KR102117281B1 - Method for generating chatbot utterance using frequency table - Google Patents

Abstract

The present invention relates to a method for generating a chatbot utterance using a frequency table. The method includes generating a frequency table using the user's past remarks, and generating a chatbot's remark to respond to the user's new remarks using the frequency table, and the frequency table is generated. Generating comprises: generating a first dependent parse tree for one or more sentences of the user's past remarks, i) a first object corresponding to a predetermined first vertex from the first dependent parse tree , ii) extracting a predetermined one or more first data sets defined as second objects corresponding to a predetermined second vertex, and generating a first table in which the extracted one or more first data sets are input; Calculating a frequency of the one or more first data sets in the first table, and one or more second data sets having a predetermined frequency and meaningful to the user from the one or more first data sets in the first table. Extracting and generating the frequency table, and generating the chatbot's speech comprises: generating a second dependent parsing tree for one or more sentences of the user's new speech, and the second dependent syntax Extracting a set of one or more third data sets defined by i) a first object corresponding to a predetermined first vertex, and ii) a second object corresponding to a predetermined second vertex, from the frequency table Comparing the one or more second data sets with the one or more third data sets, and wherein the one or more second data sets and the one or more third data sets in the frequency table have a predetermined similarity 3 extracting a data set as a speech candidate data set, using a natural language generation algorithm, generating a conversation candidate sentence corresponding to the extracted speech candidate data set, and using the conversation candidate sentence to make a chatbot's speech And generating.

Description

METHOD FOR GENERATING CHATBOT UTTERANCE USING FREQUENCY TABLE

The present invention relates to a method for generating chatbot speech using a frequency table.

Recently, the demand for natural language processing that enables communication between humans and chatbots is increasing in various fields such as personal assistant, messenger, finance, and distribution. In the case of a chatbot that processes natural language in real time, the method of storing the entire conversation log obtained through the conversation between the user and the chatbot and processing information when there is a user's request is not suitable for rapid natural language processing due to the large amount of data to be processed. . Accordingly, there is a need for a technique that systematically stores information and enables rapid processing of information when there is a user's request.

Publication Patent Publication No. 10-2016-0082996, 2016.06.30

The problem to be solved by the present invention is to provide a method for generating a chatbot utterance using a frequency table.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the following description.

Method for generating a chatbot utterance using a frequency table according to an aspect of the present invention for solving the above-described problems is a step of generating a frequency table using the user's past speech, and the user's new speech using the frequency table Generating a utterance of the chatbot for responding to, and generating the frequency table comprises: generating a first dependent parsing tree for one or more sentences of the user's past utterance; 1) extracting, from the dependent parsing tree, i) a first object corresponding to a predetermined first vertex, and ii) extracting a predetermined one or more first data sets defined as a second object corresponding to the predetermined second vertex. Generating a first table in which the one or more first data sets are input, calculating a frequency of the one or more first data sets in the first table, and the one or more first data in the first table And generating the frequency table by extracting one or more second data sets having a predetermined frequency from the set and meaningful to the user, and generating the speech of the chatbot, wherein one or more of the new speech of the user is generated. Generating a second dependent parsing tree for the sentence, from the second dependent parsing tree i) to a first object corresponding to a predetermined first vertex, and ii) to a second object corresponding to a predetermined second vertex. Extracting a defined one or more third data sets, comparing the one or more second data sets in the frequency table with the one or more third data sets, and the one or more second data sets in the frequency table And when the one or more third data sets have a predetermined similarity, extracting the one or more third data sets as a speech candidate data set, and a conversation corresponding to the extracted speech candidate data set using a natural language generation algorithm. Generating a candidate sentence, and generating a chatbot remark using the conversation candidate sentence.

In some embodiments, generating the frequency table by extracting one or more second data sets having a predetermined frequency and meaningful to the user from the one or more first data sets of the first table includes: a plurality of different users Generating a third dependent parsing tree for one or more sentences of the past remark of i) from the third dependent parsing tree i) a third object corresponding to a predetermined third vertex, ii) a predetermined fourth vertex Extracting a predetermined one or more fourth data sets defined as a fourth object corresponding to, generating a second table in which the one or more fourth data sets are input, and the one or more fourth ones in the second table Calculating a frequency of the data set, and comparing the frequency calculation result of the fourth data set with the frequency calculation result of the first data set, and based on the comparison result, the one or more first data sets to the user And extracting the meaningful one or more second data sets.

In some embodiments, the first object or the third object is a noun phrase, and the second object or the fourth object is defined as a verb phrase.

In some embodiments, the first object, the second object, the third object, or the fourth object is converted into a title word with derivative elements removed through morphological and dictionary analysis.

In some embodiments, when the first object, the second object, the third object, or the fourth object form a meaning by combining a plurality of words, a plurality of words are treated as one object. do.

In some embodiments, when the one or more second data sets of the frequency table and the one or more third data sets have a predetermined similarity, extracting the one or more third data sets as a speech candidate data set may include: The similarity between the one or more second data sets and the one or more third data sets is determined using a Word2Vec technique.

Other specific matters of the present invention are included in the detailed description and drawings.

According to the method for generating chatbot utterances using the frequency table of the present invention, it is possible to store and manage the user's past utterances into a frequency table in a structured and organized manner.

In addition, by making the user's remarks into a frequency table and sorting them in order of frequency, it is possible to detect a group of words meaningful to the user according to the Jeep's law.

In addition, before applying the natural language processing mechanism, the frequency table is used to determine whether it is a meaningful word in natural language generation, and the similarity between old and new words is determined, and chatbot comments are generated by grouping words in the frequency table. I know if it can be done.

Therefore, by using the frequency table to determine whether the user's new speech is meaningful, and whether the new speech is similar to the past speech, a chatbot's speech can be quickly generated when the user's response is needed, enabling real-time conversation between the user and the chatbot. To do.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a schematic flowchart of a method for generating a chatbot utterance using a frequency table according to an embodiment of the present invention.
2 is a schematic flowchart of a method for generating a frequency table according to an embodiment of the present invention.
3 is an exemplary diagram of a dependent parsing tree according to an embodiment of the present invention.
4 is a schematic flowchart of a method of extracting a second data set meaningful to a user according to an embodiment of the present invention.
5 is a schematic flowchart of a method for generating a chatbot's remark according to an embodiment of the present invention.
6 is a diagram illustrating a computer system for generating a chatbot service program according to an embodiment of the present invention.
7 is a diagram illustrating a chatbot system that generates a chatbot's comments according to an embodiment of the present invention.

Advantages and features of the present invention, and methods for achieving them will be clarified with reference to embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, and only the present embodiments allow the disclosure of the present invention to be complete, and are common in the technical field to which the present invention pertains. It is provided to fully inform the skilled person of the scope of the present invention, and the present invention is only defined by the scope of the claims.

The terminology used herein is for describing the embodiments and is not intended to limit the present invention. In the present specification, the singular form also includes the plural form unless otherwise specified in the phrase. As used herein, “comprises” and / or “comprising” does not exclude the presence or addition of one or more other components other than the components mentioned. The same reference numerals throughout the specification refer to the same components, and “and / or” includes each and every combination of one or more of the components mentioned. Although "first", "second", etc. are used to describe various components, it goes without saying that these components are not limited by these terms. These terms are only used to distinguish one component from another component. Therefore, it goes without saying that the first component mentioned below may be the second component within the technical spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used in the present specification may be used as meanings commonly understood by those skilled in the art to which the present invention pertains. In addition, terms defined in the commonly used dictionary are not ideally or excessively interpreted unless explicitly defined.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a schematic flowchart of a method for generating a chatbot utterance using a frequency table according to an embodiment of the present invention.

Referring to FIG. 1, a method of generating a chatbot utterance using a frequency table includes generating a frequency table (S100) and generating a chatbot utterance (S200).

In step S100, the computer generates a frequency table using the user's past remarks.

In step S200, the computer generates a chatbot's utterance to respond to the user's new utterance using the frequency table.

2 is a schematic flowchart of a method for generating a frequency table according to an embodiment of the present invention.

Referring to FIG. 2, a method of generating a frequency table includes: generating a first dependent parsing tree 300 (S110), extracting a first data set (S120), and generating a first table (S130), calculating the frequency of the first data set (S140) and extracting the second data set to generate a frequency table (S150).

In step S110, the computer generates a first dependent parsing tree 300 relating to one or more sentences of the past remarks of the user (which may be referred to as a specific user).

The dependency parsing tree 300 is a plurality of words included in one or more sentences of a user's past remarks, and is a plurality of vertices, and a plurality of vertices are connected according to a dependency relationship between the plurality of words.

The dependency relationship between words indicates, for example, how the words and words are grammatically or semantically related in a sentence, as shown in Table 1.

Dependence Explanation Dependence Explanation nsubj Sentence subject det Limited company dobj Direct object poss possessive case iobj An indirect object conj conjunctive ccomp Clause bore cc coordinator nmod Modifiers of nouns compound compound word amod Adjectives punct full stop

The dependent parsing tree 300 is a matter obvious to those skilled in the art to which the present invention pertains, and a detailed description thereof will be omitted below because it may obscure the subject matter of the present invention.

In some embodiments, before the computer generates the first dependent parsing tree 300, if there is a plurality of words representing the same subject, such as pronouns, in one or more sentences of the user's past remarks, they are distinguished from pronouns representing other objects. To make it possible. The computer performs the same operation in not only pronouns, but also expressions referring to directive pronouns and third parties. For example, “Jane likes apple. She is vegetarian. ”And“ Alice goes to mountain. In the sentence “She likes hiking.”, “She” in the first sentence means “Jane”, and “She” in the second sentence means “Alice”. At this time, “She” in the first sentence and “She” in the second sentence are distinguished by pronouns indicating different objects, so they are converted into separate expressions.

In some embodiments, the computer may also include words such as conjunctions, articles, adverbs, qualifiers, such as “and”, “the”, “also”, and “which” as a vertex of the dependent parsing tree 300. .

In some embodiments, when a plurality of words corresponding to a plurality of vertices of the first dependent parsing tree 300 are combined with each other to form a meaning, the computer merges the plurality of vertices into one vertex. For example, a computer combines two words to form a compound word, or combines possessive and noun into one vertex. For example, “us” and “grandmother” can be combined and integrated into “my grandmother”. In addition, “bag” and “baggage” can be combined to form “baggage”. In addition, “coming” and “going” can be combined to form “coming and going”.

Examples of rules for integrating a plurality of vertices of the dependency parsing tree 300 into one vertex are not limited thereto, and other arbitrary rules not illustrated may be applied.

In step S120, the computer from the first dependent parsing tree 300 i) a predetermined object defined by a first object corresponding to a predetermined first vertex, ii) a second object corresponding to a predetermined second vertex The first data set is extracted.

The first object of the one or more first data sets may be the subject of the sentence, and the second object may be the verb of the sentence. Also, the first object of the one or more first data sets may be a noun or a noun phrase, and the second object may be a verb or a verb phrase.

The first object or the second object may be converted into a title word in which a derivative element is removed through morphological and dictionary analysis, and thus may be a first data set. For example, “walk”, “walk”, and “walk” can be converted to “walk”.

The computer may generate a semantic graph database having a first object and a second object of the extracted one or more first data sets as vertices, and a relationship between the first object and the second object as a directional edge.

In some embodiments, the directional trunk includes relationship information between the first object and the second object, as well as time stamp information of the sentence corresponding to the data set. For example, time information at a time when a user defines a relationship between a first object and a second object by speaking a sentence corresponding to a data set may be included in the directional trunk.

In step S130, the computer generates a first table in which the extracted one or more first data sets are input.

The computer initializes the first table before inputting the extracted one or more first data sets to the first table. Here, “initializing” means securing (or allocating) memory space for storing table data.

In some embodiments, the computer may pair the first object corresponding to the subject and the second object corresponding to the verb in the subject column and the verb column of the first table, respectively.

In step S140, the computer calculates the frequency of one or more first data sets in the first table.

The computer may follow the Jeep's law by calculating the frequency of the first data set input to the first table, and listing the first data set in the order in which the calculated frequency is high. For example, “I-be”, “I-take”, and “I-get”, which are frequently found in sentences, count the ranks low.

The simplest example of the Jeep's law is the “1 / f function”, where the frequency following the Jeep distribution is given in order of magnitude, and the frequency corresponding to 2 is half the frequency of 1 above. The frequency above 3 is 1/3 of the frequency above 1. In this way, the frequency above n becomes 1 / n of the frequency above 1. That is, when word frequencies are sorted in order of word frequency, there is an inverse relationship between frequency and ranking. That is, the word frequency can be expressed by the following formula.

In step S150, the computer generates a frequency table by extracting one or more second data sets having a predetermined frequency and meaningful to a user from one or more first data sets of the first table.

A method of extracting a data set meaningful to the user of step S150 will be described in detail in FIG. 4 below.

3 is an exemplary diagram of a dependent parsing tree according to an embodiment of the present invention.

Referring to FIG. 3, a dependency parsing tree 300 of an example user's remark “My brother and I like the new Star Wars and also Star Trek which are coming out this year.” Is shown.

In the exemplary dependency parsing tree 300, “My brother and I like the new Star Wars and also Star Trek which are coming out this year.” The sentences are “My” (301), “brother” (302), “and” (309), “I” (305), “like” (303), “the” (310), “new” (311) , “Star” (312), “Wars” (306), “and” (309), “also” (313), “Star” (312), “Trek” (307), “which” (314), “Are” (315), “coming” (308), “out” (316), “this” (318), “year” (317), “.” (304) as the sign of each word and sentence Is divided.

In the example dependency parsing tree 300, “My” 301, “brother” 302, “and” 309, “I” 305, “like” 303, “the” ( 310), “new” (311), “Star” (312), “Wars” (306), “and” (309), “also” (313), “Star” (312), “Trek” (307 ), “Which” (314), “are” (315), “coming” (308), “out” (316), “this” (318), “year” (317), “.” (304) The sign of each word and sentence of is the peak, and the dependency relationship of each word is connected by the directional trunk. For example, “My” (301) and “brother” (302) have a possessive relationship, “brother” (302) and “I” (305) have a conjunctive relationship, and “Star” ( 312) and “Wars” 306 are analyzed to have a compound relationship, and are connected by directional trunks.

As described above, a plurality of predetermined vertices of the dependent parsing tree 300 may be integrated and transformed into one vertex. In the exemplary dependent parsing tree 300, “My” 301 and “brother” 302 can be combined and incorporated into “My brother”. In addition, “Star” 312 and “Trek” 307 can be combined and integrated into “Star Trek”. In addition, “are” 315 and “coming” 308 and “out” 316 can be combined to be integrated into “are coming out”.

The computer extracts the data set from the dependency parsing tree 300 of the example user's remark “My brother and I like the new Star Wars and also Star Trek which are coming out this year.”, And extracts the extracted data set. Create a semantic graph database with vertices and directional edges.

4 is a schematic flowchart of a method of extracting a second data set meaningful to a user according to an embodiment of the present invention.

Referring to FIG. 4, a method for extracting a second data set meaningful to a user includes generating a third dependent parsing tree 300 (S151), extracting a fourth data set (S152), and second It includes creating a table (S153), calculating a frequency of the fourth data set (S154), and extracting a second data set meaningful to the user (S155).

In step S151, the computer generates a third dependent parsing tree 300 for one or more sentences of the past remarks of a plurality of different users (which may be referred to as global users).

The computer generates a third dependent parsing tree 300 for one or more sentences of past remarks of a plurality of different users by the method described above in FIG. 2. The computer stores the past remarks of various users in the database, and uses this to generate the third dependent parsing tree 300. The computer may cause the user's past and present statements composed of the first dependent parsing tree 300 and the second dependent parsing tree 300 to also be composed of the third dependent parsing tree 300.

In step S152, the computer from the third dependent parsing tree 300 i) a third object corresponding to a predetermined third vertex, ii) a predetermined one or more objects defined as a fourth object corresponding to the predetermined fourth vertex The fourth data set is extracted.

The third object of the one or more fourth data sets may be the subject of the sentence, and the fourth object may be the verb of the sentence. Also, the third object of the one or more fourth data sets may be a noun or a noun phrase, and the fourth object may be a verb or a verb phrase.

The third object or the fourth object may be converted into a title word in which a derivative element is removed through morphological and dictionary analysis, and thus may be a fourth data set. For example, “walk”, “walk”, and “walk” can be converted to “walk”.

In step S153, the computer generates a second table in which one or more fourth data sets are input.

In some embodiments, the computer may pair the third object corresponding to the subject and the fourth object corresponding to the verb in the subject column and the verb column of the second table, respectively.

In step S154, the computer calculates the frequency of one or more fourth data sets in the second table.

The computer can follow the Jeep's law by calculating the frequency of the fourth data set input to the second table and listing the fourth data set in the order of the highest calculated frequency. For example, “I-be”, “I-take”, and “I-get”, which are frequently found in sentences, count low.

In step S155, the computer compares the frequency calculation result of the fourth data set with the frequency calculation result of the first data set, and based on the comparison result, one or more first data sets into one or more second data sets meaningful to the user. To extract.

The computer determines that the first data set is meaningful to the user when the first data set is not found in the fourth data set (that is, the frequency is 0) or when the frequency is lower than a predetermined reference value. And extract it as a second data set.

Of the specific user data sets, only the low-frequency pair is meaningful in a sentence. Accordingly, if a pair of high frequency among the global user's data set is included in the data set of a specific user, it may be determined that the data set is meaningless in a sentence. For example, assuming that the occurrence of a pair (my father, eat) of a particular user's data set is x% of a particular user pair, and the occurrence of a pair of (my father, eat) of a global user's data set is y% of a global user pair. Assuming that, if the difference between y% and x% is greater than or equal to a predetermined reference value (my father, eat), it can be determined that the pair is meaningful for a specific user.

5 is a schematic flowchart of a method for generating a chatbot's remark according to an embodiment of the present invention.

Referring to FIG. 5, a method for generating a chatbot's remarks includes: generating a second dependent parsing tree 300 (S210), extracting a third data set (S220), second data set and third Comparing the data set (S230), extracting the third data set as a speech candidate data set (S240), generating a conversation candidate sentence (S250), and generating a chatbot's speech (S260) do.

In step S210, the computer generates a second dependent parsing tree 300 for one or more sentences of the new remark of the user (which may be referred to as a specific user).

The computer generates a second dependent parse tree 300 for one or more sentences of the user's new remark by the method described above in FIG. 2.

In step S220, the computer from the second dependent parsing tree 300 i) a predetermined object defined by a first object corresponding to a predetermined first vertex, ii) a second object corresponding to the predetermined second vertex Extract the third data set.

The first object of the one or more third data sets may be the subject of the sentence, and the second object may be the verb of the sentence. Also, the first object of the one or more third data sets may be a noun or a noun phrase, and the second object may be a verb or a verb phrase.

The first object or the second object may be converted into a title word in which a derivative element is removed through morphological and dictionary analysis, and thus may be a third data set. For example, “walk”, “walk”, and “walk” can be converted to “walk”.

In step S230, the computer compares the one or more second data sets of the frequency table with the one or more third data sets.

The computer compares one or more second data sets and one or more third data sets to determine whether they have a predetermined similarity.

In step S240, the computer extracts one or more third data sets as candidate speech data sets when the one or more second data sets of the frequency table and the one or more third data sets have a predetermined similarity.

The computer determines whether the third data set extracted from the new speech of the specific user is meaningful compared to the fourth data set extracted from the global user's speech, and the meaningful second data set and the third data among the past speeches of the specific user The set of similar candidates is extracted by determining whether the three are similar.

In some embodiments, if a particular user says “I cooked with my dad last weekend.”, The computer relies on the nouns “my dad” and “I”, the verb “cook” as an object in the dependency parsing tree 300 Can be extracted with Then, the computer can extract data sets of (I, cook), (my dad, cook). Since the data sets of (I, cook) and (my dad, cook) are meaningful compared to the data set of the global user's speech, the similarity of the data set of the user's past speech is determined. The dataset (my dad, cook) of a particular user's new speech is similar to the dataset (my dad, eat) of a particular user's past speech. Therefore, the computer can extract (my dad, eat) as a speech candidate data set. The computer may generate “Did your father eat a lot?” With a natural language generation algorithm that reflects (my dad, eat) extracted as a speech candidate data set in the following S250 to S260 methods.

In another embodiment, if the user says "I watched a movie with my father last night." It can be extracted as an object of 300. Then, the computer can extract data sets of (I, watch), (movie, watch), and (my father, watch). The data sets of (I, watch), (movie, watch), and (my father, watch) are meaningful compared to the data set of the global user's speech, so the similarity of the data set of the user's past speech is judged. . However, the datasets of a specific user's new remarks (I, watch), (movie, watch), and (my father, watch) are not similar to the dataset of a specific user's past remarks. Therefore, the computer may not extract (I, watch), (movie, watch), and (my father, watch) as a candidate speech data set.

In another embodiment, if the user says “I am tired.”, The computer may extract the nouns “I” and the verb “be” as objects in the dependent parsing tree 300. Then, the computer can extract the data set of (I, be). However, since the data set of (I, be) has no meaning compared to the data set of the global user's speech, the computer may not extract (I, be) as the speech candidate data set.

When the computer determines the similarity between a specific user's new utterance 3rd data set and the past utterance 2nd data set, the second data set and the third data set may be used by using the Word2Vec technique or above a predetermined threshold. By determining whether they are similar, similarity can be determined.

In step S250, the computer generates a dialogue candidate sentence corresponding to the extracted speech candidate data set using a natural language generation algorithm. As a natural language generation algorithm, various algorithms well known in the art to which the present invention pertains can be used.

In step S260, the computer generates a chatbot's remark using the dialogue candidate sentence.

The computer uses a deep learning algorithm in generating a chatbot's remark using one or more candidate sentences, and calculates an evaluation score of the user's chatbot's context and the user's new remark for a new remark. The computer generates a chatbot's remark using one or more candidate sentences in which the calculated evaluation score is greater than or equal to a predetermined reference value.

The computer creates a context string of the entire conversation between the user and the chatbot about the user's new speech. For example, the context string may be a concatenation of words from the user's past remarks to new remarks in the entire conversation.

The computer calculates the evaluation scores of candidate sentences generated in step S250 using a pre-trained score system using deep learning. The score system receives the form of [context string, user's new remark, candidate sentence], and calculates the evaluation score of the candidate sentence generated in step S250 based on the context string and the user's new remark.

In some embodiments, the computer may generate a chatbot's comments using candidate sentences with a high evaluation score.

In another embodiment, the computer may analyze the user's conversation pattern using the second data set generated in the frequency table and visualize and provide the user with the conversation pattern. For example, the computer may visualize the second data set as a word cloud, a time-dependent stacked bar chart over time, and provide it to the user.

6 is a diagram illustrating a computer system for generating a chatbot service program according to an embodiment of the present invention.

Referring to FIG. 6, the computer system 400 for generating a chatbot service program includes a memory 401, a bus 402, and a processor 403.

The memory 401 stores instructions and data for executing the chatbot speech generation method using the above-described frequency table.

The processor 403 interprets and operates instructions and data stored in the memory 401, and outputs it to the chatbot engine of the chatbot system 500.

The bus 402 connects the memory 401 and the processor 403 to transmit instructions and data.

7 is a diagram illustrating a chatbot system that generates a chatbot's speech according to an embodiment of the present invention.

Referring to FIG. 7, a chatbot system 500 that generates chatbot comments includes a user terminal 501, a service server 502, and a chatbot server 503.

The service server 502 provides a chatbot service to the user, and when the user's speech is input from the user terminal 501, the user's speech is transmitted to the chatbot server 503.

The chatbot server 503 includes a chatbot engine and a database. The chatbot engine is a chatbot speech generation method using the above-described frequency table, and generates a chatbot speech from the user speech. The database stores user comments transmitted from the service server 502, and transmits the user comments to the chatbot engine when the chatbot comments are generated.

The steps of a method or algorithm described in connection with an embodiment of the present invention may be implemented directly in hardware, a software module executed by hardware, or a combination thereof. The software modules may include Random Access Memory (RAM), Read Only Memory (ROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), Flash Memory, Hard Disk, Removable Disk, CD-ROM, or It may reside on any type of computer readable recording medium well known in the art.

The embodiments of the present invention have been described above with reference to the accompanying drawings, but a person skilled in the art to which the present invention pertains may implement the present invention in other specific forms without changing its technical spirit or essential features. You will understand. Therefore, it should be understood that the above-described embodiments are illustrative in all respects and not restrictive.

300: dependent parse tree
400: computer system
500: chatbot system

Claims (7)

As a method executed by a computer,
Generating a frequency table using the user's past comments; And
Generating a chatbot's utterance for responding to the user's new utterance using the frequency table;
The step of generating the frequency table,
Generating a first dependent parse tree for one or more sentences of the user's past remarks;
Extracting a predetermined one or more first data sets defined as i) a first object corresponding to a predetermined first vertex, and ii) a second object corresponding to a predetermined second vertex from the first dependent parsing tree. ;
Generating a first table inputting the extracted one or more first data sets;
Calculating a frequency of the one or more first data sets in the first table; And
Generating the frequency table by extracting one or more second data sets having a predetermined frequency and meaningful to the user from the one or more first data sets of the first table,
Generating the frequency table by extracting one or more second data sets having a predetermined frequency and meaningful to the user from the one or more first data sets of the first table,
Generating a third dependent parsing tree for one or more sentences of past remarks of a plurality of different users;
Extracting a predetermined one or more fourth data sets defined as i) a third object corresponding to a predetermined third vertex, and ii) a fourth object corresponding to a predetermined fourth vertex from the third dependent parsing tree. ;
Generating a second table in which the one or more fourth data sets are input;
Calculating a frequency of the one or more fourth data sets in the second table;
Comparing the frequency calculation result of the fourth data set with the frequency calculation result of the first data set to determine whether the one or more first data sets are meaningful to the user; And
If the one or more data sets are determined to be a data set meaningful to a user, the method includes extracting the one or more first data sets to the one or more second data sets meaningful to the user,
Determining whether the at least one first data set is a data set meaningful to the user,
Determining a first data set having a difference between a frequency calculation result of the fourth data set and a frequency calculation result of the first data set equal to or greater than a predetermined reference value or a frequency of a corresponding fourth data set is 0 as a meaningful data set Will
The step of generating a remark of the chatbot,
Generating a second dependent parsing tree for one or more sentences of the user's new utterance;
Extracting a set of one or more third data sets defined as i) a first object corresponding to a predetermined first vertex, and ii) a second object corresponding to a predetermined second vertex from the second dependent parsing tree. ;
Comparing the one or more second data sets and the one or more third data sets in the frequency table;
Extracting the one or more third data sets as a speech candidate data set when the one or more second data sets and the one or more third data sets of the frequency table have a predetermined similarity;
Generating a conversation candidate sentence corresponding to the extracted speech candidate data set using a natural language generation algorithm; And
And generating a chatbot's remark using the dialogue candidate sentence.
How to create a chatbot remark using a frequency table. delete According to claim 1,
The first object or the third object is a noun phrase, and the second object or the fourth object is defined as a verb phrase
How to create a chatbot remark using a frequency table. According to claim 1,
The first object, the second object, the third object or the fourth object,
Derivative elements are transformed into removed headwords through morphological and dictionary analysis
How to create a chatbot remark using a frequency table. According to claim 1,
The first object, the second object, the third object or the fourth object,
When a plurality of words are combined with each other to form a meaning, a plurality of words are processed as one object
How to create a chatbot remark using a frequency table. According to claim 1,
When the one or more second data sets of the frequency table and the one or more third data sets have a predetermined degree of similarity, extracting the one or more third data sets as a speech candidate data set,
Determining the similarity between the one or more second data sets and the one or more third data sets using a Word2Vec technique
How to create a chatbot remark using a frequency table. A computer program stored in a computer-readable recording medium in combination with a computer to execute a method for generating a chatbot utterance using the frequency table of any one of claims 1, 3 to 6.

Images