KR101886373B1 - Platform for providing task based on deep learning - Google Patents

Abstract

Deep Learning provides a task-providing platform based on artificial neural networks. A computer-implemented method includes extracting dialog information from at least one message application installed in an electronic device; Extracting task information from the conversation information using a CNN (convolutional neural network) learning model for natural language processing and a recurrent neural network (RNN) learning model for generating a sentence; And providing the task information through a task providing application installed in the electronic device.

Description

{PLATFORM FOR PROVIDING TASK BASED ON DEEP LEARNING}

The following description relates to platform services based on deep-running artificial neural networks.

Today, with the rapid development of communication network such as the Internet, messenger service through a communication network is becoming common.

Using the Internet, users can easily exchange messages with other users anywhere and anytime, and this messenger service has brought about many changes throughout our lives.

As a mobile messenger, a messenger service is being used as a popular communication tool that transcends borders as the mobile communication develops. The number of users is gradually increasing, and a messenger-based service is gradually increasing in line with this.

In this way, the messenger service is a service that allows users to communicate information in a conversational manner

As an example of a related art of the messenger service, Korean Patent Laid-Open No. 10-2004-0055027 (published on June 26, 2004) discloses a technique for implementing a messenger function in a portable terminal.

It provides a task providing platform that extracts and provides tasks (tasks) from the conversation contents sent and received by users based on DNN (deep neural network).

It provides a task-providing platform for extracting tasks through natural language process due to DNN.

Provides a learning algorithm that can accurately and effectively distinguish task-related words from non-task words in conversation contexts.

And provides a task visualization method that can visualize tasks extracted from conversation contents sent and received by the user in a form that the user can see.

A computer-implemented method, comprising: extracting dialog information from at least one message application installed in an electronic device; Extracting task information from the conversation information using a CNN (convolutional neural network) learning model for natural language processing and a recurrent neural network (RNN) learning model for generating a sentence; And providing the task information through a task providing application installed in the electronic device.

According to an aspect of the present invention, the step of extracting task information includes extracting a core word corresponding to the task related word in a sentence included in the conversation information through the CNN learning model in which task related words are learned; And generating a task sentence using the key word through the RNN learning model.

According to another aspect of the present invention, the step of extracting task information includes analyzing positive and negative expressions in a sentence included in the conversation information through the RNN learning model including an emotional analysis specialized learning model, Generating a sentence; And extracting a key word corresponding to the task related word from the confirmed sentence through the CNN learning model in which the task related word is learned.

According to another aspect, in the step of extracting the task information, a task is extracted from a sentence included in the conversation information through the CNN learning model in which a task related word is learned, ) As the attributes of the attribute information.

According to another aspect of the present invention, the step of extracting the task information may include analyzing the positive negation in the sentence included in the conversation information through the emotion analysis specialized learning model, and reflecting the positive negative analysis result in the task extraction.

According to another aspect of the present invention, the CNN learning model is a method for expressing words of learning data by a word vector through a word embedding method, then defining the task related word as an objective function, It is possible to learn by adjusting the distance of the embedding vector.

According to another aspect of the present invention, the CNN learning model can be learned by defining a word corresponding to a place, a time, and a purpose as an objective function as the task related word.

According to another aspect of the present invention, the CNN learning model and the RNN learning model are generated by adding a cache model to the learning data added when learning data is added, Lt; / RTI >

A computer program recorded on a computer-readable recording medium for executing a task providing method, the method comprising: extracting dialog information from at least one message application installed in an electronic device; Extracting task information from the dialog information using a CNN learning model for natural language processing and an RNN learning model for generating a sentence; And providing the task information through a task providing application installed in the electronic device.

What is claimed is: 1. A computer-implemented system comprising: at least one processor configured to execute computer-readable instructions, the at least one processor comprising: a dialog extractor for extracting dialog information from at least one message application installed in the electronic device; part; A task extracting unit for extracting task information from the conversation information using a CNN learning model for natural language processing and an RNN learning model for generating a sentence; And a task providing unit for providing the task information through a task providing application installed in the electronic device.

According to embodiments of the present invention, a task can be extracted and provided from conversation contents exchanged by a user based on a DNN.

According to embodiments of the present invention, a task can be accurately extracted through natural language processing with high reliability due to DNN.

According to the embodiments of the present invention, task-related words and non-task words can be accurately and effectively distinguished from conversation contents.

According to the embodiments of the present invention, a valid task can be accurately extracted by applying the emotion analysis specialized learning model and reflecting the result of affirmative negative determination in conversation to task extraction.

According to embodiments of the present invention, when a learning model for task extraction is added to learning data, a real-time learning process can be implemented by creating a cache model and merging the model with an existing model .

According to the embodiments of the present invention, the task can be extracted and provided as attributes of 5W1H (When, Where, Who, What, Why, How) among various conversation contents.

According to embodiments of the present invention, tasks extracted from conversation contents exchanged by a user can be visualized according to an optimized user interface (UI) / UX (user experience).

According to the embodiments of the present invention, tasks are automatically extracted and provided from conversation contents, thereby reducing the time for thinking and action for the user to judge or record the schedule without separately managing or memorizing the tasks due to conversation .

1 is a diagram illustrating an example of a network environment according to an embodiment of the present invention.
2 is a block diagram illustrating an internal configuration of an electronic device and a server according to an embodiment of the present invention.
FIG. 3 is a block diagram illustrating an example of a component included in a server providing a DNN-based platform service according to an exemplary embodiment of the present invention. Referring to FIG.
4 is a flowchart illustrating an example of a method that can be performed by a server providing a platform service based on DNN in an embodiment of the present invention.
5 is a diagram showing an example of an LSTM learning model for task processing in an embodiment of the present invention.
6 is an exemplary diagram for explaining a general deep learning reinforcement learning process.
7 is an exemplary diagram for explaining a real-time reinforcement learning processing process in an embodiment of the present invention.
FIG. 8 is a diagram illustrating an example of a CNN learning model for extracting task-related words in an embodiment of the present invention.
9 is an exemplary diagram for explaining a task sentence generation process using an LSTM model in an embodiment of the present invention.
FIG. 10 is an exemplary diagram for explaining a task extraction process reflecting an affirmative negative analysis result in an embodiment of the present invention. FIG.
11 is an exemplary diagram for explaining a process of extracting a task having a 5W1H attribute according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Embodiments of the present invention relate to a technology for providing a platform service based on a deep learning artificial neural network. More specifically, the present invention relates to a task providing platform capable of extracting and visualizing a task from a conversation message based on a DNN.

Embodiments including those specifically disclosed herein can implement a platform service based on a deep learning artificial neural network, thereby enabling a task to be extracted and visualized from troublesome conversation contents, and further, to improve efficiency, rationality, effectiveness, Cost savings, and the like.

1 is a diagram illustrating an example of a network environment according to an embodiment of the present invention. 1 shows an example in which a plurality of electronic devices 110, 120, 130, 140, a plurality of servers 150, 160, and a network 170 are included. 1, the number of electronic devices and the number of servers are not limited to those shown in FIG.

The plurality of electronic devices 110, 120, 130, 140 may be a fixed terminal implemented as a computer device or a mobile terminal. Examples of the plurality of electronic devices 110, 120, 130 and 140 include a smart phone, a mobile phone, a navigation device, a computer, a notebook, a digital broadcast terminal, a PDA (Personal Digital Assistants) ), And tablet PCs. For example, the electronic device 1 110 may communicate with other electronic devices 120, 130, 140 and / or the servers 150, 160 via the network 170 using a wireless or wired communication scheme.

The communication method is not limited, and may include a communication method using a communication network (for example, a mobile communication network, a wired Internet, a wireless Internet, a broadcasting network) that the network 170 may include, as well as a short-range wireless communication between the devices. For example, the network 170 may be a personal area network (LAN), a local area network (LAN), a campus area network (CAN), a metropolitan area network (MAN), a wide area network (WAN) , A network such as the Internet, and the like. The network 170 may also include any one or more of a network topology including a bus network, a star network, a ring network, a mesh network, a star-bus network, a tree or a hierarchical network, It is not limited.

Each of the servers 150 and 160 is a computer device or a plurality of computers that communicate with a plurality of electronic devices 110, 120, 130 and 140 through a network 170 to provide commands, codes, files, Lt; / RTI > devices. In one example, the server 160 may provide a file for installation of the application to the electronic device 1 (110) connected via the network 170. [ In this case, the electronic device 1 (110) can install an application using a file provided from the server (160). The server 150 is connected to the server 150 in accordance with an operating system (OS) included in the electronic device 1 110 or under control of at least one program (for example, a browser or an installed application) I can receive contents. For example, when the electronic device 1 (110) transmits a service request message to the server 150 via the network 170 under the control of the application, the server 150 transmits a code corresponding to the service request message to the electronic device 1 The first electronic device 110 can provide contents to the user by displaying and displaying a screen according to the code according to the control of the application.

2 is a block diagram illustrating an internal configuration of an electronic device and a server according to an embodiment of the present invention. In FIG. 2, an internal configuration of the electronic device 1 (110) as an example of one electronic device and the server 150 as an example of one server will be described. Other electronic devices 120, 130, 140 or server 160 may have the same or similar internal configurations.

The electronic device 1 110 and the server 150 may include memories 211 and 221, processors 212 and 222, communication modules 213 and 223 and input / output interfaces 214 and 224. The memories 211 and 221 may be a computer-readable recording medium and may include a permanent mass storage device such as a random access memory (RAM), a read only memory (ROM), and a disk drive. In addition, the operating system or at least one program code (for example, a code for a browser or a dedicated application installed in the electronic apparatus 1 (110) and running) may be stored in the memories 211 and 221. [ These software components may be loaded from a computer readable recording medium separate from the memories 211 and 221. [ Such a computer-readable recording medium may include a computer-readable recording medium such as a floppy drive, a disk, a tape, a DVD / CD-ROM drive, and a memory card. In other embodiments, the software components may be loaded into memory 211, 221 via communication modules 213, 223 rather than a computer readable recording medium. For example, at least one program may be a program installed by a file distribution system (for example, the server 160 described above) that distributes installation files of developers or applications, May be loaded into the memory 211, 221 based on the application described above.

Processors 212 and 222 may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input / output operations. The instructions may be provided to the processors 212 and 222 by the memories 211 and 221 or the communication modules 213 and 223. For example, the processor 212, 222 may be configured to execute a command received in accordance with a program code stored in a recording device, such as the memory 211, 221.

The communication modules 213 and 223 may provide functions for the electronic device 1 110 and the server 150 to communicate with each other through the network 170 and may provide functions for communicating with other electronic devices (for example, the electronic device 2 120) Or to communicate with another server (e.g., server 160). The request generated by the processor 212 of the electronic device 1 110 according to the program code stored in the recording device such as the memory 211 is transmitted to the server 170 via the network 170 under the control of the communication module 213 150 < / RTI > Conversely, control signals, commands, contents, files, and the like provided under the control of the processor 222 of the server 150 are transmitted to the communication module 223 of the electronic device 110 via the communication module 223 and the network 170 213 to the electronic device 1 (110). For example, the control signal or command of the server 150 received through the communication module 213 may be transmitted to the processor 212 or the memory 211, May be stored as a storage medium that may further include a < RTI ID = 0.0 >

The input / output interface 214 may be a means for interfacing with the input / output device 215. For example, the input device may include a device such as a keyboard or a mouse, and the output device may include a device such as a display for displaying a communication session of the application. As another example, the input / output interface 214 may be a means for interfacing with a device having integrated functions for input and output, such as a touch screen. More specifically, the processor 212 of the electronic device 1 (110) uses the data provided by the server 150 or the electronic device 2 (120) in processing commands of the computer program loaded in the memory 211 A service screen or contents can be displayed on the display through the input / output interface 214. [

Also, in other embodiments, electronic device 1 110 and server 150 may include more components than the components of FIG. However, there is no need to clearly illustrate most prior art components. For example, electronic device 1 110 may be implemented to include at least a portion of input / output devices 215 described above, or may be implemented with other components such as a transceiver, Global Positioning System (GPS) module, camera, Elements. More specifically, when the electronic device 1 (110) is a smart phone, it may be an acceleration sensor, a gyro sensor, a camera, various physical buttons, buttons using a touch panel, input / output ports, It is to be understood that the present invention may be embodied in various forms without departing from the spirit or scope of the invention.

Hereinafter, a specific embodiment of a DNN-based platform service method and system will be described.

FIG. 3 is a configuration diagram illustrating an example of a component that can be included in a server providing a DNN-based platform service according to an exemplary embodiment of the present invention. FIG. Fig. 2 is a flowchart showing an example of a method that can be performed.

The server 150 serves as a platform for automatically extracting a task from a conversation message based on the DNN and visualizing the task.

3, the processor 222 of the server 150 may include a dialogue extracting unit 301, a task extracting unit 303, and a task providing unit 305 as components. The server 150 basically includes the components described with reference to FIG. 2, further serves as a big data platform for providing a DNN-based platform service, and stores the dialog information extracted from the dialogue extracting unit 301 And a task information database 304 for storing task information extracted from the task extraction unit 303. The task information database 304 may include a task information database 304 for storing task information extracted from the task extraction unit 303,

The components of the processor 222 and the processor 222 may control the server 150 to perform the steps S410 through S430 included in the method of FIG. At this time, the components of the processor 222 and the processor 222 may be implemented to execute instructions according to the code of the operating system and the code of at least one program that the memory 221 contains. In addition, components of processor 222 may be representations of different functions performed by processor 222 in accordance with control commands provided by the operating system or by at least one program. For example, the dialog extraction unit 301 may be used as a functional expression in which the processor 222 extracts dialog information according to the control command described above.

As shown in FIG. 3, the platform service provided by the server 150 is interlocked with a medium for exchanging conversations between users via the Internet. In this case, the medium means a message application 10 providing a service for exchanging messages in various formats such as text, voice, and moving pictures, and may be an instant messenger, a social network service (SNS), an e-mail, message service, and a multi-media message service (MMS). For example, it is natural that the message application 10 can be various applications such as Kakao Talk, LINE, NATE ON, FACE BOOK, and LINKED IN. The server 150 closely communicates information with the message application 10 and can provide a separate service to the user. In other words, the server 150 summarizes the task extracted from the message application 10 and provides information corresponding thereto can do.

In step S410, the dialogue extracting unit 301 extracts the dialogue extracting unit 301 from at least one message application (for example, messenger A, messenger B, SNS C, etc.) 10 installed in the electronic device 110, Can be extracted and stored in the conversation information database 302. In other words, the dialogue extracting unit 301 extracts, from the message application 10, the information of the talker communicating with the user using the developer tool (SDK) provided by the manufacturer of the application, The conversation information can be extracted. At this time, the dialog information extracted from the message application can be traced to the dialog information database 302, which is a big data platform of a heterogeneous and distributed system.

In step S420, the task extraction unit 303 extracts the conversation information 302 stored in the conversation information database 302 using a CNN (convolutional neural network) learning model for natural language processing and a recurrent neural network (RNN) And store the extracted task information in the task information database 304. [ The task extraction unit 303 is a combination of a CNN algorithm for extracting words related to a task (hereinafter referred to as task-related words) from conversation contents through natural language processing and an RNN algorithm for generating task sentences from conversation contents So that task information can be extracted from the conversation information. The task extracting unit 303 extracts task-related words through an RNN learning model or an algorithm that generates a task sentence through an RNN learning model after a line learning through a CNN learning model or generates a task sentence using a CNN learning model Type algorithm can be applied. In the present invention, the learned DNN algorithm for task information extraction can extract and organize tasks as attributes of 5W1H (When, Where, Who, What, Why, How). The task information extracted from the conversation information can be traced back to the task information database 304, which is a big data platform of a heterogeneous and distributed system. A learning model specialized for task information extraction and a specific task information extraction process will be described below.

In step S430, the task provisioning unit 305 can provide task information to the user of the electronic device 1 (110) by visualizing the task information through the task providing application installed in the electronic device 1 (110). The task providing unit 305 learns the tasks tracked in the task information database 304 and the conversation information extracted from the message application in real time, and can perform a process for real time displaying the task in the task providing application. At this time, the task providing application may be configured as a PC-based program or an application dedicated to the mobile terminal. In addition, the task providing application may be implemented in a program form that operates independently, or in an in-app form of a specific application (e.g., a message application) so as to be able to operate on the specific application . In other words, the task provisioning unit 305 can display tasks to be processed in real time according to the optimized UI / UX through a task providing application provided in the form of a web service or a mobile terminal dedicated application.

According to the present invention, a natural learning process and a task extraction algorithm can be applied, which are implemented by a proprietary general artificial intelligence algorithm based on a deep learning basis capable of thinking the same as a human being. Natural language processing technology uses the analysis technique of word expression to analyze the sentences and phrases by learning the relation between words in DNN, and by combining emotional analysis algorithm, it is possible to detect and predict not only negative negatives of sentences, It is a proprietary algorithm. Opinion mining is a technology that provides useful information by analyzing conversation contents. It is a technology that can analyze the text and change the sensibility and opinions of the user who created the text by statistical / numerical analysis into objective information. By applying this emotion analysis algorithm, it is possible to analyze affirmative and negative expressions in conversation contents exchanged through the message application and reflect them in task extraction. The task extraction algorithm is an algorithm that learns vast amount of high quality data based on its own natural language processing technology and automatically learns the task dictionary by automatically determining the probability of task establishment in conversation contents.

The construction process of the learning model specialized for extracting task information will be described as follows.

The server 150 is configured to establish a learning model for extracting tasks by users of each of the electronic devices 110, 120, 130 and 140 through a message application installed in the plurality of electronic devices 110, 120, 130 and 140 (Not shown), which is a component that can be included in the processor 222 of the server 150, can collect the conversation message as the learning data, A learning model specialized for information extraction can be constructed.

1. The learning module can perform vectorization of words by expressing words in the learning data as word vectors.

end. The learning module can apply the word embedding method specialized for task distribution using the skip-gram (N-gram) model of the machine learning model word2vec. For example, in the learning sentence, The corresponding word, the word corresponding to the time, the word corresponding to the purpose, and the like.

First, the process of decentralizing a place in a learning sentence is described as follows.

(1) As an example, let's say "Let's have a beer in Gangnam" as a learning sentence.

(2) Matches a predetermined number of noise samples from the noise distribution corresponding to the P (w) distribution, which is a unigram distribution (unigram distribution).

(3) When the set noise number is set to 1, it is assumed that the word 'at' is selected as a noise word in the learning sentence.

(4) In the learning phase t, the objective function is defined as shown in Equation (1), and then the first learning is performed.

[Equation 1]

는 임베딩 벡터 θ를 학습하면서 데이터셋 D에서 컨텍스트 h에서 단어 w가 나올 확률을 계산하는 이진 로지스틱 회귀분석 확률(binary logistic regression probability) 모델을 의미하고,

는 잡음분포에서 k개의 노이즈 단어를 의미한다.here,

Denotes a binary logistic regression probability model that calculates the probability that a word w occurs in the context h in the data set D while learning the embedding vector θ,

Means the k noise words in the noise distribution.

(5) Update the embedding factor θ to maximize the objective function.

(θ값에 해당하는 목적함수의 미분값)를 산출한다.(6) To derive a gradient for the loss of the embedding factor [theta]

(the derivative value of the objective function corresponding to the? value).

(7) Adjust the distance of the embedding vectors until the distinction between the task-related word ('Gangnam', which is the place in the learning sentence) and the non-task word (in the learning sentence, do. By adjusting the distance of the embedding vector, it is possible to obtain a vector close to 'place'.

The objective and time can be decentralized in the learning sentence in the above-mentioned (1) to (7).

I. The learning module can adjust the range and slope of the task related words. Table 1 shows the range of clusters of task related words and examples of slope adjustment.

class TaskPredictor
def _init_ (self):
self.vectorizer = Tfidvectorizer ()

def tran (self, data):
self.vectorizer.fit (np.append (data.context.values, data.Utterance.values))

def predict (self, context, utterances):
vector_context = self.vectorizer.transform ([context])
vector_doc = self f.vectorizer.transform (utterances)

result = np.dot (vector_doc, vector_context.T) .todense ()
result = np.asarray (result) .flatten ()

return np.argsort (result, axis = 0) [:: - 1}

2. Learning Algorithm

end. The learning module can apply the CNN algorithm for natural language processing, and the CNN learning model has the following structure.

(1) List the vector values (word2vec) of the words in the learning sentence as a matrix in the input layer of the CNN learning model.

(2) A convolutional layer with multiple filters is connected to a max-pooling layer.

(3) Select the softmax classifier as the classification layer.

I. The learning module can apply the following RNN algorithm for sentence generation. The RNN learning model has the following structure.

(1) The RNN learning model uses a long-term short-term memory (LSTM) model (see FIG. 5) for task processing.

(2) The RNN learning model can determine positive or negative in the flow of the context of the learning sentence.

&Quot; (2) "

Where h is a vector value representing the output of the hidden layer, W is a weight vector value for attaining attention, b is a bias, e is an exponential function, a is a negative negligible weight, which is a weight for h when a context vector is obtained.

All. The learning module can apply real-time reinforcement learning processing to learning data.

(1) The general deep learning reinforcement learning process is to generate the learning model 61 (CNN learning model, RNN learning model) by learning the learning data as shown in Fig. 6, The learning model 61 is deleted and the entire learning data including the new data is again learned to generate a new learning model 63. [

(2) The real-time reinforcement learning process is a method of creating a cache model and merging it with an existing learning model, and real-time learning processing can be implemented. 7, a learning model 71 is generated by learning learning data. Then, when learning data is added, a cache model 73 is generated for the added learning data, and the learning model is merged with the existing learning model 71 The learning model 75 can be generated. Therefore, by implementing the real-time learning process and the reinforcement learning process for the learning model by utilizing the cache model, it is possible to reduce the resources and the cost for constructing the learning model.

As a specific task information extraction process, a learning algorithm for extracting task information is as follows.

1. The task extraction unit 303 can apply an algorithm (CNN + RNN learning algorithm) that generates a task sentence through an RNN learning model after a line learning through a CNN learning model.

end. The task extraction unit 303 can extract a task-related word that is a core word in the sentence using the CNN learning model (see FIG. 8).

(1) Embed a word contained in a sentence as a vector value.

(2) A feature map is generated by performing a composite multiplication on three sentence matrices with three filters (sets of row lengths of 2, 3 and 4, respectively).

(3) Max-pooling of each map is performed to extract the maximum value.

(4) Connect the six univariate feature vectors (max-pooling result values) to the feature vectors of the classification layer (softmax classifier).

(5) Extract task related words (place, purpose, time words) through classification layer (softmax classifier).

I. The task extraction unit 303 can generate a task sentence using the LNT learning model LSTM. The task extraction unit 303 can learn the words selected through the CNN learning model from the input data of the RNN learning model, and can also perform word correction processing such as spoken language and the like. FIG. 9 is a diagram illustrating an RNN learning model in which words selected through the CNN learning model are input data.

(1) Word representation Layer: The vector values of words selected by CNN are arranged in time series (including spoken and written words).

(2) sentence composition layer: regenerates a sentence using the LSTM algorithm for the vector value input from the word expression layer.

(3) Sentence representation layer: Places the sentence generated in the sentence structure layer.

(4) Document composition Layer: Replaces non-learning data (spoken word, etc.) in the generated sentence with words of the closest meaning in the learned conversation (identified by a combination of vector values of similar sentences) .

(5) Document representation Layer: Select the final sentence using the softmax function for the sentences containing the replaced word.

2. The task extraction unit 303 may apply an algorithm (RNN + CNN learning algorithm) that generates a task sentence in the CNN learning model after extracting task related words through the RNN learning model.

end. The task extraction unit 303 can summarize the conversation flow using the RNN learning model.

(1) The task extraction unit 303 can grasp the negative or negative in the sentence by applying the emotion analysis specialized learning model. The emotion analysis specialized learning model can be defined as Equation (3).

&Quot; (3) "

는 시그모이드(sigmoid) 함수, x_t는 이전 시퀀스의 입력, h_t-1은 이전 시퀀스의 출력, W_z, U_z, W_r, U_r, W_m, U_m은 각각 게이트(gate)와 셀 메모리(cell memory)를 위한 가중치 행렬, r은 이전 상태를 유닛의 입력에 반영하는 비율을 결정하는 리셋 게이트(reset gate), z는 이전 상태를 보존하여 반영하는 정도를 결정하는 업데이트 게이트(update gate), act는 활성화(activation) 함수,

는 요소별 결과물(element-wise product)을 의미한다.here,

Is sigmoid (sigmoid) function, x _t is input, h _t-1 of the previous sequence is output from the previous _{sequence, W z, U z, W} r, U r, W m, U m are each gate (gate) And r is a reset gate for determining the ratio of the previous state to the input of the unit, and z is an update gate for determining the extent to which the previous state is preserved and reflected update gate), act is an activation function,

Means an element-wise product.

(2) The task extraction unit 303 can determine a range in which a negative word affects the context.

(3) Referring to FIG. 10, the task extraction unit 303 may generate a task confirmation sentence by combining task related words including affirmative expression words, excluding negative expression words among task related words. In other words, affirmative expression words are established as task related words, but negative expression words are excluded from task related words.

I. The task extracting unit 303 can extract the task related words as attributes of 5W1H (When, Where, Who, What, Why, How) using the CNN learning model and calculate the task probability to determine the final task sentence. 11 is a diagram illustrating a CNN learning model for extracting a task with an attribute of 5W1H.

(1) List the vector values of each word in a sentence as a matrix. In FIG. 11, it is shown in a six-dimensional form, but this is for convenience of the city, and actually has a shape of more than 100 dimensions.

(2) Perform a convolutional transformation using multiple filters.

(3) Max-pooling is performed by using the 5W1H feature map.

(4) The task probability is calculated through max-pooling.

For example, the task extracting unit 303 can determine a sentence having a task probability greater than or equal to a set value among the task confirmation sentences as the final task sentence by applying the CNN learning model to each of the task confirmation sentences generated through the RNN learning model .

Examples of learning algorithms to be applied in the present invention include Hadoop + Hbase, Spark + Word2vec, Tensorflow + Theano + Kerar, and the like.

In summary, a task extraction method for a DNN-based platform service according to the present invention may include a data collection step, a data normalization step, a data classification step, and a pattern mining step.

The data collection step is a process of accessing the message application used by the user and collecting all available data including the conversation contents exchanged through the message application.

In the data normalization step, the text is quantified using Word2Vec, one of the functions of Spark MLlib, and the quantified data is converted into a vector value (numerical value). In other words, the word representation is converted into a value that can be represented on the vector space using artificial neural networks. Word2Vec is an algorithm to determine the proximity by looking at the relation between words. It is a non-bipod learning algorithm through deep learning. It is easy to extract new words, spoken words, swallow words, etc. It is advantageous in the point of being.

The data classification step classifies data having a vector tendency similar to a common data set. That is, data having a similar pattern and tendency are clustered and classified. For example, in a sentence, it is possible to group words corresponding to place, purpose, and time.

The pattern mining step is a process of recognizing the same vector pattern frequently occurring in a common data cluster. Based on the recognized pattern, information for task extraction can be learned from the conversation contents.

The natural language processing technology can be applied to the above algorithm to extract the task from the conversation contents. First, we learn massive data based on natural language processing technology. Next, the task dictionary is automatically constructed and learned by judging the probability of task establishment by itself in the dialogue contents. Through this learning model, a task can be extracted from conversation contents on various message applications used by the user. The natural language processing part may correspond to an algorithm that skips the coordination step of task words in word2vec and goes through the correction step of non-learning data (hermaphroditic, spoken word, etc.) for all words through RNN.

As described above, according to the embodiments of the present invention, it is possible to extract and visualize a task from the conversation contents exchanged by the user based on the DNN, Time of thought and action can be shortened. According to embodiments of the present invention, task-related words and non-task words can be accurately and effectively distinguished by extracting tasks from conversation contents by applying a natural language processing technique with high reliability. In addition, according to embodiments of the present invention, a valid task can be accurately extracted by applying the emotion analysis specialized learning model to the result of affirmative negative determination in conversation to the task extraction. Furthermore, according to embodiments of the present invention, when a learning model for task extraction is added to learning data, a cache model is generated and merged with an existing model to perform real-time learning processing Implementation can reduce the resources and costs required to build a learning model.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the apparatus and components described in the embodiments may be implemented as a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit, a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device As shown in FIG. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. At this time, the medium may be a program that continuously stores a computer executable program, or temporarily stores the program for execution or downloading. Further, the medium may be a variety of recording means or storage means in the form of a combination of a single hardware or a plurality of hardware, but is not limited to a medium directly connected to any computer system, but may be dispersed on a network. Examples of the medium include a magnetic medium such as a hard disk, a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floptical disk, And program instructions including ROM, RAM, flash memory, and the like. As another example of the medium, a recording medium or a storage medium that is managed by a site or a server that supplies or distributes an application store or various other software is also enumerated.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (10)

In a computer implemented method,
Extracting dialog information from at least one message application installed in the electronic device;
Extracting task information from the conversation information using a CNN (convolutional neural network) learning model for natural language processing and a recurrent neural network (RNN) learning model for generating a sentence; And
Providing the task information through a task providing application installed in the electronic device
Lt; / RTI >
The step of extracting the task information includes:
Analyzing a positive expression and a negative expression in a sentence included in the conversation information through the RNN learning model including a sensitivity analysis specific learning model to generate a confirmed sentence excluding the negative expression; And
Extracting a key word corresponding to the task related word from the confirmed sentence through the CNN learning model in which a task related word is learned
≪ / RTI > The method according to claim 1,
The step of extracting the task information includes:
Extracting a key word corresponding to the task related word in a sentence included in the conversation information through the CNN learning model in which a task related word is learned; And
Generating a task sentence using the key word through the RNN learning model
≪ / RTI > delete The method according to claim 1,
The step of extracting the task information includes:
Extracting a task as a 5W1H (When, Where, Who, What, Why, How) attribute in a sentence included in the conversation information through the CNN learning model in which a task related word is learned
≪ / RTI > In a computer implemented method,
Extracting dialog information from at least one message application installed in the electronic device;
Extracting task information from the conversation information using a CNN (convolutional neural network) learning model for natural language processing and a recurrent neural network (RNN) learning model for generating a sentence; And
Providing the task information through a task providing application installed in the electronic device
Lt; / RTI >
The step of extracting the task information includes:
The emotional analysis analysis model is used to analyze positive and negative statements in sentences included in the conversation information to reflect positive negative analysis results in task extraction
≪ / RTI > 6. The method according to claim 1 or 5,
In the CNN learning model,
The word of the learning data is represented by a word vector through a word embedding method and then the task related word is defined as an objective function and the distance of the embedding vector is adjusted for classifying the task related word and the non- that
≪ / RTI > The method according to claim 6,
In the CNN learning model,
Learning by defining the words corresponding to the task, time, and purpose as the task related words as objective functions
≪ / RTI > In a computer implemented method,
Extracting dialog information from at least one message application installed in the electronic device;
Extracting task information from the conversation information using a CNN (convolutional neural network) learning model for natural language processing and a recurrent neural network (RNN) learning model for generating a sentence; And
Providing the task information through a task providing application installed in the electronic device
Lt; / RTI >
Wherein the CNN learning model and the RNN learning model include:
When the learning data is added, a cache model is created for the added learning data, and then a reinforcement learning model is generated by merging with the existing learning model
≪ / RTI > A computer program recorded on a computer-readable recording medium for executing a task providing method,
The task providing method includes:
Extracting dialog information from at least one message application installed in the electronic device;
Extracting task information from the dialog information using a CNN learning model for natural language processing and an RNN learning model for generating a sentence; And
Providing the task information through a task providing application installed in the electronic device
Lt; / RTI >
The step of extracting the task information includes:
The emotional analysis analysis model is used to analyze positive and negative statements in sentences included in the conversation information to reflect positive negative analysis results in task extraction
And a computer program product. In a computer implemented system,
At least one processor configured to execute computer readable instructions,
Lt; / RTI >
Wherein the at least one processor comprises:
A dialogue extractor for extracting dialogue information from at least one message application installed in the electronic device;
A task extracting unit for extracting task information from the conversation information using a CNN learning model for natural language processing and an RNN learning model for generating a sentence; And
A task providing unit for providing the task information through a task providing application installed in the electronic equipment;
Lt; / RTI >
The task extracting unit,
The emotional analysis analysis model is used to analyze positive and negative statements in sentences included in the conversation information to reflect positive negative analysis results in task extraction
Lt; / RTI >

Images