KR102156249B1 - Indexing system and method using variational recurrent autoencoding - Google Patents

Abstract

The index system of the variable recurrent autoencoding method according to an embodiment of the present invention learns the value of a latent variable through a training process of encoding and decoding a predetermined keyword set, and based on this A first variational recurrent autoencoder for calculating a value of a latent variable by encoding a keyword set corresponding to voice or text input; A second variable cycle autoencoder for calculating a value of a latent variable by learning a value of a latent variable through a training process of encoding and decoding a predetermined image, and encoding a user's image input based thereon; A result of comparing the value of the latent variable learned by the first and second differential autoencoder and the value of the latent variable calculated by the first variable cycle autoencoder or the second variable cycle autoencoder And an indexing unit for indexing an input from a user based on, and the values of the latent variables learned in the first and second cyclic autoencoders are shared.

Description

Automatic indexing system and method of variable cycle auto-encoding method {INDEXING SYSTEM AND METHOD USING VARIATIONAL RECURRENT AUTOENCODING}

The present invention relates to an automatic indexing system and method using a differential cycle auto-encoding method.

Recently, intelligent information search systems and chatbot services that provide various services and information are increasing. The existing intelligent information search engine uses a database that stores answers to questions, and calculates the relationship between the term frequency and the inverse document frequency, Use a method of finding the appropriate answer.

In this regard, Korean Patent Application Laid-Open No.-2018-0042763 provides a communication module for receiving a user query in the form of a completed or incomplete sentence input through an access channel and transmitting a response to the user query; A natural language processing engine that performs syntax analysis and morpheme analysis on user queries received through the communication module; A database including a language dictionary database, a financial knowledge database, and a conversation scenario database for providing financial services; A conversation that interprets the dictionary meaning and contextual meaning of the user query based on the syntax and morpheme analyzed by the natural language processing engine, and generates a response to the user query based on the financial service provision scenario stored in the conversation scenario database. It provides a chat-type financial robot including a processing engine.

However, this is about finding answers (responses) directly, not indexing questions to find answers.

In the present invention, when an input question is a natural language or an image, it is indexed (classified) using the value of a latent variable of a variable cyclic encoder, so that it is easy to find an answer in a database using a Variational Recurrent Autoencoding method. We would like to provide an automatic indexing system and method.

The index system of the variable cyclic auto-encoding method according to an embodiment of the present invention learns the value of a latent variable through a training process of encoding and decoding a preset predetermined keyword set, and based on this, the user's voice or text input A first variational recurrent autoencoder for calculating a value of a latent variable by encoding a corresponding keyword set; A second variable cycle autoencoder for calculating a value of a latent variable by learning a value of a latent variable through a training process of encoding and decoding a predetermined image, and encoding a user's image input based thereon; A result of comparing the value of the latent variable learned by the first and second differential autoencoder and the value of the latent variable calculated by the first variable cycle autoencoder or the second variable cycle autoencoder And an indexing unit for indexing an input from a user based on, and values of latent variables learned in the first and second cyclic autoencoders are shared.

The index system of the differential cycle auto-encoding method may further include a keyword generator configured to generate a keyword set corresponding to the user's text or voice input by analyzing a morpheme of the user's text or voice input.

The first differential cycle autoencoder may include: a first encoding unit encoding the preset predetermined keyword set or a keyword set corresponding to the user's voice or text input; And a first decoding unit for decoding a value of a latent variable generated by the first encoding unit encoding the predetermined keyword set.

The first encoding unit includes a number of LSTMs (Long Short Term Memory) equal to the number of keywords constituting the input keyword set, and the keywords are respectively input to the LSTM in the order of input, and the LSTM is a chain format. Can be connected to.

The second differential cyclic autoencoder may include: a second encoding unit encoding the predetermined image or an image input of the user; And a second decoding unit for decoding a value of a latent variable generated by the second encoding unit encoding the predetermined image, and the second encoding unit and the second decoding unit include a convolutional neural network (CNN). can do.

It may further include an update unit for updating the index of the index unit by using the value of the latent variable calculated by the first cyclical autoencoder or the second cyclical autoencoder.

The preset predetermined keyword set relates to tax affairs, and the predetermined predetermined image may be an image capable of distinguishing the type of document related to tax affairs.

In the indexing method of the variable cycle auto-encoding method according to an embodiment of the present invention, a value of a latent variable is learned through a training process of encoding and decoding a preset predetermined keyword set using a first variable cycle auto-encoder, and Learning a value of a latent variable shared with a latent variable learned in the first variable cycle autoencoder through a training process of encoding and decoding a set predetermined image using a second variable cycle autoencoder; Calculating a value of a latent variable corresponding to a user's input by using the first differential cycle autoencoder or the second differential cycle autoencoder; And indexing an input from a user based on a result of comparing the learned value of the latent variable with the calculated value of the latent variable.

The indexing method of the differential cycle auto-encoding method further includes the step of generating a keyword set corresponding to the user's text or voice input by analyzing the morpheme of the user's text or voice input, and in the calculating step, the user When the input of is text or voice, the value of the latent variable corresponding to the generated keyword set may be calculated using the first differential cycle autoencoder.

The indexing method of the differential cycle auto-encoding method includes the step of updating an index used in the indexing step by using a value of a latent variable calculated by the first differential cycle autoencoder or the second differential cycle autoencoder. It may contain more.

The preset predetermined keyword set relates to tax affairs, and the predetermined predetermined image may be an image capable of distinguishing the type of document related to tax affairs.

A keyword generator for generating a keyword set corresponding to the user's natural language input by analyzing a morpheme of the user's natural language input;

The index system of the variable cyclic auto-encoding method according to an embodiment of the present invention learns a value of a latent variable through a training process of encoding and decoding a predetermined keyword set, and is generated by the keyword generator based on this. A variable cycle autoencoder that calculates a value of a latent variable by encoding a keyword set; An index unit for indexing an input from a user based on a result of comparing the value of the latent variable learned by the differential cycle autoencoder with the value of the latent variable calculated by the differential cycle autoencoder; And an update unit for updating the index of the index unit by using the value of the latent variable calculated by the differential cycle autoencoder.

According to an embodiment of the present invention, since the input question is indexed using the variance cycle auto-encoding method, indexing with high accuracy is possible.

1 is a diagram showing the configuration of an index system of an auto-encoding method according to an exemplary embodiment of the present invention.
2 is a diagram showing the operation of the morpheme analyzer of FIG. 1.
FIG. 3 is a diagram for explaining the operation of the differential cycle autoencoder of FIG. 1.
4 is a diagram illustrating an example of the structure of the LSTM of FIG. 3.
5 is a diagram illustrating a configuration of an index system of an auto-encoding method according to a differential cycle according to another embodiment of the present invention.
6 is a diagram for describing an operation of some components of FIG. 5.
FIG. 7 is a diagram showing the configuration of a specialized field response service system using a differential cycle auto-encoding method according to an embodiment of the present invention.
FIG. 8 is a diagram illustrating a specialized field response service method using a variable cycle auto-encoding method using a chatbot according to an embodiment of the present invention.
9 is a diagram showing an experiment result of a specialized field response service system and method of a differential cycle auto-encoding method according to an embodiment of the present invention.

Based on the principle that the inventor can appropriately define the concept of terms in order to describe his own invention in the best way, terms or words used in the present specification and claims are consistent with the technical idea of the present invention. It should be interpreted as meaning and concept.

Throughout the specification, when a part "includes" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary. In addition, when one component is said to be "connected", "transmitted", "transmitted", "received" or "transmitted" to another component, it is not only the case that is directly connected, transmitted, transmitted, received, or transmitted It also includes cases that are indirectly connected, transmitted, transmitted, received or transmitted through components. In addition, terms such as "...unit", "...group", "module", and "device" described in the specification mean units that process at least one function or operation, which is a combination of hardware or software or hardware and software. It can be implemented as

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

First, a first embodiment of the present invention will be described with reference to FIGS. 1 to 4.

FIG. 1 is a diagram showing the configuration of an index system 1 of a variant recurrent autoencoding method according to an embodiment of the present invention.

Referring to FIG. 1, the index system 1 is a system that receives natural language and generates an index value for searching a database for a response corresponding thereto. The index system 1 includes a morpheme analyzer 100, an autoencoder 200 for variance circulation, an index unit 300, and an update unit 400.

The morpheme analyzer 100 analyzes the morpheme of the natural language and generates a keyword set of the natural language. The natural language input to the morpheme analyzer 100 may be a completed sentence or an incomplete sentence composed of one or more words. The natural language may be in Korean form, for example. The keyword set may be a general noun, proper noun, verb, etc. excluding investigations that do not have special meaning in natural language input. The keyword set includes at least one keyword.

The variable cycle autoencoder 200 generates a latent variable by processing a keyword set in a variable cycle auto encoding method.

The index unit 300 indexes the keyword set according to the value of the latent variable generated by the variable cycle autoencoder 200.

The update unit 400 updates the index by using the value of the latent variable generated by the variance cycle autoencoder 200.

2 is a diagram showing the operation of the morpheme analyzer 100 of FIG. 1. The morpheme analyzer 100 generates a keyword set by tokenizing natural language.

Referring to FIG. 2, when the natural language "This year is corporate tax" is input, it indicates that {this year, corporate tax} is generated as a set of keywords. In this embodiment, the keyword set {this year, corporate tax} is composed of two keywords "this year" and "corporate tax". For example, the morpheme analyzer 100 may be used in a Lucene-based search engine.

Next, the operation of the variable cycle autoencoder 200 will be described with reference to FIG. 3.

The variable cycle autoencoder is a kind of unsupervised learning and is widely used in dimensional reduction and generation models. The key to the variable cycle autoencoder is to learn that the latent variable (Z) follows a normal distribution (diagonal Gaussian) of mean (μ) and variance (σ). Since the posterior distribution p(z|x) is difficult to calculate (Intractable), we approximate q(z|x) and p(z|x) using Variational Inference. Kullback-Leibler Divergence can be used to minimize the difference between q(z|x) and p(z|x) as follows.

는 항상 0보다 크거나 같다.

를 최대화하면

가 최대가 되도록 하는 하한(Evidence Lower Bound)이며 목적함수(Objective Function)가 된다.In the above equation

Is always greater than or equal to zero.

Maximizing

It is the Evidence Lower Bound that makes is the maximum and becomes the Objective Function.

FIG. 3 is a diagram for explaining the operation of the variable cycle autoencoder 200 of FIG. 1.

Referring to FIG. 3, the variable cyclic autoencoder 200 includes encoding Long Short Term Memory (LSTM; encoding units; 210, 220) into which one or more keywords constituting a keyword set are respectively input, and encoding LSTMs (210, 220). It may include a decoding LSTM (decoder; 230, 240, 250) in which the latent variable Z generated in is decoded.

First, the encoding LSTMs 210 and 220 and the decoding LSTMs 230, 240 and 250 of the variable cyclic autoencoder 200 may perform a training operation. The encoding LSTMs 210 and 220 calculate a latent variable by encoding each of a plurality of preset keyword sets, and perform training in a manner of performing decoding on the calculated latent variable. Accordingly, a value of a potential variable suitable for a plurality of preset keyword sets may be calculated. 3 shows that two encoding LSTMs 210, 220 and three decoding LSTMs 230, 240, 250 are included, but the number of encoding LSTMs and decoding LSTMs varies according to the number of keywords in the input keyword set. I can.

As shown in FIG. 3, the output of the LSTM 210 at the front end may be a chain structure that becomes the input of the LSTM 220 at the rear end. In the variable cyclic autoencoder 200 of the present embodiment, since each keyword is sequentially input to the LSTM and encoded, the linkage between the keywords can be considered.

Next, when a natural language is input from a user and a keyword set is generated by the morpheme analysis unit 100, the encoding LSTMs 210 and 220 calculate a latent variable for a keyword set corresponding to the natural language. The calculated latent variables can be used as a standard for indexing in the future.

As shown in FIG. 3, when the keyword set {this year, corporate tax} is input, "this year" is input to the LSTM 210, and the output of the LSTM 210 and the "corporate tax" are input to the LSTM 220. . Accordingly, a latent variable (Z) having an average (μ) and a variance (σ) in the LSTM 220 is generated. 3 illustrates that two encoding LSTMs 210 and LSTMs corresponding to two keywords are included as an example, but the number of encoding LSTMs may vary according to the number of keywords constituting the keyword set.

4 is a diagram illustrating an example of the structure of the LSTM of FIG. 3.

Referring to FIG. 4, keywords of a keyword set generated by the morpheme analyzer 100 are sequentially input to Xt-1 and Xt. In the embodiment of FIG. 4, "this year" may be input to Xt-1, and "corporation tax" may be input to Xt. ht-1 and ht are encoded vector values, and ht finally output may correspond to the latent variable Z of FIG. 3.

The LSTM 210 and the LSTM 220 may have the same structure. Therefore, in the following description, the LSTM 220 will be mainly described.

The LSTM 220 changes the cell state from Ct-1 to Ct when viewed as a whole. In FIG. 4, square marks 41, 42, 43, 47 represent neural network layers, and circle marks 44, 45, 46, 49 represent point calculations.

First, the sigmoid layer 41 receives ht-1 and xt and outputs ft with a value between 0 and 1. And ft is multiplied by Ct-1 representing the previous cell state (see 44) to determine how much to reflect the previous cell state. ft can be expressed by the following equation.

The sigmoid layer 42 and the tanh layer 43 receive inputs of ht-1 and xt and determine how much to reflect in the current cell state Ct (see 46).

Similar to the sigmoid layer 41, the sigmoid layer 42 receives ht-1 and xt and outputs it having a value between 0 and 1, and it can be expressed by the following equation.

를 생성한다. tanh layer(213)의 출력값

는 다음의 수식으로 표현될 수 있다.The tanh layer 43 receives ht-1 and xt, and is a new candidate value (vector).

Create output value of tanh layer(213)

Can be expressed by the following equation.

Accordingly, the current cell state Ct can be expressed by the following equation (refer to 44, 45, and 46).

Next, the sigmoid layer 47 receives ht-1 and xt and outputs a value between 0 and 1, and the output ot of the sigmoid layer 47 can be expressed by the following equation.

ht is output through point calculations 48 and 49, and ht can be expressed as follows.

In the above, the LSTM having the structure of FIG. 4 has been described as an example, but the scope of the present invention is not limited thereto. For example, some of the neural network layers 41, 42, 43, 47 may be omitted, and the neural network layers 41, 42, 43, 47 and point operations 44, 45, 46, 48, and 49 are connected. The relationship may be different.

Next, a second embodiment of the present invention will be described with reference to FIGS. 5 and 6.

5 is a diagram illustrating a configuration of an index system 1000 of a differential cycle auto-encoding method according to an embodiment of the present invention.

Referring to FIG. 5, the index system 1000 is a system that receives natural language or images in the form of text or voice, and outputs an index value to search for a response corresponding thereto. The index system 2 includes a morpheme analysis unit 1100, a variance circulation autoencoder 1200, an index unit 1300, and an update unit 1400.

Referring to FIG. 5, the morpheme analysis unit 1100 generates a keyword set by receiving a user's natural language in the form of text or voice. Since the function of the morpheme analysis unit 1100 is the same as that of the morpheme analysis unit 100 described with reference to FIGS. 1 and 2, a detailed description is omitted.

The incremental circulation autoencoder 1200 includes a first incremental circulation autoencoder VAE1 and a second incremental circulation autoencoder VAE2.

The first variable cyclic autoencoder VAE1 learns the value of the latent variable Z through a training process of encoding and decoding a preset predetermined keyword set, and based on this, a keyword set corresponding to the user's voice or text input The value of the latent variable Z is calculated by encoding.

The second variable cyclic autoencoder VAE2 learns the value of the latent variable Z through a training process for encoding and decoding a predetermined image, and encoding the user's image input based on the value of the latent variable Z. Calculate the value of

The index unit 1300 includes the value of the latent variable Z learned by the first variable cycle autoencoder VAE1 or the second variable cycle autoencoder VAE2, and the first variable cycle autoencoder VAE1 or the first The input from the user is indexed based on the result of comparing the value of the latent variable Z calculated by the two-variable cyclic autoencoder VAE2.

The update unit 1400 updates the index by using the value of the latent variable Z calculated by the first variable cycle autoencoder VAE1 or the second variable cycle autoencoder VAE2.

FIG. 6 is a diagram for explaining the operations of the first cyclical autoencoder VAE1, the second cyclical autoencoder VAE2, and the updater 1400 of FIG. 5.

The first variable cycle autoencoder VAE1 is the same as the variable cycle autoencoder 200 except that the latent variable Z is shared with the second variable cycle autoencoder VAE2. Specifically, the first variable cyclic autoencoder VAE1 includes: first encoding units 1210 and 1220 for encoding a preset predetermined keyword set or a keyword set corresponding to the user's voice or text input; And first decoding units 1230, 1240, and 1250 for decoding a value of the latent variable Z generated by encoding a preset keyword set by the first encoding units 1210 and 1220.

The first encoding units 1210 and 1220 include the same number of LSTMs 1210 and 1220 as the number of keywords constituting the input keyword set, and the keywords are respectively input to the LSTMs 1210 and 1220 in the order of input. The LSTM (1210, 1220) can be connected in a chain.

The second variable cyclic autoencoder VAE2 includes: second encoding units 1211 and 1221 for encoding a predetermined image or an image input of the user; And second decoding units 1231 and 1241 for decoding a value of the latent variable Z generated by the second encoding units 1211 and 1221 encoding a predetermined image, and a second encoding unit 1211 , 1221) and the second decoding units 1231 and 1241 may include a convolutional neural network (CNN).

The second differential cycle autoencoder VAE2 inputs an image from a user. For example, tax-related documents may include different images depending on the type. Depending on the type of corporate tax delay, income tax refund, and VAT, the image or color of the format included in the document may be different. The second differential cycle autoencoder VAE2 calculates the value of the latent variable Z according to this image.

As shown in FIG. 6, the second variable cyclic autoencoder VAE2 includes a Convolutional Neural Network (CNN). CNN refers to a structure that copes with the FC (Fully-Connected Layer) of the existing DNN (Deep Neural etwork) as a convolution layer. Since FC can only receive 1D data, spatial information is lost (inappropriate) when processing 3D image data (Width, Height, Channel (3 Channel for RGB, 1 Channel for Black and White)). CNN generally means composed of Convolutional Layer-Pooling-Convolutional Layer-Pooling-Fully Connected Layer, and is suitable for image processing. The Convolutional Layer creates a Feature Map by obtaining the sum of the filter (kernel) input image and element-wise multiplication.

At this time, the value of the latent variable Z learned by the first variable cycle autoencoder VAE1 and the second variable cycle autoencoder VAE2 is shared. Accordingly, the first variable cycle autoencoder VAE1 and the second variable cycle autoencoder VAE2 learn and calculate the value of the latent variable Z under the influence of each other. In other words, the first variable cyclic autoencoder (VAE1) is based on the value of the latent variable (Z) learned and calculated by the second cyclical autoencoder (VAE2) as well as the first cyclical autoencoder (VAE2). The value of the latent variable (Z) corresponding to the input is calculated, and the second variable cycle autoencoder (VAE2) is learned and calculated from the second variable cycle autoencoder (VAE2) as well as the first variable cycle autoencoder (VAE1). The value of the latent variable Z corresponding to the user's input is calculated based on the value of the latent variable Z.

The update unit 1400 updates the index of the index unit 1300 by using the value of the latent variable Z calculated by the first variable cycle autoencoder VAE1 or the second variable cycle autoencoder VAE2.

For example, as shown in FIG. 6, the user inputs the text "Corporate tax this year", and the value of the latent variable Z from the first variable circulation autoencoder VAE1 is [0.5, 2.0, 0.4 ] Is output, the update unit 1400 updates the index unit so that the value of the potential variable of [0.5, 2.0, 0.4] and the index value of the corporate tax item correspond.

Alternatively, if the user inputs the image shown in FIG. 6 and the value of the latent variable Z is [1.5, 0.7, 0.9] from the second differential cycle autoencoder VAE2, the update unit 1400 [ 1.5, 0.7, 0.9] and the index value of the year-end settlement item are updated to correspond to each other.

Next, a description will be given of a specialized field response service system 2 of a differential cycle auto-encoding method according to an embodiment of the present invention with reference to FIG. 7.

Referring to FIG. 7, the specialized field response service system 2 includes an input/output unit 2100, an index system 2200, a search unit 2300, and a database 2400.

The input/output unit 2100 transmits the input question to the index system 2200 when a user inputs a question in a chat window displayed on a screen of an input terminal such as a mobile phone or a computer. Also, when an answer corresponding to an input question is searched in the database 2400, the input/output unit 2100 outputs the searched answer to the chat window. The input/output unit 2100 may perform a function of converting the input into text when the user inputs voice.

Additionally, the input/output unit 2100 may block indiscriminate access to data by issuing an authentication key for each user. Specifically, the input/output unit 210 may generate an authentication key for each user, store it in a user database (not shown), and provide the generated authentication key to the user.

The index system 2200 may be the index system 1 according to the first embodiment of the present invention or the index system 1000 according to the second embodiment of the present invention. The index system 2200 calculates a value of a latent variable according to a user's input, and outputs an index value corresponding to the calculated value of the latent variable.

The search unit 2300 searches for answers in the database 2400 according to the index value received from the index system 2200. Then, the searched answer is provided to the input/output unit 2100 to be output to the user.

The database 2400 stores a plurality of questions and their answers, and the plurality of questions are classified according to index values. The questions stored in the database 2400 and answers corresponding thereto may be related to a specific field of expertise, for example, a tax field. However, the scope of the present invention is not limited thereto, and may be related to other specialized fields such as law and finance.

FIG. 8 is a diagram illustrating a specialized field response service method using a variable cycle auto-encoding method using a chatbot according to an embodiment of the present invention.

Referring to FIG. 8, in the specialized field response service method, first, a database 2400 of the specialized field is created (S100). As described above, the database 2400 stores answers to questions in a specific field of expertise, and the questions may be indexed (classified). In addition, with the creation of the database 2400, the index system 2200 learns the value of the latent variable using the variable cycle auto-encoder. For example, in the case of the first embodiment, a value of a latent variable may be learned through a training process of encoding and decoding a preset predetermined keyword set using a cyclic autoencoder including an LSTM. In the case of the second embodiment, the value of the latent variable is learned through a training process of encoding and decoding a preset predetermined keyword set using a first variable cyclic autoencoder including an LSTM, and a predetermined image is CNN. A value of a latent variable shared with a latent variable learned in the first variable cycle autoencoder may be learned through a training process of encoding and decoding using a second variable cycle autoencoder including.

Next, an input is received from the user (S110). The user can input a question about a specialized field in text or voice form or input in the form of an image through the user terminal. The input/output unit 2100 of FIG. 7 transmits the input question to the index system 2200.

Next, the morpheme analysis unit of the index system 2200 analyzes the morphemes of the received natural language to generate a keyword set corresponding to the natural language (S120). The morpheme analysis unit may generate a keyword set composed of one or more keywords excluding investigations that do not have a special meaning. This step is performed only when the user's input is in the form of text or voice, and is omitted in the case of an image form.

Next, the differential cycle autoencoder of the index system 2200 calculates a latent variable by encoding a keyword set generated by the morpheme analysis unit or an image input from a user (S130).

Next, the update unit of the index system 2200 updates the index value according to the value of the latent variable generated by the differential cycle autoencoder (S140).

Next, the search unit 2300 searches for an answer in the database 2400 according to the index value (S150).

Depending on the embodiment, the order of the index update step S140 and the search step S150 may be changed.

9 is a diagram showing the experimental results of the specialized field response service system and method of the differential cycle auto-encoding method according to the first embodiment of the present invention.

The data used in the experiment were more than 1000 sets of questions and expert answers, such as corporate tax, value added tax, and year-end settlement, in the knowledge IN Naver tax field. The learning rate was set to 0.001, the batch size was set to 32, and the latent space was trained in 2D and 3D, respectively.

In (a) and (b) of Fig. 9, blue represents corporate tax, red represents VAT, and green represents year-end settlement. As shown in FIG. 9, it can be seen that corporate tax, value added tax, and year-end settlement are clearly classified according to the value of the potential variable.

As described above, the present invention has been described in detail through preferred embodiments, but the present invention is not limited thereto, and various modifications and applications can be made within the scope of the technical spirit of the present invention. It is self-explanatory to the technician. Therefore, the true scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be interpreted as being included in the scope of the present invention.

Claims (12)

The first variation of calculating the value of the latent variable by learning the value of the latent variable through a training process of encoding and decoding a preset predetermined keyword set, and encoding a keyword set corresponding to the user's voice or text input based on this Variational Recurrent Autoencoder;
A second variable cycle autoencoder for calculating a value of a latent variable by learning a value of a latent variable through a training process for encoding and decoding a predetermined image, and encoding an image input of a user based thereon;
A result of comparing the value of the latent variable learned in the first and second differential cycle autoencoder and the value of the latent variable calculated by the first variable cycle autoencoder or the second variable cycle autoencoder An indexing unit for indexing an input from a user based on;
Including,
The index system of the differential cycle auto-encoding method, characterized in that the values of the latent variables learned by the first differential cycle autoencoder and the second differential cycle autoencoder are shared. The method of claim 1,
Keyword generator for generating a keyword set corresponding to the user's text or voice input by analyzing the morpheme of the user's text or voice input
Indexing system of the variable cycle auto-encoding method further comprising a. The method of claim 1,
The first differential circulation autoencoder,
A first encoding unit encoding the preset predetermined keyword set or a keyword set corresponding to the user's voice or text input; And
A first decoding unit for decoding a value of a latent variable generated by the first encoding unit encoding the preset keyword set
Indexing system of the variable cycle auto-encoding method comprising a. The method of claim 3,
The first encoding unit,
Includes LSTM (Long Short Term Memory) of the same number as the number of keywords constituting the input keyword set,
The keywords are respectively input to the LSTM in the order of input,
The LSTM is an index system of a variable cycle auto-encoding method, characterized in that connected in a chain form. The method of claim 1,
The second differential circulation autoencoder,
A second encoding unit encoding the predetermined image or the image input of the user; And
A second decoding unit for decoding a value of a latent variable generated by the second encoding unit encoding the predetermined image
Including,
Wherein the second encoding unit and the second decoding unit include a convolutional neural network (CNN). The method of claim 1,
An update unit for updating the index of the index unit by using the value of the latent variable calculated by the first cyclic autoencoder or the second cyclic autoencoder
An index system of a differential cycle auto encoding method further comprising a. The method of claim 1,
The preset predetermined keyword set relates to tax affairs,
The preset predetermined image is an image capable of distinguishing types of documents related to tax affairs. Learning the value of a latent variable through a training process of encoding and decoding a preset predetermined keyword set using a first differential cycle autoencoder, and encoding and decoding a preset predetermined image using a second differential cycle autoencoder Learning a value of a latent variable shared with the latent variable learned by the first variable cycle autoencoder through a training process to be performed;
Calculating a value of a latent variable corresponding to a user's input by using the first differential cycle autoencoder or the second differential cycle autoencoder; And
Indexing an input from a user based on a result of comparing the learned value of the latent variable with the calculated value of the latent variable;
Indexing method of the incremental cycle auto-encoding method including a. The method of claim 8,
Generating a keyword set corresponding to the user's text or voice input by analyzing the morpheme of the user's text or voice input
Including more,
In the calculating step,
When a user's input is text or voice, a value of a latent variable corresponding to the generated keyword set is calculated using the first differential cycle auto-encoder. The method of claim 8,
Updating the index used in the indexing step by using the value of the latent variable calculated by the first differential cycle autoencoder or the second differential cycle autoencoder
Indexing method of the differential cycle auto-encoding method, characterized in that it further comprises. The method of claim 8,
The preset predetermined keyword set relates to tax affairs,
The indexing method of the incremental cycle auto encoding method, characterized in that the preset predetermined image is an image capable of distinguishing types of documents related to tax affairs. A keyword generator for generating a keyword set corresponding to the user's natural language input by analyzing a morpheme of the user's natural language input;
A differential cycle autoencoder for calculating a value of a latent variable by learning a value of a latent variable through a training process of encoding and decoding a predetermined keyword set, and encoding the keyword set generated by the keyword generator based on this;
An index unit for indexing an input from a user based on a result of comparing the value of the latent variable learned by the differential cycle autoencoder with the value of the latent variable calculated by the differential cycle autoencoder; And
An update unit for updating the index of the index unit by using the value of the latent variable calculated by the differential cycle autoencoder
Indexing system of the variable cycle auto-encoding method comprising a.

Images