KR102051085B1 - Device and method for providing nationality information of user name using neural networks - Google Patents

Abstract

An apparatus for providing nationality information based on an artificial neural network according to an aspect of the present invention includes a memory in which a nationality information providing program is stored, a data input / output module for performing data input / output, and a processor. At this time, a nationality information providing program is driven by the processor to predict the nationality of the subject based on the subject's name input through the data input / output module, but the nationality information providing program is RNN-LSTM (Recurrent Neural Network-Long). Based on a learning model based on Short Term Memory, a unigram that provides a nationality having the highest probability that the inputted subject's name is classified as an output value, and decomposes the subject's name in alphabetical order according to the input order. The RNN is performed by unigram embedding, bigram embedding to decompose and vectorize two letters according to an input order, and trigram embedding to decompose and vectorize three letters according to an input order. -After transferring to LSTM memory of LSTM, combine the vector value output from each LSTM memory. Perform the process of transferring to all layers.

Description

DEVICE AND METHOD FOR PROVIDING NATIONALITY INFORMATION OF USER NAME USING NEURAL NETWORKS}

The present invention relates to a nationality classification method and apparatus capable of predicting the nationality of a person from a person's name using a neural network structure.

Recently, various kinds of Internet services have been provided through web sites or applications of smart terminals, and companies are actively making efforts to provide optimal services for each user through big data. Particularly, if a user's nationality can be predicted only by the name without requesting additional information from the user, a study for this can be provided since it is possible to provide a customized service more suitable to an individual user.

In general, since a person's name has a pattern frequently used for each nationality, the name alone can roughly identify or predict a nationality. For example, suffixes such as “-sky” and “-v” appear frequently in Russians, and in Germans “-mann” and “-urg”. Suffixes often appear, so approximation of nationality is possible.

As a major factor for the prediction of nationality, the first is that the language of the country is the most important factor to consider, the second is that geographic location can provide useful information, and the third is the decisive role of each country's history. Is that you can.

As described above, there are various considerations, so that it is difficult to predict nationality only by its name, and in the present invention, this technical difficulty is solved through a new platform based on a recurrent neural network.

Chinese Patent Application No. 106600283 (Method and system for identifying the name nationalities as well as method and system for determining transaction risk)

An embodiment of the present invention is to solve the problems of the prior art, a nationality information providing apparatus and method that can improve the learning and classification performance in the process of predicting and providing nationality information of the name based on the cyclic neural network To provide.

However, the technical problem to be achieved by the present embodiment is not limited to the technical problem as described above, and other technical problems may exist.

As a technical means for achieving the above-described technical problem, an apparatus for providing nationality information based on an artificial neural network according to an aspect of the present invention includes a memory, a data input / output module and a processor for performing data input / output and a nationality information providing program Include. At this time, a nationality information providing program is driven by the processor to predict the nationality of the subject based on the subject's name input through the data input / output module, but the nationality information providing program is RNN-LSTM (Recurrent Neural Network-Long). Based on a learning model based on Short Term Memory, a unigram that provides a nationality having the highest probability of classifying the inputted subject's name as an output value, and decomposes the subject's name in alphabetical order according to the input order. The RNN is performed by unigram embedding, bigram embedding to decompose and vectorize two letters according to an input order, and trigram embedding to decompose and vectorize three letters according to an input order. -After transferring to LSTM memory of LSTM, combine the vector value output from each LSTM memory. Perform the process of transferring to all layers.

In addition, the method for providing nationality information based on an artificial neural network according to another aspect of the present invention includes the steps of: receiving name data that is a classification target; Character embedding the received name data; Transferring the character embedded data to a LSTM memory of RNN-LSTM (Recurrent Neural Network-Long Short Term Memory); Combining the vector values output from the LSTM memory and transferring the combined vector values to a hidden layer; And predicting nationality information on the name data according to a probability value calculated through the hidden layer. In this case, the character embedding may include a unigram embedding process of decomposing the subject's name into alphabetical characters according to an input order, and vectorizing the bigram embedding process to decompose and vectorize each character according to the input order. ) And trigram embedding to vectorize and decompose three characters according to the input order.

According to any one of the problem solving means of the present invention described above, it is possible to further improve the performance of the artificial neural network predicting the nationality from the name of the person. In particular, by performing character embedding such as unigram embedding, bigram embedding, and trigram embedding on the names and inputting them into the ANN neural network of the RNN-LSTM structure, the learning and classification performance can be improved.

1 is a block diagram of a nationality information providing apparatus according to an embodiment of the present invention.
2 is a view for explaining the configuration of an artificial neural network model according to an embodiment of the present invention.
3 is a flowchart illustrating a method for providing nationality information based on a name according to an embodiment of the present invention.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted in the drawings, and like reference numerals designate like parts throughout the specification. In addition, while describing with reference to the drawings, even if the configuration shown by the same name may be different according to the drawing number, the drawing number is just described for convenience of description and the concept, features, functions of each configuration by the corresponding reference number Or the effects are not to be construed as limiting.

Throughout the specification, when a part is "connected" to another part, this includes not only "directly connected" but also "electrically connected" with another element in between. . In addition, when a part is said to "include" a certain component, which means that it may further include other components, except to exclude other components unless specifically stated otherwise, one or more other It is to be understood that the present invention does not exclude the possibility of the presence or the addition of features, numbers, steps, operations, components, parts, or combinations thereof.

In the present specification, the term 'part' or 'module' includes a unit realized by hardware or software, and a unit realized using both, and one unit is realized by using two or more pieces of hardware. Two or more units may be implemented by one hardware.

1 is a block diagram of a nationality information providing apparatus according to an embodiment of the present invention.

The nationality information providing apparatus 100 illustrated in FIG. 1 includes a data input / output module 110, a memory 120, and a processor 130.

The data input / output module 110 may perform a function of receiving or transmitting data through a communication module. The data input / output module 110 may be a device including hardware and software necessary for transmitting and receiving a signal such as a control signal or a data signal through a wired or wireless connection with another network device.

The memory 120 stores a program that predicts the user's nationality from the user's name information. The program is implemented based on an artificial neural network that collects and learns user's name information and performs classification processing for predicting nationality information of an input user's name. The program is executed by the processor 130 to perform learning function and classification. Perform a function.

The memory 120 may be referred to as a nonvolatile storage device that maintains stored information even when power is not supplied, or a volatile storage device that requires power to maintain stored information.

2 is a view for explaining the configuration of an artificial neural network model according to an embodiment of the present invention.

In the present invention, a nationality verification service is provided based on generally known Recurrent Neural Network-Long Short Term Memory (RNN-LSTM). RNN is a structure that can be used to share previous information and is known as a neural network structure that is suitable for learning continuous data in time series.However, if the temporal gap between each data becomes long, it can not be remembered. Dependency problem has been known, and the RNN-LSTM technique has been developed to improve it. RNN-LSTM is known to solve the problem of gradient vanishing or exploding in the existing RNN technique, and to better convey the error value during backpropagation. LSTM is mainly used to learn the representation of long time series connected data, and is replaced by a memory block consisting of three gates: an input gate, an output gate, and a forge gate. Has The LSTM is known to be able to learn the representation of data longer than the conventional RNN by utilizing the memory c and the state h. The detailed structure and operation of the LSTM block are the same as those known in the art, and thus the detailed description thereof Will be omitted.

According to the present invention, there is a technical feature in that character embedding is performed on an input name and then inputted to each LSTM module.

That is, unigram embedding processing (202) for decomposing and inputting the input name by alphabetical characters according to the input order (Bigram embedding, 204), input sequence for decomposing and transmitting two letters according to the input order. According to the trigram embedding process (Trigram embedding, 206) to be decomposed by three letters each to perform.

Given the input name X = [x_0, x_1, ..., x_ (n-1)], x_k means each letter of the name. Here, x_k may be unigram, bigram, or trigram. For example, when the input data is "James", X may be represented by [J, a, m, e, s] according to the unigram embedding process, or [Ja, am, me, es] can be expressed as X, or [Jam, ame, mes] can be expressed by the trigram embedding process. As such, in order to increase the expressive power of a unigram, a biggram, or a trigram, it may be expressed as a vector having a lower dimension, which is called a character embedding or a distributed representation. In the present invention, the input name data is decomposed into unigrams, bigrams, and trigrams, respectively, and unigram embedding, bigram embedding, and trigram embedding processing for expressing the data as vectors are performed. Each of the vectors embedded in the character embedding process is obtained as three states (h) through a separate LSTM, and in the present invention, each state (h) is concatenated into one long vector (concatenate, 208) so that the hidden layer and the soft layer are soft. Through the maxmax layer, the probability value (Equation 1) in which the input data belongs to the classification of a specific country is calculated.

On the other hand, the classification model that classifies names according to nationality is learned to maximize the conditional probability according to the following equation.

[Equation 1]

는 해당 이름의 국적(또는 인종) 정보이며, D( )는 θ라는 학습 파라메터를 갖고 있는 분류 모델이며, g( )는 특징(feature)을 추출하기 위한 함수이다. 이때, 국적은 127개 정도의 경우의 수가 존재하지만 인종은 13 개 정도로 그 수가 현저히 줄어들기 때문에, 국적을 정확하게 분류하는 것이 더욱 어려운 문제에 해당한다. here

Is the nationality (or race) information of the name, D () is a classification model with a learning parameter of θ, and g () is a function for extracting features. At this time, since the number of nationalities is about 127, but the number of races is considerably reduced to about 13, it is more difficult to classify nationalities accurately.

In more detail, according to the RNN technique, the hidden state h may be defined as in Equation 2.

[Equation 2]

는 l 번째 레이어에서의 시간 구간 t 에서의 RNN의 히든 상태를 나타내는 것이다.

,

및

는 학습해야할 파라미터를 나타내고,

는 기본 시그모이드(sigmoid) 함수를 나타낸다. 또한, 각 국적

클래스에 대한 확은은 아래 수학식 3과 같이 정의될 수 있다.

Denotes the hidden state of the RNN in the time interval t in the l th layer.

,

And

Represents the parameters to be learned,

Denotes a basic sigmoid function. In addition, each nationality

The confirmation of the class may be defined as in Equation 3 below.

[Equation 3]

와

는 훈련가능한 파라미터들이다. 이러한 RNN 기반의 학습모델에 앞서 소개한 LSTM 모델을 병합하면 다음 수학식 4와 같이 구현될 수 있다.At this time,

Wow

Are trainable parameters. Merging the aforementioned LSTM model to the RNN-based learning model can be implemented as shown in Equation 4 below.

[Equation 4]

와

는 훈련가능한 파라미터들이고,

는 행렬 요소별(element wise) 곱을 나타낸다. 또한, 장기 메모리(Long Term memory)는 입력 게이트(input gate) i 와 이전 셀

이 저장된

로부터 복원될 수 있고, 포겟 게이트 f 에 업데이트될 수 있다.

Wow

Are the trainable parameters,

Denotes a matrix element wise product. Long Term memory also includes input gate i and the previous cell.

This stored

Can be recovered from, and updated to the forge gate f.

The probabilities finally obtained through the above process are used to calculate the negative log likelihood (NLL) through the cross entropy technique. You can learn a model that maximizes that probability.

를 통해 해당 확률이 최대화도록 한다.For example, cross-entropy loss function as shown in Equation 5 below.

To maximize the probability.

3 is a flowchart illustrating a method for providing nationality information based on a name according to an embodiment of the present invention.

First, name information for confirming nationality information is received as input data (S310). Input data is received and stored through the data input / output module 110 of the nationality information providing device 100 of FIG. 1.

Next, character embedding is performed on the input name data (S320).

In the present invention, a unigram embedding process for decomposing and vectorizing characters by an alphabetical order according to an input order, a bigram embedding process for decomposing and vectorizing two characters according to an input order, and decomposing three characters according to an input order. Trigram embedding to vectorize is performed.

Next, the character-embedded data is input to each LSTM, and a single vector is generated by connecting values output from the LSTM (S330).

The generated vector is transmitted to the hidden generator, and based on this, a probability value indicating which country is most suitable is calculated (S340).

As described above, character embedding can be carried out in several stages to design the RNN-LSTM learning model more precisely, thereby further improving the accuracy of the service of estimating nationality based only on the name of a person.

The apparatus and method for providing nationality information according to the embodiment of the present invention described above may be implemented in the form of a recording medium including instructions executable by a computer, such as a program module executed by a computer. Such recording media includes computer readable media, and computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, computer readable media includes computer storage media, which are volatile and nonvolatile implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Both removable and non-removable media.

The foregoing description of the present invention is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

In addition, while the methods and systems of the present invention have been described in connection with specific embodiments, some or all of their components or operations may be implemented using a computer system having a general purpose hardware architecture.

The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. .

100: nationality information providing device 110: data input / output module
120: memory 30: processor

Claims (4)

In the device for providing nationality information based on artificial neural network,
Memory where the nationality information providing program is stored;
A data input / output module for performing data input / output, and
Include processors,
The nationality information providing program is driven by the processor to predict the nationality of the subject based on the subject's name input through the data input / output module, but the nationality information providing program is RNN-LSTM (Recurrent Neural Network-Long Short). Based on a learning model based on Term Memory, the output of a nationality having the highest probability that the inputted subject's name is classified.
Unigram embedding processing to decompose the subject's name into alphabetical characters according to the input order and vectorization, Bigram embedding processing to decompose and vectorize two characters according to the input order The apparatus for providing nationality information comprising performing trigram embedding processing to decompose and vectorizing and transferring the result to the LSTM memory of the RNN-LSTM, and then combining the vector values output from each LSTM memory to the hidden layer. The method of claim 1,
The processor generates a combination of a value output from the LSTM memory after the unigram embedding process, a value output from the LSTM memory after the bigram embedding process, and a value output from the LSTM memory after the trigram embedding process. The calculated vector value is transmitted to the hidden layer, and the hidden layer calculates a conditional probability that the vector value is classified as a predetermined nationality based on a classification model classifying names according to nationality, and the calculated conditional probability is maximum. Nationality information providing apparatus to perform a learning operation to be. In the nationality information providing method using a nationality information providing device for providing nationality information based on an artificial neural network,
Receiving, by the nationality information providing apparatus, name data of a subject to be classified;
Character embedding, by the nationality information providing apparatus, the name data of the received subject;
Transmitting, by the nationality information providing device, the character embedded data to an LSTM memory of a Recurrent Neural Network-Long Short Term Memory (RNN-LSTM);
Transmitting, by the nationality information providing apparatus, a vector value output from the LSTM memory to a hidden layer; And
And predicting, by the apparatus for providing nationality information, nationality information on the name data according to a probability value calculated through the hidden layer.
Embedding the character,
Unigram embedding processing to decompose the subject's name data into alphabetical characters according to the input order and vectorize, Bigram embedding processing to decompose and vectorize two characters according to the input order A method of providing nationality information, wherein trigram embedding is performed by decomposing and vectorizing each other. The method of claim 3, wherein
And performing a learning operation to maximize the calculated probability value.

Images