KR102682790B1 - Computer device and method to create question data - Google Patents

Abstract

A computer device may include at least one processor; and a communicator configured to communicate with the user terminal in response to control of the at least one processor. The processor selects at least one target word from the input text; Obtaining a word embedding vector of a target word from a first database that stores word embedding vectors corresponding to words of a plurality of source sentences, obtained by learning a plurality of source sentences; Searching a first database for a predetermined number of word embedding vectors that are closest to the word embedding vector of the target word in vector space and extracting words from the searched word embedding vectors as first candidate words; extracting at least some of the first candidate words as one or more second candidate words by verifying the association between the first candidate words and the target word by querying a second database storing related words; It is configured to generate problem data corresponding to the input text using the second candidate words.

Description

COMPUTER DEVICE AND METHOD TO CREATE QUESTION DATA}

The present invention relates to computer devices and methods, and more particularly to computer devices and methods for generating problem data.

As the Internet becomes widely used, various websites allow people to learn about a variety of topics. For example, people can access websites through the Internet to download and play multimedia content. Furthermore, websites can provide users with problems related to the information they wish to convey in order to evaluate the degree of learning, receive answers to the problems from the user, and provide the user with the results of the evaluation. These questions are typically created by humans, so the website may additionally provide a user interface that allows the person to enter the questions.

The content described above is only intended to help understand the background technology of the technical ideas of the present invention, and therefore, it cannot be understood as content corresponding to prior art known to those skilled in the art of the present invention.

Embodiments of the present invention are intended to provide a computer device and method that can automatically generate problems appropriate for information intended to be conveyed.

A computer device for communicating with a user terminal according to an embodiment of the present invention includes at least one processor; and a communicator configured to communicate with the user terminal in response to control of the at least one processor, wherein the at least one processor selects at least one target word from the input text; Obtaining a word embedding vector of the target word from a first database that stores word embedding vectors corresponding to words of the plurality of source sentences, obtained by learning a plurality of source sentences; Searching the first database for a predetermined number of word embedding vectors that are closest to the word embedding vector of the target word in a vector space and extracting words from the searched word embedding vectors as first candidate words; extracting at least some of the first candidate words as one or more second candidate words by verifying the association between the first candidate words and the target word by querying a second database storing related words; ; generating problem data corresponding to the input text using the second candidate words; and transmit the problem data to the user terminal through the communicator.

The at least one processor may be configured to determine, among the first candidate words, those that match the target word in the second database as the second candidate words.

The at least one processor may be configured to filter the second candidate words based on words stored in a third database selected by a user and generate the problem data according to the filtered second candidate words.

The input text may be determined according to a request signal received from the user terminal through the communicator.

Another aspect of the present invention relates to a method for communicating with a user terminal. The method includes selecting at least one target word from input text; Obtaining a word embedding vector of the target word from a first database that stores word embedding vectors corresponding to words of the plurality of source sentences, obtained by learning a plurality of source sentences; Searching the first database for a predetermined number of word embedding vectors that are closest to the word embedding vector of the target word in vector space and extracting words from the searched word embedding vectors as first candidate words; Extracting at least some of the first candidate words as one or more second candidate words by verifying the association between each of the first candidate words and the target word by searching a second database storing related words steps; generating problem data corresponding to the input text using the second candidate words; And it may include transmitting the problem data to the user terminal.

In the step of extracting the second candidate words, those that match the target word in the second database among the first candidate words may be determined as the second candidate words.

Generating the problem data may include filtering the second candidate words based on words stored in a third database selected by a user; and generating the problem data according to the filtered second candidate words.

The input text may be determined according to a request signal received from the user terminal through the communicator.

According to embodiments of the present invention, a computer device and method are provided that can automatically generate problems suitable for information to be conveyed. For example, a computer device may first extract candidate words related to a target word based on pre-learned word embedding vectors, and then filter the extracted candidate words by referring to a language database that stores words related to each other. there is. Problems can be generated based on filtered candidate words. Accordingly, the computer device can generate problems suitable for the information intended to be conveyed with high reliability, while saving resources required for problem creation.

1 is a block diagram showing a network system according to an embodiment of the present invention.
FIG. 2 is a block diagram showing a problem generator included in the network server of FIG. 1.
FIG. 3 is a diagram conceptually showing an example of word embedding vectors stored in the word embedding database of FIG. 2.
FIG. 4 is a diagram showing an example of first candidate words extracted by the candidate word extraction unit of FIG. 2.
FIG. 5 is a diagram conceptually showing some of the data stored in the synonym database of FIG. 2.
FIG. 6 is a diagram showing an example of a problem created by the problem generator of FIG. 2.
Figure 7 is a diagram conceptually showing the data sizes of first to third candidate words.
Figure 8 is a flowchart showing a method of operating a network server according to an embodiment of the present invention.
Figure 9 is a flowchart showing an example of step S150 of Figure 8.
FIG. 10 is a block diagram showing an embodiment of a computer device suitable for implementing the network server of FIG. 1.

Hereinafter, a preferred embodiment according to the present invention will be described in detail with reference to the attached drawings. It should be noted that in the following description, only the parts necessary to understand the operation according to the present invention will be described, and the description of other parts will be omitted in order to not obscure the gist of the present invention. Additionally, the present invention is not limited to the embodiments described herein and may be embodied in other forms. However, the embodiments described herein are provided to explain in detail enough to enable those skilled in the art to easily implement the technical idea of the present invention.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case where it is "directly connected," but also the case where it is "indirectly connected" with another element in between. . The terminology used herein is for the purpose of describing specific embodiments and is not intended to limit the invention. Throughout the specification, when a part is said to “include” a certain element, this means that it may further include other elements rather than excluding other elements, unless specifically stated to the contrary. “at least one of X, Y, and Z”, and “at least one selected from the group consisting of It can be interpreted as any combination (e.g., XYZ, XYY, YZ, ZZ). Here, “and/or” includes any combination of one or more of the constituents.

1 is a block diagram showing a network system according to an embodiment of the present invention.

Referring to FIG. 1 , the network system 100 may include a network 50, a user terminal 110, and a network server 120.

Network system 100 may include a plurality of devices, servers, and/or software components operative to perform various methods in accordance with embodiments of the invention described herein. The devices and/or servers shown in FIG. 1 may be configured in different ways, and the operations and services provided by the devices and/or servers may be combined or separated for the embodiments described herein. and may be performed by more or fewer devices and/or servers. One or more devices and/or servers may be operated and/or maintained by the same or different entities.

The network 50 connects components within the network system 100, such as the user terminal 110 and the network server 120. Network 50 may include at least one of a public network, at least one private network, a wired network, a wireless network, another suitable type of network, and combinations thereof. Each of the components in the network system 100 may include at least one of a wired communication function and a wireless communication function, and thus may communicate with each other through the network 50 .

The user terminal 110 is configured to communicate with the network server 120 through the network 50. The user terminal 110 receives content data (CT) from the network server 120 through the network 50 and outputs information corresponding to the received content data (CT) through a display device and/or an audio device. Allows users to learn the information.

The user terminal 110 may receive problem data from the network server 120 and output information (i.e., problem) corresponding to the problem data through a display device and/or an audio device. At this time, the problem data may be associated with content data (CT). Additionally, the user terminal 110 may receive answer data corresponding to the question data from the network server 120 and output it through a display device and/or an audio device.

The user terminal 110 may include hardware, software, and/or firmware suitable for performing the above operations. In embodiments, the user terminal 110 may be a computer device, an Ultra Mobile PC (UMPC), a workstation, a net-book, a Personal Digital Assistant (PDA), a portable computer, a web tablet, Information such as wireless phones, mobile phones, smart phones, e-books, PMPs (portable multimedia players), portable game consoles, etc. are stored in a wired and/or wireless environment. It may include devices capable of transmitting and receiving.

The network server 120 is configured to communicate with the user terminal 110 through the network 50. The network server 120 may provide content data (CT) stored in the database (DB) to the user terminal 110 in response to a request from the user terminal 110. In addition, according to an embodiment of the present invention, the network server 120 automatically generates problem data and answer data corresponding to the content data (CT) in response to a request from the user terminal 110, and generates the problem data may be provided to the user terminal 110, and the correct answer data may be provided to the user terminal 110. This will be explained in more detail with reference to FIG. 3 .

FIG. 2 is a block diagram showing a problem generator included in the network server of FIG. 1. FIG. 3 is a diagram conceptually showing an example of word embedding vectors stored in the word embedding database of FIG. 2. FIG. 4 is a diagram showing an example of first candidate words extracted by the candidate word extraction unit of FIG. 2. FIG. 5 is a diagram conceptually showing some of the data stored in the synonym database of FIG. 2. FIG. 6 is a diagram showing an example of a problem created by the problem generator of FIG. 2.

Referring to FIG. 2, the problem generator 200 includes a target word selection unit 220, a candidate word extraction unit 230, a first filtering unit 240, a second filtering unit 250, and a problem generating unit 260. , and may include a user terminal interface 270.

The target word selection unit 220 is configured to select a target word (TW) from words in the input text (TXT). The input text (TXT) may be determined according to a request signal received from the user terminal 110 through the communicator of the network server 120 (see FIG. 1). For example, one of the sentences included in the content data (CT) may be selected by the user of the user terminal 110, and the selected sentence is sent to the target word selection unit 220 as input text (TXT). can be provided. In addition, the input text (TXT) can be determined through various methods. For example, when the input text (TXT) is “but the price tag says it’s only 19 dollars and 99 cents,” the target word selection unit 220 may select “says” as the target word (TW).

In embodiments, the target word selection unit 220 includes a morpheme analyzer configured to identify morphemes included in the input text (TXT), and a parser configured to identify phrases included in the input text (TXT). can do. Each of the morphological analyzer and syntactic analyzer may operate according to various methods and/or algorithms well known in the art. The target word selection unit 220 may determine a word with relatively high importance (or highest importance) in the input text (TXT) as the target word (TW) according to the identified morphemes and phrases. In other embodiments, the target word (TW) may be selected by the user.

The candidate word extraction unit 230 may receive the target word (TW) from the target word selection unit 220. The candidate word extractor 230 can access the word embedding database DB1.

The word embedding database DB1 stores word embedding vectors corresponding to words of a plurality of source sentences, obtained by performing deep learning on the plurality of source sentences. Through a learning module such as a neural network, words before and after a random word in source sentences can be learned to generate a word embedding vector for the word. For example, if the same and/or similar words appear repeatedly before and after each of the arbitrary words in the source sentences, the arbitrary words may have word embedding vectors adjacent to each other, which may It may mean that they are related to each other (e.g., similar). Referring to FIG. 3, word embedding vectors corresponding to multiple words are located in a vector space (VS), and each word embedding vector has two or more scales (e.g., x-axis and y-axis in FIG. 3). It can be defined by . Within a vector space (VS), words that are related to each other may have dense vectors adjacent to each other, that is, word embedding vectors. In Figure 3, the dense vectors containing the word embedding vector of “woman” and its adjacent word embedding vectors, and the dense vectors containing the word embedding vector of “future” and its adjacent word embedding vectors are indicated by dotted circles. . In this way, words corresponding to adjacent word embedding vectors are related to each other and may be similar to each other. The closer word embedding vectors are to each other, the higher the degree of correlation between the words can be. In an embodiment, word embedding vectors may be generated by learning source sentences using the Sense2Vec model or the Word2Vec model.

Referring again to FIG. 2, the candidate word extractor 230 obtains a word embedding vector corresponding to the target word (TW) from the word embedding database DB1, and extracts the target word ( A predetermined number of word embedding vectors that are closest to the word embedding vector of TW) are searched in the word embedding database (DB1), and the words of the searched word embedding vectors are extracted as first candidate words (CW1). Referring to FIG. 4, “know”, “tell”, “admit”, “realize”, “believe”, etc. are extracted as first candidate words (CW1) related to “say”. For example, five first candidate words (CW1) may be extracted. Depending on the distance from the word embedding vector of “say” in the vector space (VS, see FIG. 3), the degree of relevance of each of the first candidate words (CW1) with “say” may be determined. In Figure 4, “know”, “tell”, “admit”, “realize”, and “believe” have associations with “say” of 88%, 85%, 82%, 74%, and 74%, respectively. .

Referring again to FIG. 2, the first filtering unit 240 receives first candidate words CW1 from the candidate word extracting unit 230. The first filtering unit 240 may access the synonym database DB2. The synonym database (DB2) groups words that are mapped to each other, for example, synonyms that have the same or similar meaning. Referring to FIG. 5, the synonym database (DB2) may include “state,” “tell,” “allege,” “aver,” “write,” etc. as words that are mapped to (e.g., similar to) “say.” You can. In an embodiment, the synonym database DB2 may include a linguistic database such as WordNet. In embodiments, synonym database DB2 may be a general-purpose language database.

Referring to FIG. 2 along with FIGS. 4 and 5, the first filtering unit 240 searches the synonym database DB2 and filters the first candidate words CW1, and filters the first candidate words CW1. ) can be output as at least one second candidate word (CW2). Among the first candidate words (CW1), “know,” “tell,” “admit,” “realize,” and “believe,” the word mapped to “say” in the synonym database (DB2) is “tell.” In this case, the first filtering unit 240 may output “tell” as the second candidate word (CW), excluding “know,” “admit,” “realize,” and “believe.”

The second filtering unit 250 receives one or more second candidate words CW2 from the first filtering unit 240. The second filtering unit 250 can access the curriculum content database DB3. The curriculum content database DB3 may store at least some of the words included in the content data CT. The second filtering unit 250 may verify whether each of the second candidate words CW2 exists in the curriculum content database DB3. The second filtering unit 250 may output those existing in the curriculum content database DB3 among the second candidate words CW2 as the correct answer words AW. In other embodiments, the second candidate word (CW) may be provided as the correct answer word (AW) without going through the second filtering unit 250.

The problem generator 260 is configured to generate problem data based on the input text (TXT), target word (TW), and correct answer word (AW). A variety of methods and/or algorithms known in the art may be used to generate the problem. Referring to Figure 6, based on the input text (TXT) “but the price tag says it's only 19 dollars and 99 cents”, the target word (TW) “says”, and the correct word (AW) “tells” So a problem was created. In embodiments, “knows,” “admits,” “realizes,” and “believes,” which are words excluding the correct word (AW) among the first candidate words (CW1, see FIG. 2), together with “says,” are included in a multiple-choice question. can be provided as

Referring again to FIG. 2 , the user terminal interface 270 may transmit problem data generated by the problem generator 260 to the user terminal 110 through the network 50 (see FIG. 1 ).

In embodiments, a word embedding database (DB1), a synonym database (DB2), and a curriculum content database (DB3) may be included in the database (DB) shown in FIG. 1. In embodiments, the word embedding database DB1, the synonym database DB2, and the curriculum content database DB3 may be provided as components included in the network server 120 (see FIG. 1). In other embodiments, at least some of the word embedding database DB1, the synonym database DB2, and the curriculum content database DB3 may be provided as components separate from the network server 120 and network 50, FIG. Its data can be provided to the network server 120 through (see 1).

Figure 7 is a diagram conceptually showing the data sizes of first to third candidate words.

2 and 7, after the target word (TW) is selected from the input text (TXT), a predetermined number of first candidate words (CW1) are selected using word embedding vectors of the word embedding database (DB1). Second candidate words (CW2) are extracted by first extracting and then filtering using the synonym database (DB2). Accordingly, the second candidate words CW2 may have a smaller data size than the first candidate words CW1. The second candidate words (CW2) can be further filtered (or verified) using the curriculum content database (DB3) to extract the correct answer word (AW).

Assume that candidate words are extracted by first using the synonym database (DB2). In this case, considering that there are many different words similar to a specific word (e.g., target word) in a general-purpose synonym database (DB2) such as WordNet, a relatively large number of candidate words must be extracted. In this case, the resources required to extract the correct word and create a problem may increase due to the large number of candidate words.

On the other hand, according to an embodiment of the present invention, the word embedding database DB1 is preferentially used to extract a predetermined number of first candidate words CW1. Words with a high degree of association (e.g., similarity) with a specific word may vary depending on the learner of the content data (CT, see Figure 1), the type and content of the content data (CT), and the content data depending on the user. Considering that word embedding vectors generated by learning source sentences adaptively selected for (CT) are used to extract first candidate words (CW1) associated with the target word (TW), the first candidate words ( Even though CW1) is limited to a predetermined number, it may contain the correct word (AW) that matches the target word (TW) with a relatively high probability. Additionally, extracting a limited number of first candidate words (CW1) means that resources for problem creation can be reduced.

Afterwards, the second candidate word (CW2) is extracted by filtering the first candidate words (CW1) through the synonym database (DB2), so that the second candidate word (CW2) contains the correct word (AW) with a higher probability. You can. Subsequently, the second candidate word (CW2) can be verified using the curriculum content database (DB3) and provided as the correct answer word (AW).

In this way, the problem generator 200 can extract the correct word (AW) with improved reliability while determining the correct word (AW) with reduced resources.

Figure 8 is a flowchart showing a method of operating a network server according to an embodiment of the present invention.

Referring to FIGS. 1 and 8 , in step S110, the network server 120 selects a target word from the input text. Target words may be selected from the input text using various morphological analysis algorithms and/or parsing algorithms well known in the art.

In step S120, the network server 120 selects a target from the word embedding database (DB1, see FIG. 2) that stores word embedding vectors corresponding to a plurality of words, obtained by performing deep learning on a plurality of source sentences. Obtain the embedding vector of the word.

In step S130, the network server 120 extracts words corresponding to a predetermined number of word embedding vectors that are closest to the embedding vector of the target word as first candidate words from the word embedding database DB1. As described above, first candidate words are extracted using word embedding vectors, so that the first candidate words may include correct words that match the target word with a relatively high probability even if the first candidate words are limited to a predetermined number.

In step S140, the network server 120 searches the synonym database (DB2, see FIG. 2) to verify the degree of association (e.g., similarity) between each of the first candidate words and the target word, thereby selecting one of the first candidate words. Some are extracted as at least one second candidate word. Accordingly, at least one second candidate word may include the correct word with a higher probability.

In step S150, the network server 120 generates problem data corresponding to the input text using the second candidate word. The network server 120 may generate problem data based on the input text, target word, and second candidate word.

In step S160, the network server 120 may transmit problem data to the user terminal 110. Thereafter, the network server 120 may provide a second candidate word (i.e., the correct word) to the user terminal 110 in response to a request from the user terminal 110.

Figure 9 is a flowchart showing an example of step S150 of Figure 8.

Referring to FIGS. 1 and 9 , in step S210, the network server 120 may verify the second candidate word based on the curriculum content database (DB3, see FIG. 2). Subsequently, in step S220, the network server 120 may create a problem using the verified second candidate word (i.e., the correct word).

FIG. 10 is a block diagram showing an embodiment of a computer device suitable for implementing the network server of FIG. 1.

Referring to FIG. 10, the computer device 1000 includes a bus 1100, at least one processor 1200, a system memory 1300, a storage interface 1400, a communication interface 1500, a storage medium 1600, and Includes a communicator 1700.

Bus 1100 is connected to various components of computer device 1000 to transfer data, signals, and information. The processor 1200 may be either a general-purpose or dedicated processor, and may control overall operations of the computer device 1000.

The processor 1200 is configured to load program codes and instructions that provide various functions when executed into the system memory 1300 and process the loaded program codes and instructions. The system memory 1300 may be provided as a working memory for the processor 1200 or the problem creation module 1310 loaded into the system memory 1300. As an embodiment, the system memory 1300 may include at least one of random access memory (RAM), read only memory (ROM), and media readable by other types of computers.

The processor 1200 may load the problem creation module 1310, which provides the functions of the problem generator 200 of FIG. 2 when executed by the processor 1200, into the system memory 1300. Such program codes and/or instructions may be executed by processor 1200 to perform the operations of problem generator 200 described with reference to FIG. 2 . Program codes and/or instructions may be loaded into the system memory 1300 from the storage medium 1600, which is a recording medium readable by a separate computer. Alternatively, program codes and/or instructions may be loaded into the system memory 1300 from outside the computer device 1000 through the communicator 1700. In addition, the processor 1200 loads an operating system 1320 into the system memory 1300 to provide a suitable environment for the problem creation module 1310 to run when executed by the processor 1200, and the loaded operating system (1320) can be executed. The operating system 1320 interfaces between the problem creation module 1310 and components such as the storage interface 1400 and the communication interface 1500 of the computer device 1000 so that the problem creation module 1310 can utilize them. can do.

In FIG. 10 , the system memory 1300 is shown as a separate component from the processor 1200 , but at least a portion of the system memory 1300 may be included in the processor 1200 .

Storage interface 1400 is connected to storage medium 1600. The storage interface 1400 may interface between components such as the processor 1200 and the system memory 1300 connected to the bus 1100 and the storage medium 1600. Communication interface 1500 is connected to communicator 1700. The communication interface 1500 may interface between components connected to the bus 1100 and the communicator 1700. The processor 1200 may communicate with the user terminal 110 (see FIG. 1) connected to the network 50 through the communicator 1700.

In embodiments, bus 1100, processor 1200, and system memory 1300 may be integrated into one board 1050. For example, the bus 1100, processor 1200, and system memory 1300 may be mounted on one semiconductor chip. In embodiments, the board 1050 may further include a storage interface 1400 and a communication interface 1500.

The storage medium 1600 may include various types of non-volatile storage media, such as flash memory and hard disk, that retain stored data even when power is turned off. The storage medium 1600 may include at least a portion of the word embedding database DB1, the synonym database DB2, and the curriculum content database DB3 of FIG. 2 .

The communicator 1700 (or transceiver) may transmit and receive signals between the computer device 1000 and other devices and/or servers in the network system 100 (see FIG. 1) through the network 50.

Although specific embodiments and application examples are described herein, they are provided only to facilitate a more general understanding of the present invention, and the present invention is not limited to the above embodiments and can be understood by those skilled in the art. Various modifications and variations are possible from this substrate.

Accordingly, the spirit of the present invention should not be limited to the described embodiments, and the scope of the patent claims described later as well as all things that are equivalent or equivalent to the scope of this patent claim shall fall within the scope of the spirit of the present invention. .

100: Network system
110: user terminal
120: network server
200: Problem Generator
220: Target word selection unit
230: Candidate word extraction unit
240: first filtering unit
250: second filtering unit
260: Problem creation unit
270: User terminal interface

Claims (8)

In a computer device for communicating with a user terminal:
at least one processor; and
a communicator configured to communicate with the user terminal in response to control of the at least one processor,
The at least one processor,
select at least one target word from the input text;
Obtain a word embedding vector of the target word from a first database that stores word embedding vectors corresponding to words of the plurality of source sentences, obtained by learning a plurality of source sentences;
Searching the first database for a predetermined number of word embedding vectors that are closest to the word embedding vector of the target word in a vector space and extracting words from the searched word embedding vectors as first candidate words;
extracting at least some of the first candidate words as one or more second candidate words by verifying the association between the first candidate words and the target word by querying a second database storing related words; ;
generating problem data corresponding to the input text using the second candidate words;
configured to transmit the problem data to the user terminal through the communicator,
The at least one processor is configured to determine as the second candidate words those that match the target word in the second database among the first candidate words. delete According to claim 1,
The computer device is configured to filter the second candidate words based on words stored in a third database selected by a user, and generate the problem data according to the filtered second candidate words. According to claim 1,
A computer device in which the input text is determined according to a request signal received from the user terminal through the communicator. In a method for communicating with a user terminal:
selecting at least one target word from the input text;
Obtaining a word embedding vector of the target word from a first database that stores word embedding vectors corresponding to words of the plurality of source sentences, obtained by learning a plurality of source sentences;
Searching the first database for a predetermined number of word embedding vectors that are closest to the word embedding vector of the target word in vector space and extracting words from the searched word embedding vectors as first candidate words;
Extracting at least some of the first candidate words as one or more second candidate words by verifying the association between each of the first candidate words and the target word by searching a second database storing related words steps;
generating problem data corresponding to the input text using the second candidate words; and
It includes transmitting the problem data to the user terminal,
In the step of extracting the second candidate words, those that match the target word in the second database among the first candidate words are determined as the second candidate words. delete According to claim 5,
The step of generating the problem data is,
filtering the second candidate words based on words stored in a third database selected by a user; and
and generating the problem data according to the filtered second candidate words. According to claim 5,
A method wherein the input text is determined according to a request signal received from the user terminal through the communicator.

Images