the plurality of query words.

12 . The method of claim 11 , wherein the sort order of the response words is determined based on sort rankings of the response words in each of the plurality of response word sets.

13 . The method of claim 1 , wherein generating of the information on the response words comprises:

classifying the response words into a first response word which is a target keyword and a second response word which is a general keyword; and

classifying and generating information on the first response word and information on the second response word separately,

wherein the first response word includes a keyword corresponding to at least one stock that is listed on an exchange.

14 . The method of claim 1 , wherein generating of the information on the response words comprise:

identifying first total frequency in which one of the response words is included together with the query word in each of the at least one text element and a category of the response word; and

based on the first total frequency and the category, generating a page where information on the response words is displayed.

15 . The method of claim 1 , wherein generating of the information on the response words comprises:

among the response words, identifying a first type response word included in a first group and a second type response word included in a second group; and

classifying and generating information on the first type response word and information on the second type response word separately.

16 . An electronic device comprising:

a transceiver;

at least one processor; and

a memory configured to store one or more instructions executed by the at least one processor,

wherein the at least one processor, executing the one or more instructions, is configured to:

identify a text set including at least one text element;

using a named entity recognition model based on deep learning, identify keywords in the at least one text element;

based on the text set, determine degrees of association of keyword pairs included in a keyword set including the keywords by:

identifying total frequency in which the keyword pairs are included in each of the at least one text element; and

based on the keyword set and the total frequency, determining a co-occurrence graph that includes nodes and edges connecting the nodes,

wherein each of the nodes corresponds to one of the keywords included in the keyword set,

wherein a weight of each of the edges includes at least one of a first sub weight and a second sub weight that are related to a first node and a second node that are connected to the each of the edges,

wherein the first sub weight is identified based on first total frequency in which a keyword corresponding to the first node and a keyword corresponding to the second node are included together in each of the at least one text element and second total frequency in which the keyword corresponding to the first node is included in each of the at least one text element, and

wherein the second sub weight is identified based on the first total frequency and third total frequency in which the keyword corresponding to the second node is included in each of the at least one text element;

obtain information on a query word that is input by a user; and

based on at least one among the degrees of association, generate information on response words corresponding to the query word.

17 . A non-transitory computer-readable recording medium having a program for executing a method by a computing device, the method comprising:

identifying a text set including at least one text element;

using a named entity recognition model based on deep learning, identifying keywords in the at least one text element;

based on the text set, determining degrees of association of keyword pairs included in a keyword set including the keywords by:

identifying total frequency in which the keyword pairs are included in each of the at least one text element; and

based on the keyword set and the total frequency, determining a directed weighted co-occurrence graph;

obtaining information on a query word that is input by a user; and

based on at least one among the degrees of association, generating information on response words corresponding to the query word by:

based on the directed weighted co-occurrence graph, identifying at least one keyword related to the query word among the keywords included in the keyword set; and

based on a degree of association of a keyword pair consisting of one of the at least one keyword and the query word, generating the information on the response words corresponding to the query word,

wherein the degree of association of a keyword pair consisting of one of the at least one keyword and the query word is determined through a predetermined calculation using at least one sub weight related to a node corresponding to the one of the at least one keyword and a node corresponding to the query word on the directed weighted co-occurrence graph.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (21)