KR20220150122A - Method and apparatus for correcting text - Google Patents

Abstract

According to one embodiment of the present disclosure, disclosed is a contract text correction method performed by a computing device including at least one processor. The method includes the steps of: acquiring input text; predicting a target character based on at least a part of the input text using a first prediction model based on a learned neural network; and changing an object character corresponding to the target character among a plurality of characters included in the input text into the target character.

Description

Text proofing method and apparatus {METHOD AND APPARATUS FOR CORRECTING TEXT}

The present invention relates to a method of correcting text, and more particularly, to a method of correcting text using an artificial neural network.

Text proofreading not only requires correction of typos, but also requires conversion of words themselves according to semantic errors of sentences more broadly.

Recently, along with the development of IT technology, text correction methods using machine learning technology are newly emerging. Currently, in the case of a general machine learning-based text proofing method in the art, a seq2seq model with an encoder for input and a decoder for output is used. However, in this case, there is a problem in that it is not easy to find a typo due to a contextual error, and when a plurality of typos occur in one input, text correction cannot be easily performed due to the mutual influence of the typos.

Therefore, there has been a continuous demand in the art for a more effective text proofing method.

Korean Patent Registration KR10-2236639 discloses "a method for correcting Hangul errors and a system for correcting Hangul errors using syllable-based vectors".

The present disclosure has been devised in response to the background art described above, and an object of the present disclosure is to provide a method for correcting text using an artificial neural network.

Disclosed is a contract text proofreading method performed by a computing device including at least one processor according to an embodiment of the present disclosure for realizing the above-described problems. The method includes: obtaining input text; predicting a target character based on at least a part of the input text using a first prediction model based on the learned neural network; and changing a target character corresponding to the target character among a plurality of characters included in the input text to the target character.

In an alternative embodiment, predicting the target character may be performed based on at least one surrounding character having a position different from the target character included in the input text.

In an alternative embodiment, the first predictive model may be learned based on a vocabulary set comprising at least one training text token consisting of a predetermined number of characters.

In an alternative embodiment, the changing may be performed according to at least one predetermined changing condition determined based on the target text and the target text.

In an alternative embodiment, the predetermined change condition is that, when one of a leading consonant, a middle consonant, or a final consonant in the syllable composition of each of the target character and the target character is inconsistent and the others are identical, the target character is changed to the target character. It may include a first change condition to change.

In an alternative embodiment, the predetermined change condition is that the target text token comprising the target character is not included in the vocabulary set used to train the first predictive model and the target text token comprising the target character When included in the vocabulary set, a second change condition for changing the target character to the target character may be included.

In an alternative embodiment, the method may further include predicting a target text token based on at least a portion of the input text using a second prediction model based on the learned neural network.

In an alternative embodiment, predicting the target text token may be performed based on at least one surrounding text token having a location different from the target text token included in the input text.

In an alternative embodiment, the second predictive model may be trained based on a text corpus comprising at least one training text consisting of a predetermined number of text tokens.

In an alternative embodiment, the changing may be performed based on a prediction result of the first prediction model and a prediction result of the second prediction model.

In an alternative embodiment, the changing may include: a probability value for each of a plurality of candidate characters included in the prediction result of the first prediction model; and a result of an ensemble operation based on a probability value for each of a plurality of candidate text tokens included in the prediction result of the second prediction model.

In an alternative embodiment, the input text is obtained by applying an optical character recognition (OCR) technique to a document image, and the predicting or the changing is a recognition obtained as a result of applying the optical character recognition technique It may be performed when the accuracy of is less than or equal to a predetermined reference accuracy.

Disclosed is a computer program stored in a computer-readable storage medium according to an embodiment of the present disclosure for realizing the above-described problems. The computer program, when executed on one or more processors, causes the computer program to perform the following operations for correcting text, the operations comprising: obtaining input text; predicting a target character based on at least a part of the input text using a first prediction model based on the learned neural network; and changing a target character corresponding to the target character among a plurality of characters included in the input text into the target character.

A text proofing apparatus is disclosed according to an embodiment of the present disclosure for realizing the above-described problems. The apparatus may include one or more processors; Memory; and a network unit, wherein the one or more processors obtain an input text, predict a target character based on at least a portion of the input text, using a first predictive model based on a learned neural network, and A target character corresponding to the target character among a plurality of characters included in the text may be changed to the target character.

The present disclosure may provide a method for correcting text using an artificial neural network.

1 is a block diagram of a computing device for correcting text according to an embodiment of the present disclosure.
2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure;
3 is an exemplary diagram illustrating a result of a text correction method according to an embodiment of the present disclosure.
4 is an exemplary diagram illustrating a target text token prediction method using a second prediction model according to an embodiment of the present disclosure.
5 is a flowchart of a text proofing method according to an embodiment of the present disclosure.
6 is a flowchart of a text correction method according to another embodiment of the present disclosure.
7 is a performance comparison screen illustrating an effect of a text correction method according to an embodiment of the present disclosure.
8 is a simplified, general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented.

Various embodiments are now described with reference to the drawings. In this specification, various descriptions are presented to provide an understanding of the present disclosure. However, it is apparent that these embodiments may be practiced without these specific descriptions.

The terms “component,” “module,” “system,” and the like, as used herein, refer to a computer-related entity, hardware, firmware, software, a combination of software and hardware, or execution of software. For example, a component can be, but is not limited to being, a process running on a processor, a processor, an object, a thread of execution, a program, and/or a computer. For example, both an application running on a computing device and the computing device may be a component. One or more components may reside within a processor and/or thread of execution. A component may be localized within one computer. A component may be distributed between two or more computers. In addition, these components can execute from various computer readable media having various data structures stored therein. Components may communicate via a network such as the Internet with another system, for example via a signal having one or more data packets (eg, data and/or signals from one component interacting with another component in a local system, distributed system, etc.) may communicate via local and/or remote processes depending on the data being transmitted).

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from context, "X employs A or B" is intended to mean one of the natural implicit substitutions. That is, X employs A; X employs B; or when X employs both A and B, "X employs A or B" may apply to either of these cases. It should also be understood that the term “and/or” as used herein refers to and includes all possible combinations of one or more of the listed related items.

Also, the terms "comprises" and/or "comprising" should be understood to mean that the feature and/or element in question is present. However, it should be understood that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, elements and/or groups thereof. Also, unless otherwise specified or unless it is clear from context to refer to a singular form, the singular in the specification and claims should generally be construed to mean “one or more”.

In addition, the term “at least one of A or B” should be interpreted to mean “includes only A”, “includes only B”, and “combined with the configuration of A and B”.

Those skilled in the art will further appreciate that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, computer software, or combinations of both. It should be recognized that they can be implemented with To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

Descriptions of the presented embodiments are provided to enable those skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art of the present disclosure. The generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present invention is not limited to the embodiments presented herein. The invention is to be construed in its widest scope consistent with the principles and novel features presented herein.

1 is a block diagram of a computing device for correcting text according to an embodiment of the present disclosure.

The configuration of the computing device 100 shown in FIG. 1 is only a simplified example. In an embodiment of the present disclosure, the computing device 100 may include other components for performing the computing environment of the computing device 100 , and only some of the disclosed components may configure the computing device 100 .

The computing device 100 may include a processor 110 , a memory 130 , and a network unit 150 .

The processor 110 may include one or more cores, and a central processing unit (CPU) of a computing device, a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU). unit) and the like, and may include a processor for data analysis and deep learning. The processor 110 may read a computer program stored in the memory 130 and perform data processing for text correction of the present disclosure.

According to an embodiment of the present disclosure, the memory 130 may store any type of information generated or determined by the processor 110 and any type of information received by the network unit 150 .

According to an embodiment of the present disclosure, the memory 130 is a flash memory type, a hard disk type, a multimedia card micro type, and a card type memory (eg, For example, SD or XD memory), Random Access Memory (RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read (PROM) -Only Memory), a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium. The computing device 100 may operate in relation to a web storage that performs a storage function of the memory 130 on the Internet. The description of the above-described memory is only an example, and the present disclosure is not limited thereto.

The network unit 150 according to an embodiment of the present disclosure may include any wired/wireless communication network capable of transmitting and receiving any type of data and signals, etc., in the network represented in the present disclosure.

In the present disclosure, the network unit 150 may be configured regardless of its communication mode, such as wired and wireless, and may be configured with various communication networks such as a personal area network (PAN) and a wide area network (WAN). can In addition, the network may be a well-known World Wide Web (WWW), and may use a wireless transmission technology used for short-range communication such as Infrared Data Association (IrDA) or Bluetooth.

The techniques described herein may be used in the networks mentioned above, as well as in other networks.

The configuration of the computing device 100 illustrated in FIG. 1 is merely an example of a simplified configuration of the computing device. In an embodiment of the present disclosure, the computing device 100 may include other components for performing the computing environment of the computing device 100 , and only some of the disclosed components may configure the computing device 100 .

The computing device 100 according to the present disclosure may obtain an input text. In the present disclosure, “input text” may be used interchangeably with “a character string to be corrected for text” according to the present disclosure. The computing device 100 may acquire the input text by receiving the text stored in the memory 130 . The computing device 100 may acquire the input text by receiving the text from the external server through the network unit 150 . The computing device 100 may acquire the input text through an input unit (not shown) included in the computing device 100 .

2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure; The text correction method according to the present disclosure may be performed by a neural network-based prediction model including at least one node.

In the present disclosure, terms such as "neural network-based X model", "neural network-based Y model", etc. mean that an operation is performed on input data by at least some of the nodes included in the neural network, and an operation is performed on the result of the operation. It can be used to refer to a neural network-based model that generates output data based on it. In terms of "neural network-based X model" and "neural network-based Y model", "X" and "Y" may be used to distinguish a structure of a neural network from each other. Throughout this specification, "neural network-based X model" and "neural network-based Y model" may be used interchangeably as "X model" and "Y model", respectively, for short.

In the present disclosure, the terms 'neural network', 'artificial neural network', 'network function', and 'neural network' may be used interchangeably. A neural network may be composed of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. The neural network is configured to include at least one node. Nodes (or neurons) constituting the neural networks may be interconnected by one or more links.

In the neural network, one or more nodes connected through a link may relatively form a relationship between an input node and an output node. The concepts of an input node and an output node are relative, and any node in an output node relationship with respect to one node may be in an input node relationship in a relationship with another node, and vice versa. As described above, an input node-to-output node relationship may be created around a link. One or more output nodes may be connected to one input node through a link, and vice versa.

In the relationship between the input node and the output node connected through one link, the value of the data of the output node may be determined based on data input to the input node. Here, a link interconnecting the input node and the output node may have a weight. The weight may be variable, and may be changed by the user or algorithm in order for the neural network to perform a desired function. For example, when one or more input nodes are interconnected to one output node by respective links, the output node sets values input to input nodes connected to the output node and links corresponding to the respective input nodes. An output node value may be determined based on the weight.

As described above, in a neural network, one or more nodes are interconnected through one or more links to form an input node and an output node relationship in the neural network. The characteristics of the neural network may be determined according to the number of nodes and links in the neural network, the correlation between the nodes and the links, and the value of a weight assigned to each of the links. For example, when the same number of nodes and links exist and there are two neural networks having different weight values of the links, the two neural networks may be recognized as different from each other.

A neural network may include one or more layers. A layer may contain one or more nodes. Some of the nodes constituting the neural network may configure one layer based on distances from the initial input node. For example, a set of nodes having a distance n from the first input node may constitute an n-th layer. The distance from the initial input node may be defined by the minimum number of links that must be traversed to reach the corresponding node from the initial input node. However, the definition of such a layer is arbitrary for description, and the order of the layer in the neural network may be defined in a different way from the above. For example, a layer of nodes may be defined by a distance from the final output node.

The initial input node may mean one or more nodes to which data is directly input without going through a link in a relationship with other nodes among nodes in the neural network. Alternatively, in a relationship between nodes based on a link in a neural network, it may mean nodes that do not have other input nodes connected by a link. Similarly, the final output node may refer to one or more nodes that do not have an output node in relation to other nodes among nodes in the neural network. In addition, the hidden node may mean nodes constituting the neural network other than the first input node and the last output node.

The neural network according to an embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be the same as the number of nodes in the output layer, and the number of nodes decreases and then increases again as the input layer progresses to the hidden layer. can In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be less than the number of nodes in the output layer, and the number of nodes decreases as the number of nodes progresses from the input layer to the hidden layer. have. In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be greater than the number of nodes in the output layer, and the number of nodes increases as the number of nodes progresses from the input layer to the hidden layer. can The neural network according to another embodiment of the present disclosure may be a neural network in a combined form of the aforementioned neural networks.

A neural network according to an embodiment of the present disclosure may include a plurality of neural network layers. The plurality of neural network layers may constitute a sequence having a predetermined order in the neural network according to their functions and roles.

The neural network layer according to an embodiment of the present disclosure may include at least one node. A weight or a bias may be assigned to each node included in the neural network layer. The memory 130 of the computing device 100 according to the present disclosure may store a weight or bias value assigned to at least one node included in the neural network layer. Each neural network layer may receive as an input the first input to the convolutional neural network or the output of the previous neural network layer. For example, in a sequence including a plurality of neural network layers, the N-th neural network layer may receive the output of the N-1 th neural network layer as an input. Each neural network layer can generate an output from its input. When the neural network layer is the highest and last neural network layer in the sequence, the output of this neural network layer can be treated as the output of the entire neural network.

A deep neural network (DNN) may refer to a neural network including a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks can be used to identify the latent structures of data. In other words, it can identify the potential structure of photos, texts, videos, voices, and music (e.g., what objects are in the photos, what the text and emotions are, what the texts and emotions are, etc.) . Deep neural networks include convolutional neural networks (CNNs), recurrent neural networks (RNNs), auto encoders, generative adversarial networks (GANs), and restricted boltzmann machines (RBMs). machine), a deep belief network (DBN), a Q network, a U network, a Siamese network, and a Generative Adversarial Network (GAN). The description of the deep neural network described above is only an example, and the present disclosure is not limited thereto.

In an embodiment of the present disclosure, the network function may include an autoencoder. The auto-encoder may be a kind of artificial neural network for outputting output data similar to input data. The auto encoder may include at least one hidden layer, and an odd number of hidden layers may be disposed between the input/output layers. The number of nodes in each layer may be reduced from the number of nodes in the input layer to an intermediate layer called the bottleneck layer (encoding), and then expanded symmetrically with reduction from the bottleneck layer to the output layer (symmetrical to the input layer). Autoencoders can perform non-linear dimensionality reduction. The number of input layers and output layers may correspond to a dimension after preprocessing the input data. In the auto-encoder structure, the number of nodes of the hidden layer included in the encoder may have a structure that decreases as the distance from the input layer increases. If the number of nodes in the bottleneck layer (the layer with the fewest nodes between the encoder and decoder) is too small, a sufficient amount of information may not be conveyed, so a certain number or more (e.g., more than half of the input layer, etc.) ) may be maintained.

The neural network may be trained using at least one of supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. Learning of the neural network may be a process of applying knowledge for the neural network to perform a specific operation to the neural network.

A neural network can be trained in a way that minimizes output errors. In the training of a neural network, iteratively input the training data into the neural network, calculate the output of the neural network and the target error for the training data, and calculate the error of the neural network from the output layer of the neural network to the input layer in the direction to reduce the error. It is the process of updating the weight of each node of the neural network by backpropagation in the direction. In the case of teacher learning, learning data in which the correct answer is labeled in each learning data is used (ie, labeled learning data), and in the case of comparative learning, the correct answer may not be labeled in each learning data. That is, for example, the learning data in the case of teacher learning regarding data classification may be data in which categories are labeled for each of the learning data. Labeled training data is input to the neural network, and an error can be calculated by comparing the output (category) of the neural network with the label of the training data. As another example, in the case of comparison learning about data classification, an error may be calculated by comparing the input training data with the output of the neural network. The calculated error is back propagated in the reverse direction (ie, from the output layer to the input layer) in the neural network, and the connection weight of each node of each layer of the neural network may be updated according to the back propagation. A change amount of the connection weight of each node to be updated may be determined according to a learning rate. The computation of the neural network on the input data and the backpropagation of errors can constitute a learning cycle (epoch). The learning rate may be applied differently depending on the number of repetitions of the learning cycle of the neural network. For example, in the early stage of learning of a neural network, a high learning rate can be used to enable the neural network to quickly acquire a certain level of performance, thereby increasing efficiency, and using a low learning rate at the end of learning can increase accuracy.

In training of a neural network, in general, the training data may be a subset of real data (that is, data to be processed using the trained neural network), and thus the error on the training data is reduced, but the error on the real data is reduced. There may be increasing learning cycles. Overfitting is a phenomenon in which errors on actual data increase by over-learning on training data as described above. For example, a phenomenon in which a neural network that has learned a cat by seeing a yellow cat does not recognize that it is a cat when it sees a cat other than yellow may be a type of overfitting. Overfitting can act as a cause of increasing errors in machine learning algorithms. In order to prevent such overfitting, various optimization methods can be used. In order to prevent overfitting, methods such as increasing the training data, regularization, and dropout that deactivate some of the nodes of the network in the process of learning, and the use of a batch normalization layer are applied. can

The computing device 100 according to the present disclosure may predict a target character based on at least a part of the input text using the first prediction model based on the learned neural network. In the present disclosure, terms such as “first” and “second” are used to distinguish one component from other components, and are used only to maintain consistency of the referenced object throughout the specification. The scope of rights should not be limited. Accordingly, if necessary, the “first predictive model” may be changed to “the second predictive model” and the “second predictive model” may be changed to “the first predictive model” throughout the specification.

The computing device 100 may calculate a prediction result regarding the target character by inputting at least a portion of the input text into the first prediction model.

In the present disclosure, the input data input to the first predictive model may be a 'character embedding vector', which is vector expression data about a character. The character 　 embedding vector means a vector calculated according to an arbitrary method for expressing a character as a vector. In a specific embodiment, any method for representing a character as a vector may be a sparse representation method. The sparse representation method may include a vector representation method of a one-hot encoding method. In a one-hot vector, an index value of a corresponding character may be 1, and the remaining index values may be expressed as 0. The size of the one-hot vector may be equal to the total number of representable characters. In a sparse representation, most of the values of elements of a vector or matrix can be expressed as 0. In another embodiment, any method for representing a character as a vector may be a dense representation method. In the dense expression method, element values of vectors for characters may have real values. The dense representation method may include a neural network-based vector operation method. A neural network-based vector operation method is a method of inputting a one-hot encoding vector for at least one character to an artificial neural network model including at least one layer, and then outputting a one-hot encoding vector of a specific character through operation. includes The dense expression method includes a method of determining an intermediate vector calculated by an artificial neural network model including at least one layer during a calculation process as an embedding vector for an input character. As an example, the dense expression method may include word2vec, FastText, Glove, and the like. The detailed description of the character embedding vector described above is merely illustrative of various embodiments of expressing a character as a vector, and the present disclosure includes, without limitation, any method for expressing a character as a vector. In the present disclosure, input of at least one character to the first predictive model should be interpreted as input of at least one character embedding vector, which is vector representation data about a character, into the first predictive model.

The computing device 100 according to the present disclosure may predict the target character based on at least one surrounding character having a position different from the target character included in the input text. In the present disclosure, the term “target character” may be used interchangeably with terms such as “corrected character”, “corrected character”, “correction result character” or “correction result character”. In the present disclosure, “target character” may be used interchangeably with “inspection target character”. The computing device 100 according to the present disclosure may calculate a target character by examining at least one target character included in the input text, and change the target character to the target character. The computing device 100 may calculate a target character corresponding to the target character based on surrounding characters excluding the target character. Specifically, in the present disclosure, the target character and the surrounding characters may be distinguished based on the position of the character included in the input text. The surrounding character may be a character having a position different from that of the target character among a plurality of characters included in the input text. For example, suppose there is an input text "St. Mary's Hospital". In this case, when "surname" is the target character, the surrounding characters may be 'mo', 'byeong', and 'won'. Also, when “circle” is a target character, the surrounding characters may be 'surname', 'mo', and 'byeong'. The first prediction model according to the present disclosure may predict the target character corresponding to the position of the target character based on at least some of the surrounding characters excluding the target character. The computing device 100 according to the present disclosure may sequentially move from the very beginning of the input text to examine the input text and select one character as the character to be tested. The computing device 100 may select a target character to be examined in the input text based on an arbitrary algorithm. The computing device may select the target character based on the recognition accuracy of the optical character recognition technology. This will be described later in detail.

Hereinafter, some embodiments according to the present disclosure for changing the target character included in the input text to the target character will be described in detail with reference to FIG. 3 .

3 is an exemplary diagram illustrating a result of a text correction method according to an embodiment of the present disclosure. Reference numeral 311 denotes the target character “bell” having the position of the first syllable in the input text “jonggyobi”. Reference numeral 313 denotes the target character “gun” corresponding to the target character “bell”. The computing device 100 according to the present disclosure receives the input text and receives at least one of the surrounding characters “Jin”, “Ryo”, and “Bi” which are surrounding characters having a different position from the target character “Jong” 311 Based on , the target character “total” 313 may be calculated. As a result, the input text of 'end of medical care' may be corrected as "total medical expenses" by the text correction method of the present disclosure. As another example, reference numeral 331 denotes “M” as the target character having the position of the fourth syllable in the input text “Cash Sunam Geum”. Reference number 333 indicates the target character “lead” corresponding to the target character “male”. The computing device 100 according to the present disclosure receives the input text and receives at least one of the surrounding characters “hyun,” “gold,” “number,” and “gold,” which are surrounding characters having different positions from the target character “male” 331 . A target character “lead” 333 may be calculated based on one surrounding character. As a result, the input text marked as “cash receipts” may be corrected to “cash receipts”.

The first predictive model according to the present disclosure may be learned based on a vocabulary set including at least one training text token. In the present disclosure, a “text token” refers to a natural language processing unit used for natural language processing. Generating text tokens for the input text may mean segmenting/tokenizing the input text into arbitrary units. The text token may be generated from the input text according to any predetermined criteria. In the present disclosure, a text token means a natural language processing unit including two or more characters. The computing device 100 according to the present disclosure may perform text inspection in a more comprehensive range than individual characters by using a natural language processing unit including two or more characters. The “learning text token” is a text token used for learning, and may be a text token having a predetermined number of characters (N, N being a natural number equal to or greater than 2). For example, when N is 2, the vocabulary set may include training text tokens such as “resident”, “resident”, “number”, and the like. As another example, when N is 3, the vocabulary set may include training text tokens such as “ultrasound”, “sonic wave value”, and “wave therapy”. The vocabulary set may include two or more training text tokens with different character counts.

In an embodiment of the present disclosure, a vocabulary set used to train the first predictive model may be generated based on an operation of tokenizing an arbitrary sentence by a predetermined number of characters. For example, given a sentence such as “unpaid ultrasound therapy” and the predetermined number for the tokenization task is 3, the vocabulary set is “non-pay”, “salary_“, “female_second”, “_ultrasound” , “ultrasound”, “sonic wave”, and “wave therapy” may include training text tokens. As in the example above, the learning text token included in the vocabulary set can be treated as a character by replacing the space in the sentence with a single character such as “_”.

The first prediction model according to the present disclosure predicts a target character based on at least one surrounding character having a different position from the target character included in the training text token, and may be learned by comparing the predicted target character and the target character. have. That is, in the learning process of the first predictive model, the first predictive model is trained to predict what the target character is by using surrounding characters having different positions from the target character, and the target character included in the training text token is the first predictive model. It can be used as correct answer data in the learning process of In the learning process, the computing device 100 of the present disclosure may calculate a prediction result as an output after inputting at least one surrounding character into the first prediction model. The position of the surrounding characters or the number of surrounding characters input to the first predictive model may be determined according to a predetermined setting. Each character input to the first prediction model may be data expressed as a vector as described above. Hereinafter, a learning method of the first predictive model and an output form of the first predictive model will be described.

To describe some embodiments of the first predictive model learning method according to the present disclosure, it is assumed that the learning text token is “Catholic University”. When the target character for learning is 'Rick' and the window size is 5, the first prediction model uses two characters (i.e. 'A', 'Toll', 'Large', 'Learning') can be input and learned. If the learning target character is 'A' and the window size is 5, 'A' is the starting character of the training text token, so there are no other characters on the left. In this case, the first predictive model may be learned by inputting four characters (i.e. 'toll', 'rick', 'dae', and 'hak') existing to the right of 'a'. In addition, when the learning target character is 'gyo' and the window size is 3, 'gyo' is the last character of the training text token, so there are no other characters on the right side. In this case, the first predictive model may be learned by inputting two characters (i.e. 'dae', 'hak') existing to the left of 'gyo'. The detailed description of the above-described learning text token and window size is only an example for description and does not limit the present disclosure. As described above, the first predictive model according to the present disclosure can learn which characters each individual character exist in the vicinity of together based on the training text token included in the vocabulary set. This has the effect of improving the text proofing accuracy at the character level.

The computing device 100 according to the present disclosure may change the target text corresponding to the target text to the target text according to at least one predetermined change condition determined based on the target text and the target text.

The predetermined change condition according to an embodiment of the present disclosure is “If one of the leading, middle, or final consonants in the syllable composition of the target character and the target character does not match and the others match, the target character is changed to the target character” The first change condition may be included. As an example, if the target character to be tested in the input text is a 'star' and the target character predicted based on the surrounding characters around the target character is 'ill', the target character and the target character have a different finality, but , , and neutrals match, so that the first change condition is satisfied, and the computing device 100 may change the target character to the target character. In another embodiment, when the target character to be inspected is 'chi' and the target character predicted based on the surrounding characters around the target character is 'in', the target character and the target character have different initials and neutrals. Therefore, the first change condition may not be satisfied.

When a target character is changed to a target character according to the first change condition described in the present disclosure, instead of simply changing all the characters according to the prediction result of the first prediction model, a rule-based text correction method can be additionally introduced. there is That is, by introducing the rule-based correction method serving as an upper limit on the degree of change as the first change condition, excessive deformation of the input text can be prevented. For example, in the respective texts of “medical fee” and “medical fee”, the middle character is 'Ryo' and 'Cal', but both texts may be texts that do not require correction. In this case, the first predictive model may calculate a dominant probability value for only one of 'Ryo' or 'Chal' when 'Jin' and 'Bi' are input as surrounding words according to the vocabulary set used during learning. Accordingly, the text correction method according to the present disclosure discloses a first change condition serving as a kind of 'limitation of typographical recognition'. When the initial consonant, middle, and final consonants of the target character included in the text written by the user and the target character output by the first prediction model all do not match, there is a possibility that it is a new expression rather than a typo. Accordingly, the present disclosure may change the target character to the target character when only one of the initial consonant, the middle consonant, and the final consonant is different and the rest match through the first change condition.

The predetermined change condition according to another embodiment of the present disclosure is: “The target text token including the target character is not included in the vocabulary set used to train the first predictive model and the target text token including the target character When included in the vocabulary set used to train the first predictive model, it may include a second change condition of “change the target character to the target character”. The “target text token” is a text token including a target character to be examined by the computing device 100, and the “target text token” is a text token that is predicted for the target character by the first prediction model. A text token containing the predicted target character. For example, in the input text, a target character to be examined may be 'Jo,' and a target text token including the target character may be 'Articulation wave'. Also, the target character output by the first predictive model may be 'seconds', and the target text token including the target character may be 'ultrasound'. In this case, the target text token 'articulation wave' may be a token that does not exist in the vocabulary set used to train the first prediction model. Conversely, 'ultrasound' can be a token present in a vocabulary set. In this case, since the second change condition is satisfied, the computing device 100 may change the target character to the target character.

When a target character is changed to a target character according to the second change condition described in the present disclosure, token unit verification is possible. Since the learned first prediction model predicts the target character through the operation result using a plurality of internal parameters, the causal relationship cannot be accurately known. On the other hand, the text token composition of the vocabulary set used when the first prediction model is trained is directly managed by the user. Therefore, even if the first prediction model predicts the target character, whether the word is actually used can be known through the vocabulary set. That is, the second change condition according to the present disclosure is not to simply use the result of the learned artificial neural network-based prediction model in the correction of characters, but to check once whether the corresponding word is included in the vocabulary set used when the model is trained. By further verification, more accurate and sophisticated text proofing work can be performed.

The text correction method according to the present disclosure may include an additional step of using a second prediction model based on the learned neural network. The computing device 100 may predict the target text token based on at least a portion of the input text using the second prediction model based on the learned neural network. The computing device 100 may predict the target text token based on at least one surrounding text token having a location different from the target text token included in the input text.

In the present disclosure, the term “target text token” refers to “text token comprising redacted characters”, “text token comprising corrected characters”, “text token comprising correction result characters” or “correction result characters”. It may be used interchangeably with terms such as “including text token”. In the present disclosure, the target text token may be determined as 'a text token including a target character to be examined'.

When examining at least one target character included in the input text, the computing device 100 according to the present disclosure determines a target text token including the target character, and based on surrounding text tokens excluding the target text token, the target text Tokens can be generated. A method of changing the target text to the target text based on the target text token will be described later in detail. In the present disclosure, the target text token and the surrounding text token may be distinguished based on the position of the text token included in the input text. The surrounding text token may be a text token having a position different from that of the target text token among a plurality of text tokens included in the input text. For example, suppose there is an input text "Comprehensive health insurance that covers even cancer". At this time, the text input by the computing device 100 according to a predetermined setting is a plurality of text tokens (e.g. 'up to cancer', '_guarantee', 'do_', 'comprehensive_', 'health_', ' insurance'). For example, when the target character to be examined by the computing device 100 is 'cancer', the computing device 100 may determine 'until cancer' as the target text token. The computing device 100 may determine at least one of '_guarantee', 'do_', 'comprehensive_', 'health_', and 'insurance' as the surrounding text token. As another example, when the target character to be examined by the computing device 100 is 'sum', the computing device 100 may determine 'synthesis_' as the target text token. The computing device 100 may determine at least one of 'up to cancer', 'guarantee', 'do_', 'health_', and 'insurance' as the surrounding text token.

The second prediction model according to the present disclosure may predict the target text token corresponding to the location of the target text token based on at least some of the surrounding text tokens except for the target text token. The second prediction model according to the present disclosure may receive at least one surrounding text token and predict a target text token.

In the present disclosure, input data input to the second predictive model may be a 'token embedding vector' that is vector representation data about a text token. The token embedding vector may be calculated in the same or similar manner as the aforementioned character embedding vector. In the present disclosure, input of at least one text token to the second predictive model should be interpreted as input of at least one token embedding vector that is vector representation data about the text token into the second predictive model.

The second predictive model according to the present disclosure may be trained based on a text corpus including at least one training text. In the present disclosure, the “learning text” may be a character string having an arbitrary length. The “training text” may be a character string composed of a predetermined number of text tokens (M and M are natural numbers). Each text token may again be composed of a predetermined number of characters (N and N are natural numbers). For example, if the predetermined number of text tokens (M) is 5 and the predetermined number of characters (N) constituting each text token is 5, the text corpus contains ['mild dementia_', 'severe dementia_', 'all Learning texts such as 'guarantee', 'be_insured', 'only here'] may be included. As another example, when the predetermined number of text tokens (M) is 5 and the predetermined number of characters (N) constituting each text token is 3, the text corpus includes ['direct_', 'subscribed', '_customer ', 'this_', 'recommended'] may be included in the training text. The computing device 100 may collect arbitrary sentences and generate various text corpuses according to the number of arbitrary text tokens constituting the training text and the number of arbitrary characters constituting the text token.

The second prediction model according to the present disclosure predicts a target text token based on at least one surrounding text token having a different location from the target text token included in the training text, and compares the predicted target text token with the target text token. can be learned through That is, in the learning process of the second prediction model, the second prediction model is trained to predict the target text token using surrounding text tokens except for the target text token, and the target text token included in the training text is the second prediction model of the second prediction model. It can be used as correct answer data in the learning process. In the learning process, the computing device 100 of the present disclosure may calculate a prediction result as an output after inputting at least one surrounding text token into the second prediction model. The position of the surrounding text tokens or the number of surrounding text tokens input to the second prediction model may be determined according to a predetermined setting. Each text token input to the second prediction model may be data expressed as a vector as described above.

4 is an exemplary diagram illustrating a target text token prediction method using a second prediction model according to an embodiment of the present disclosure. Reference numeral 400 designates a second prediction model. The second prediction model 400 receives ['I', <M>, 'going', <M>, 'work'] as input data and ['I', 'am', 'going'] as output data. , 'to', 'work'] can be output. <M> of the input data corresponds to a hidden text token and means a text token to be predicted by the second prediction model 400 . In the embodiment of FIG. 4 , the second prediction model 400 receives 'I' and 'going', which are surrounding text tokens, as input to predict the target text token located first from the left among the text tokens indicated by <M>. You can print the target text token 'am'. In addition, the second prediction model 400 receives the surrounding text tokens 'going' and 'work' to predict the target text token located second from the left among the text tokens marked with <M>, and the target text 'to' You can print tokens. The detailed description of the learning text described with reference to FIG. 4 is only an example for description and does not limit the present disclosure. As such, the second predictive model according to the present disclosure can learn which text tokens each individual text tokens exist in the vicinity of based on the training text included in the text corpus. This has the effect of improving the accuracy of text proofreading at the text token level.

Hereinafter, an embodiment in which the target character corresponding to the test target in the input text is finally changed to the target character based on the target text token calculated by the second prediction model according to the present disclosure will be described.

The computing device 100 according to the present disclosure may change the target character to the target character based on the prediction result of the first prediction model and the prediction result of the second prediction model.

In the present disclosure, the term “prediction result of the first prediction model” may be used interchangeably with “output of the first prediction model with respect to the input”. The prediction result of the first prediction model may include a plurality of probability values calculated for each of the plurality of candidate characters. The plurality of probability values may be prediction confidence scores for each of the plurality of candidate characters. The first prediction model may be learned in such a way that, among the probability values corresponding to the plurality of candidate characters included in the prediction result, the probability values for the same candidate characters as the correct answer characters increase and the probability values for the remaining candidate characters decrease. The computing device 100 may determine a candidate character corresponding to the largest value among the probability values included in the prediction result of the first prediction model as the target character. When the probability value included in the prediction result of the first prediction model is equal to or greater than a predetermined threshold value, the computing device 100 may determine a candidate character corresponding to the corresponding probability value as the target character. Here, a plurality of candidate character groups to be subjected to reliability calculation by the first prediction model may be preset. The plurality of preset candidate character groups may be a set of individual characters constituting at least one training text token included in the aforementioned vocabulary set.

Meanwhile, in the present disclosure, the term “prediction result of the second prediction model” may be used interchangeably with “output of the second prediction model with respect to the input”. The prediction result of the second prediction model may include a plurality of probability values calculated for each of the plurality of candidate text tokens. The plurality of probability values may be prediction confidence scores for each of the plurality of candidate text tokens. The second prediction model may be learned in such a way that, among the probability values corresponding to the plurality of candidate text tokens included in the prediction result, the probability value for the same candidate text token as the correct text token is increased and the probability value for the remaining candidate text tokens is decreased. . The computing device 100 may determine a candidate text token corresponding to the largest value among probability values included in the prediction result of the second prediction model as the target text token. When the probability value included in the prediction result of the second prediction model is equal to or greater than a predetermined threshold, the computing device 100 may determine a candidate text token corresponding to the corresponding probability value as the target text token. Here, a plurality of candidate text token groups to be subjected to reliability calculation by the second prediction model may be preset. The plurality of preset candidate text token groups may be a set of text tokens constituting at least one training text included in the aforementioned text corpus.

The computing device 100 according to the present disclosure may input at least a portion of the input text into the learned first predictive model, and then predict the target character based on the prediction result of the first predictive model.

In the first embodiment related to the first predictive model, when the input text of “elbow seat and tension” is given to the computing device 100 and 'Um' corresponds to the character to be tested, the first predictive model is ' Receive at least one of the remaining characters except for 'Um' as the surrounding character and receive [('Yeom': 0.99999), ('Tong': 2.79597e-06), ('Reverse': 3.49937e-08), ('G ': 2.02706e-08)] may be calculated. The computing device 100 may determine, as the target character, 'salt', which is a character having the highest corresponding probability value, among a plurality of characters included in the prediction result of the first prediction model.

The computing device 100 according to the present disclosure may predict a target text token based on a prediction result of the second prediction model after inputting at least a portion of the input text into the learned second prediction model.

In the second embodiment of the second predictive model, when an input text of “elbow seat and tension” is given to the computing device 100 and 'um' corresponds to the character to be tested, ['elbow', ' An input text expressed as _compassion', '_and_', 'tension.'] may be input to the second predictive model. The second predictive model is the surrounding text tokens (i.e. 'elbow', '_and_', 'tension.') excluding the target text token (i.e. '_Umjwa') including the character to be tested (i.e. 'Um'). A prediction result can be calculated based on . The prediction result of the second prediction model may include a plurality of candidate text tokens and probability values for each candidate text token. The prediction result can be calculated as follows. [('_Strain': 0.4), ('_Pain': 0.3), ('_Fracture': 2.30256e-07), ('_Dislocation': 1.02706e-08)] can The computing device 100 may determine, as the target text token, '_sprain', which is a candidate text token having the highest corresponding probability value among a plurality of candidate text tokens included in the prediction result of the second prediction model.

The computing device 100 according to the present disclosure provides a probability value for each of a plurality of candidate characters included in the prediction result of the first prediction model and a probability value for each of a plurality of candidate text tokens included in the prediction result of the second prediction model It is possible to change the target character to the target character according to the result of the operation based on . The computing device 100 may perform an ensemble operation using the prediction result of the first prediction model and the prediction result of the second prediction model, and change the target character to the target character according to the result. For example, the ensemble operation may include a weighted sum operation and an average operation on the probability values included in the prediction result.

In the third embodiment of the present disclosure relating to the operation of changing the target character to the target character according to the result of the ensemble operation, the first embodiment of the first predictive model and the second embodiment of the second predictive model are described. It will be described in detail with reference again. When “elbow seat and tension” is given as input text to the computing device 100 according to the present disclosure and the target character is 'um', referring back to the first embodiment, the first predictive model is [('salt') : 0.99999), ('Tong': 2.79597e-06), ('Reverse': 3.49937e-08), ('G': 2.02706e-08)] can be calculated. Likewise, referring back to the second embodiment, the second prediction model is [('_sprain': 0.4), ('_pain': 0.3), ('_fracture': 2.30256e-07), ('_dislocation ': 1.02706e-08)] can be calculated. In this case, the computing device 100 may generate at least one text token by substituting at least one candidate character among a plurality of candidate characters included in the prediction result of the first prediction model. According to the above embodiment, text tokens such as '_sprain', '_contract', '_reverse', '_right' may be generated. In this case, the same probability included in the prediction result of the first prediction model may be given to the text tokens generated by the substitution operation. After that, the computing device 100 includes text tokens (e.g. '_sprain', '_contract', '_reverse seat', '_region') generated by the substitution operation and the prediction result of the second prediction model. A target character may be changed to a target character according to a weighted sum operation on the probabilities of each of the candidate text tokens (e.g. '_sprain', '_pain', '_fracture', '_dislocation'). As an example of the weighted sum operation, it is assumed that a weight of 0.8 is set for the prediction result of the first prediction model and a weight of 0.2 is set with respect to the prediction result of the second prediction model. At this time, '_sprain' may have a weighted probability value of “0.8*0.99999 + 0.2*0.4 = 0.879992”, and '_sprain' has a weight of “0.8*2.79597e-06 + 0.2*0 = 2.36776*e-06” It can have a probability value. Also, '_pain' may have a weighted probability value of “0.8*0 + 0.2*0.3 = 0.06”. The same weighted sum operation may be performed on the remaining text tokens. As a result of the ensemble operation according to the third embodiment, the computing device 100 may change the target character 'um' of the input text to 'yeom' according to '_sprain' having the highest probability value.

A fourth embodiment of the present disclosure relating to an operation of changing a target character to a target character according to an ensemble operation result describes a case in which the prediction result of the second prediction model has a higher weighted probability value. When "elbow thumb and tension" is given as input text to the computing device 100 according to the present disclosure and the target character is 'um', the prediction result of the first prediction model is [('ring': 0.0359), ( 'Cube': 0.0266), ('Bone': 0.0219), ('Upper': 0.0215)]. That is, on the prediction result of the first prediction model, it may be in a state in which any one character does not have a predominantly dominant probability. This state is, for example, when the first predictive model is not fully trained, when data different from the training data is input, or a character (e.g. 'character) following the target character (e.g. 'um') as in this embodiment. ') Also, due to a typo, it may appear when the difficulty of predicting the target character based on the surrounding characters is high. At this time, the prediction results for the target text token (e.g. '_strain') of the second prediction model are [('_sprain': 0.4), ('_pain': 0.3), ('_fracture': 2.30256e- 07), ('_dislocation': 1.02706e-08)]. In this case, the computing device 100 may perform a weighted sum operation on the respective probabilities included in the prediction result of the first prediction model and the prediction result of the second prediction model. It is assumed that the weights for the prediction results of each prediction model are set to be the same as in the third embodiment. Here, the weighted probability value for '_patient', which is one of the text tokens generated by the substitution operation, may be “0.8*0.0359 + 0.2*0 = 0.02872”. On the other hand, the weighted probability value for '_sprain', which is one of the candidate text tokens included in the prediction result of the second prediction model, may be “0.8*0 + 0.2*0.4 = 0.08”. It is obvious that the same weighted sum operation may be performed on the remaining text tokens. As a result of the ensemble operation according to the fourth embodiment, the computing device 100 may change the target character 'um' of the input text to 'yeom' according to '_sprain' having the highest probability value. Also, when the target text is corrected based on the target text token of the second prediction model, the computing device 100 may also change other surrounding characters included in the target text token according to the target text token. In this case, the computing device 100 may correct the input text of “elbow thumb and strain” to “elbow sprain and strain”.

The above-described third and fourth embodiments are exemplary descriptions for specific description of a text correction method for changing a target character to a target character based on the prediction result of the first prediction model and the prediction result of the second prediction model to be. The present disclosure includes, without limitation, various ensemble methods for changing a target character to a target character based on the prediction result of the first prediction model and the prediction result of the second prediction model, including the above-mentioned contents. A first prediction model of the present disclosure is a model for predicting a target character based on surrounding characters of a target character in character units, and enables accurate correction of typos. The second prediction model of the present disclosure predicts a target text token based on the surrounding text token of the surrounding target text token in units of text tokens, and thus enables correction in consideration of the context. Accordingly, the present disclosure discloses a method of correcting text by appropriately ensembles the prediction results of the first prediction model and the second prediction model, taking into account the physical error of the written character (e.g. error due to typo) and the context and content. Disclosed is a method capable of correcting texts taking into account all semantic errors (e.g. errors in word selection).

The text correction method according to the present disclosure may be performed when the recognition accuracy obtained as a result of the application of the optical character recognition technology is less than or equal to a predetermined reference accuracy. In the present disclosure, optical character recognition (OCR) technology refers to a technology that recognizes characters, symbols, marks, etc. printed or handwritten on paper or the like through optical means (e.g. camera, scanner, etc.) and converts them into computer text. . In other words, OCR technology is a technology that converts analog text into digital text. It is a technology that converts characters included in an image created by an optical device such as a scanner or camera into characters that can be edited by a digital device such as a computer. it means. The computing device 100 may acquire the input text by applying an optical character recognition (OCR) technology to the document image. The document image to which the optical character recognition technology is applied may be obtained through an input unit included in the computing device 100 or may be obtained through the network unit 150 . The computing device 100 may compare the recognition accuracy according to the optical character recognition technology with a predetermined reference accuracy, and perform the text correction method when the recognition accuracy is equal to or less than the predetermined reference accuracy. Specifically, the recognition accuracy by the optical character recognition technology may be obtained as a natural number such as “70” or “80”, and the predetermined reference accuracy is also a natural number for comparison with the recognition accuracy as “75” or “90”, etc. can The computing device may select the target character based on the recognition accuracy of the optical character recognition technology. In this case, the recognition accuracy of the optical character recognition technology may be calculated in units of individual characters. The text correction method performed when the recognition accuracy is less than or equal to a predetermined reference accuracy may include predicting a target character using a first prediction model and changing the target character to the target character. Since the specific text correction method has been described in detail above, overlapping content will be omitted.

5 is a flowchart of a text proofing method according to an embodiment of the present disclosure. The steps shown in FIG. 5 are exemplary, and additional steps may also be included within the scope of the present disclosure. To start the text proofing method, the computing device 100 may acquire the input text ( S510 ). The computing device 100 may predict the target character based on at least a part of the input text using the first prediction model based on the learned neural network ( S530 ). The computing device 100 may input surrounding characters excluding the target character to be examined into the first prediction model and predict the target character. The number of surrounding characters input to the first prediction model may be predetermined as an arbitrary natural number. The first prediction model may calculate probability values for a plurality of candidate characters as a prediction result. The computing device 100 may determine a candidate character having the largest probability value from the prediction result of the first prediction model as the target character. The computing device 100 may change the target character corresponding to the target character among a plurality of characters included in the input text to the target character ( S550 ). The computing device 100 determines the target character based on the target character and the target character. The target character may be changed to the target character in response to at least one predetermined changing condition. The predetermined changing condition compares syllables of the target character and the target character, and when one of the leading, middle, or final consonants does not match and the others match, the first changing condition changes the target character to the target character and the target text including the target character a second change condition for changing the target character to the target character if the token is not included in the vocabulary set used to train the first predictive model and the target text token including the target character is included in the vocabulary set have.

6 is a flowchart of a text correction method according to another embodiment of the present disclosure. The steps shown in FIG. 6 are exemplary, and additional steps may also be included within the scope of the present disclosure. The computing device 100 may acquire the input text ( S610 ). The computing device 100 may predict the target character based on at least a part of the input text using the first prediction model based on the learned neural network ( S630 ). The computing device 100 may predict the target text token based on at least a part of the input text using the second prediction model based on the learned neural network ( S650 ). The computing device 100 may change the target character corresponding to the target character among the plurality of characters included in the input text to the target character based on the prediction result of the first prediction model and the prediction result of the second prediction model (S670). can After performing steps S630 and S650, the computing device 100 may perform an operation of ensembles the probability values included in the prediction result of the first prediction model and the prediction result of the second prediction model in operation S670. The computing device 100 may perform a task of checking typos and contextual errors of individual characters and correcting texts by ensembles each prediction result.

7 is a performance comparison screen illustrating an effect of a text correction method according to an embodiment of the present disclosure. 7 is a performance as of January 28, 2021 of the International Optical Character Recognition Competition (SROIE, Scanned Receipts OCR and Information Extraction) hosted by the International Conference on Document Analysis and Recognition (ICDAR). This is the improvement result screen. Reference numeral 700 denotes precision and recall calculated for test data as a result of performing the text proofing method of the present disclosure. It can be seen from FIG. 7 that the improved effect in precision and recall compared to the calibration method of other participants was shown.

8 is a simplified, general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented. Although the present disclosure has been described above generally as being capable of being implemented by a computing device, those skilled in the art will appreciate that the present disclosure is a combination of hardware and software and/or in combination with computer-executable instructions and/or other program modules that may be executed on one or more computers. It will be appreciated that it can be implemented as a combination.

Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will appreciate that the methods of the present disclosure are suitable for single-processor or multiprocessor computer systems, minicomputers, mainframe computers as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, etc. (each of which is It will be appreciated that other computer system configurations may be implemented, including those that may operate in connection with one or more associated devices.

The described embodiments of the present disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Computers typically include a variety of computer-readable media. Any medium accessible by a computer can be a computer readable medium, and such computer readable media includes volatile and nonvolatile media, transitory and non-transitory media, removable and non-transitory media. including removable media. By way of example, and not limitation, computer-readable media may include computer-readable storage media and computer-readable transmission media. Computer-readable storage media includes volatile and non-volatile media, temporary and non-transitory media, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. includes media. A computer-readable storage medium may be RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device, or other magnetic storage device. device, or any other medium that can be accessed by a computer and used to store the desired information.

Computer readable transmission media typically embodies computer readable instructions, data structures, program modules or other data, etc. in a modulated data signal such as a carrier wave or other transport mechanism, and Includes any information delivery medium. The term modulated data signal means a signal in which one or more of the characteristics of the signal is set or changed so as to encode information in the signal. By way of example, and not limitation, computer-readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer-readable transmission media.

An exemplary environment 1100 implementing various aspects of the disclosure is shown including a computer 1102 , the computer 1102 including a processing unit 1104 , a system memory 1106 , and a system bus 1108 . do. A system bus 1108 couples system components, including but not limited to system memory 1106 , to the processing device 1104 . The processing device 1104 may be any of a variety of commercially available processors. Dual processor and other multiprocessor architectures may also be used as processing unit 1104 .

The system bus 1108 may be any of several types of bus structures that may be further interconnected to a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 1106 includes read only memory (ROM) 1110 and random access memory (RAM) 1112 . A basic input/output system (BIOS) is stored in non-volatile memory 1110, such as ROM, EPROM, EEPROM, etc., which BIOS is the basic input/output system (BIOS) that helps transfer information between components within computer 1102, such as during startup. contains routines. RAM 1112 may also include high-speed RAM, such as static RAM, for caching data.

The computer 1102 may also be configured with an internal hard disk drive (HDD) 1114 (eg, EIDE, SATA) - this internal hard disk drive 1114 may also be configured for external use within a suitable chassis (not shown). Yes-, magnetic floppy disk drive (FDD) 1116 (eg, for reading from or writing to removable diskette 1118), and optical disk drive 1120 (eg, CD-ROM) for reading from, or writing to, disk 1122, or other high capacity optical media such as DVD. The hard disk drive 1114 , the magnetic disk drive 1116 , and the optical disk drive 1120 are connected to the system bus 1108 by the hard disk drive interface 1124 , the magnetic disk drive interface 1126 , and the optical drive interface 1128 , respectively. ) can be connected to The interface 1124 for implementing an external drive includes at least one or both of Universal Serial Bus (USB) and IEEE 1394 interface technologies.

These drives and their associated computer readable media provide non-volatile storage of data, data structures, computer executable instructions, and the like. For computer 1102, drives and media correspond to storing any data in a suitable digital format. Although the description of computer readable media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art will use zip drives, magnetic cassettes, flash memory cards, cartridges, etc. It will be appreciated that other tangible computer-readable media such as etc. may also be used in the exemplary operating environment and any such media may include computer-executable instructions for performing the methods of the present disclosure.

A number of program modules may be stored in the drive and RAM 1112 , including an operating system 1130 , one or more application programs 1132 , other program modules 1134 , and program data 1136 . All or portions of the operating system, applications, modules, and/or data may also be cached in RAM 1112 . It will be appreciated that the present disclosure may be implemented in various commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 1102 via one or more wired/wireless input devices, for example, a pointing device such as a keyboard 1138 and a mouse 1140 . Other input devices (not shown) may include a microphone, IR remote control, joystick, game pad, stylus pen, touch screen, and the like. Although these and other input devices are often connected to the processing unit 1104 through an input device interface 1142 that is connected to the system bus 1108, parallel ports, IEEE 1394 serial ports, game ports, USB ports, IR interfaces, It may be connected by other interfaces, etc.

A monitor 1144 or other type of display device is also coupled to the system bus 1108 via an interface, such as a video adapter 1146 . In addition to the monitor 1144, the computer typically includes other peripheral output devices (not shown), such as speakers, printers, and the like.

Computer 1102 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 1148 via wired and/or wireless communications. Remote computer(s) 1148 may be workstations, computing device computers, routers, personal computers, portable computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, and are typically connected to computer 1102 . Although it includes many or all of the components described for it, only memory storage device 1150 is shown for simplicity. The logical connections shown include wired/wireless connections to a local area network (LAN) 1152 and/or a larger network, eg, a wide area network (WAN) 1154 . Such LAN and WAN networking environments are common in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to a worldwide computer network, for example, the Internet.

When used in a LAN networking environment, the computer 1102 is connected to the local network 1152 through a wired and/or wireless communication network interface or adapter 1156 . Adapter 1156 may facilitate wired or wireless communication to LAN 1152 , which also includes a wireless access point installed therein for communicating with wireless adapter 1156 . When used in a WAN networking environment, the computer 1102 may include a modem 1158, be connected to a communication computing device on the WAN 1154, or establish communications over the WAN 1154, such as over the Internet. have other means. A modem 1158 , which may be internal or external and a wired or wireless device, is coupled to the system bus 1108 via a serial port interface 1142 . In a networked environment, program modules described for computer 1102 or portions thereof may be stored in remote memory/storage device 1150 . It will be appreciated that the network connections shown are exemplary and other means of establishing a communication link between the computers may be used.

Computer 1102 may be associated with any wireless device or object that is deployed and operates in wireless communication, for example, printers, scanners, desktop and/or portable computers, portable data assistants (PDAs), communication satellites, wireless detectable tags. It operates to communicate with any device or place, and phone. This includes at least Wi-Fi and Bluetooth wireless technologies. Accordingly, the communication may be a predefined structure as in a conventional network or may simply be an ad hoc communication between at least two devices.

Wi-Fi (Wireless Fidelity) makes it possible to connect to the Internet, etc. without a wire. Wi-Fi is a wireless technology such as cell phones that allows these devices, eg, computers, to transmit and receive data indoors and outdoors, ie anywhere within range of a base station. Wi-Fi networks use a radio technology called IEEE 802.11 (a, b, g, etc) to provide secure, reliable, and high-speed wireless connections. Wi-Fi can be used to connect computers to each other, to the Internet, and to wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks may operate in unlicensed 2.4 and 5 GHz radio bands, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or in products that include both bands (dual band). .

One of ordinary skill in the art of this disclosure will understand that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, instructions, information, signals, bits, symbols, and chips that may be referenced in the above description are voltages, currents, electromagnetic waves, magnetic fields or particles, optical field particles or particles, or any combination thereof.

A person of ordinary skill in the art of the present disclosure will recognize that the various illustrative logical blocks, modules, processors, means, circuits and algorithm steps described in connection with the embodiments disclosed herein include electronic hardware, (convenience For this purpose, it will be understood that it may be implemented by various forms of program or design code (referred to herein as software) or a combination of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. A person skilled in the art of the present disclosure may implement the described functionality in various ways for each specific application, but such implementation decisions should not be interpreted as a departure from the scope of the present disclosure.

The various embodiments presented herein may be implemented as methods, apparatus, or articles of manufacture using standard programming and/or engineering techniques. The term article of manufacture includes a computer program, carrier, or media accessible from any computer-readable storage device. For example, computer-readable storage media include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical disks (eg, CDs, DVDs, etc.), smart cards, and flash drives. memory devices (eg, EEPROMs, cards, sticks, key drives, etc.). Also, various storage media presented herein include one or more devices and/or other machine-readable media for storing information.

It is understood that the specific order or hierarchy of steps in the presented processes is an example of exemplary approaches. Based on design priorities, it is to be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure. The appended method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present disclosure is not intended to be limited to the embodiments presented herein, but is to be construed in the widest scope consistent with the principles and novel features presented herein.

Claims (14)

A text proofing method performed by a computing device comprising at least one processor, the method comprising:
obtaining input text;
predicting a target character based on at least a part of the input text using a first prediction model based on the learned neural network; and
changing a target character corresponding to the target character among a plurality of characters included in the input text into the target character;
containing,
How to proofread text.
The method of claim 1,
Predicting the target character comprises:
performed based on at least one surrounding character having a position different from the target character included in the input text,
How to proofread text.
The method of claim 1,
The first prediction model is,
learned based on a vocabulary set comprising at least one training text token consisting of a predetermined number of characters;
How to proofread text.
The method of claim 1,
The changing step is
performed according to the target text and at least one predetermined change condition determined based on the target text,
How to proofread text.
5. The method of claim 4,
The predetermined change condition is
and a first change condition for changing the target character to the target character when one of a leading consonant, a middle consonant, and a final consonant in each syllable configuration of the target character and the target character does not match and the others match,
How to proofread text.
5. The method of claim 4,
The predetermined change condition is
When the target text token including the target character is not included in the vocabulary set used for training the first predictive model and the target text token including the target character is included in the vocabulary set, Including a second change condition to change to the target character,
How to proofread text.
The method of claim 1,
predicting a target text token based on at least a portion of the input text using a second prediction model based on the learned neural network;
further comprising,
How to proofread text.
8. The method of claim 7,
Predicting the target text token comprises:
performed based on at least one surrounding text token having a position different from a target text token included in the input text,
How to proofread text.
8. The method of claim 7,
The second prediction model is,
learned based on a text corpus comprising at least one training text composed of a predetermined number of text tokens;
How to proofread text.
8. The method of claim 7,
The changing step is
performed based on the prediction result of the first prediction model and the prediction result of the second prediction model,
How to proofread text.
8. The method of claim 7,
The changing step is
a probability value for each of the plurality of candidate characters included in the prediction result of the first prediction model; and
a probability value for each of the plurality of candidate text tokens included in the prediction result of the second prediction model;
performed according to the result of an ensemble operation based on
How to proofread text.
The method of claim 1,
The input text is
obtained by applying optical character recognition (OCR) technology to the document image, and
The predicting step or the changing step is,
Performed when the accuracy of recognition obtained as a result of applying the optical character recognition technology is less than or equal to a predetermined reference accuracy,
How to proofread text.
A computer program stored on a computer-readable storage medium, wherein the computer program, when executed on one or more processors, causes the following operations to be performed for correcting text, the operations comprising:
obtaining input text;
predicting a target character based on at least a part of the input text using a first prediction model based on the learned neural network; and
changing a target character corresponding to the target character among a plurality of characters included in the input text into the target character;
comprising,
A computer program stored on a computer-readable storage medium.
A text proofing device comprising:
one or more processors;
Memory; and
network department;
includes, and
The one or more processors,
get the input text,
predicting a target character based on at least a part of the input text using a first prediction model based on the learned neural network, and
changing a target character corresponding to the target character among a plurality of characters included in the input text to the target character,
text proofing device.

Images