KR102145777B1 - Method and apparatus of object control based on self-verification type analysis platform for emotion and weather using artificial intelligence - Google Patents

Abstract

A method for controlling a device based on self-verification type emotion analysis using artificial intelligence comprises: a data collection step in which unstructured data including at least one of user body data, environmental data which can affect a user, and text data related to the user is collected; a data processing step in which the combination of two or more emotional inference classifiers and environmental variable control classifiers is generated by a learning model generation step through analysis of the unstructured data using artificial intelligence; a learning model verification step in which the optimal combination of the emotional inference classifier and the environmental variable control classifier is finally selected through evaluation and verification using the artificial intelligence from the generated combination of the two or more emotional inference classifiers and the environmental variable control classifier; and a classifier driving step in which the environmental variable control classifier is driven according to an emotion classification value generated by driving the emotion inference classifier of the optimal combination, and a device operation command for a target device to be controlled is output.

Description

Device control method and device based on self-verification type sentiment analysis using artificial intelligence {METHOD AND APPARATUS OF OBJECT CONTROL BASED ON SELF-VERIFICATION TYPE ANALYSIS PLATFORM FOR EMOTION AND WEATHER USING ARTIFICIAL INTELLIGENCE}

The present invention relates to a device control method and apparatus based on self-verification type sentiment analysis using artificial intelligence, and in more detail, user's emotion or body data, and local and global environment data that can affect the user. , Generating multiple emotional reasoning classifiers and environmental variable control classifiers through analysis using artificial intelligence on unstructured data such as text data related to users, self-verifying and finally selecting them, and driving these classifiers to control target devices. It relates to a method and apparatus for controlling.

Recently, technologies that produce useful results using artificial intelligence, such as machine learning (machine learning), have emerged.

For example, in Patent Document 1,'first data in which a plurality of first body composition data collected by magnetic resonance imaging (MRI) using radiation, dual-energy X-ray absorptiometry (DEXA), or computed tomography (FATCT) is stored. A storage unit 10; A second data storage unit 20 for storing a plurality of second body composition data collected by the BIA (Bioelectrical Impedance Analysis) type body composition measuring device; Each of a plurality of first body composition data stored in the first data storage unit 10 and a plurality of second body composition data stored in the second data storage unit 20 are used as parameters of gender, age, weight, height, and race. A parameter classifying unit 30 for classifying into; And an algorithm generator 40 for generating a body composition measurement model algorithm by using a machine learning method in comparing the first body composition data and the second body composition data classified by parameters. A body composition measurement system' is disclosed.

In addition, in Patent Document 2,'when a mobile application installed in a mobile communication terminal is executed by a user, the mobile application is first text-based messages generated by the user from the memory of the mobile communication terminal and at least one Extracting second text-based messages generated by the user from a database; Inferring sentiment information for the user from the extracted first text-based messages and the extracted second text-based messages by the mobile application; And the mobile application sends the inferred emotion information to the management server so that the management server for controlling the operation of the smart home appliance can provide personal service information corresponding to the emotion information inferred by the mobile application to the smart home appliance. Disclosed is a method for providing a service based on emotion information including the step of transmitting.

Patent Document 1: Patent Publication 10-2018-0038251 Patent Document 2: Patent Publication 10-2014-0107723

However, the user's emotions (hereinafter, referred to as'feelings' including'feelings') are determined and changed by various factors. In other words, in order to accurately determine emotion, various factors must be analyzed.

Data that affect emotions include text, climate, indoor/outdoor environment, financial situation, amount of activity, sleep time, and social issues (news). The data that allows you to guess your emotions include heart rate, consumption history, social media posts that posted your own opinions, news comments that posted your own opinions, facial expressions, and pitch.

Since the technique of Patent Document 1 is a technique for analyzing body composition data, determination of sensitivity may not be possible. In addition, since the data to be compared are the first body composition data and the second body composition data, and these are both'body composition' data such as bone density and fat percentage, they are typically defined quantitative data, and the first and second two data are It is data with complete homogeneity. Therefore, there is no difficulty in comparing each other to the extent that there is no need to mobilize artificial intelligence. In addition, machine learning methods are only being used to compare two data and generate body composition measurement model algorithms. Therefore, a technique in which a machine learning method is used for a more advanced function, such as a function of generating a classifier constituting a learning model, is neither disclosed nor suggested.

The technology of Patent Document 2 is a technology that infers sensibility through text and enables indoor devices to be manipulated according to sensibility. However, the data that influences human emotions and allows them to guess emotions are very diverse. In other words, text is just one kind, which is only a small part of it. Therefore, the method of inferring emotion through text alone, as described in Patent Document 2, has a limitation in accurately inferring emotion.

The present invention is to solve the problems of the prior art, and collects various data surrounding emotions, and through machine learning, it is possible to infer more precise emotions by grasping relationships that humans have not thought of, using artificial intelligence. It is to provide a device control method and apparatus based on self-verification type emotion analysis.

In addition, for self-verification type sentiment analysis using artificial intelligence, which controls the target device based on unstructured data such as user's emotion or body data, local and global environmental data that can affect the user, and text data related to the user. It is to provide a device control method and apparatus based. In the present invention, the reasons for the various data collection modules and the reason for the existence of the emotional reasoning classifier are that the data affecting human emotions and allowing them to guess the emotions are very diverse, and their shapes are different in various ways. Because.

And it is to provide a device control method and apparatus based on self-verification type sentiment analysis using artificial intelligence that generates an emotional reasoning classifier and an environmental variable control classifier through an analysis of such unstructured data using artificial intelligence.

In addition, it is to provide a device control method and apparatus based on self-verifying emotion analysis using artificial intelligence that can infer emotional states through correlations of heterogeneous data.

And it is to provide a device control method and apparatus based on self-verification type sentiment analysis using artificial intelligence to control the target device of control by driving the finally selected emotional reasoning classifier and environmental variable control classifier.

In addition, another object of the present invention is to further verify the classifier generated by artificial intelligence using a self-validation type analysis that borrows a genetic algorithm, to analyze the emotion more accurately, and based on this, using artificial intelligence. It is to provide a device control method and apparatus based on self-verification type emotion analysis.

Without the application of genetic algorithms, if unstructured data is preprocessed by known knowledge, and an artificial intelligence constructs a classifier based on this preprocessed result value, it is absolutely affected by the result value and can show a biased result. In other words, the criterion for how many more pre-processing models and result values can be created is as important as how artificial intelligence effectively constructs a classifier using data preprocessed values.

In the present invention, using a self-verification type analysis device borrowing a genetic algorithm, various preprocessing models for processing big data affecting emotions and result values thereof are generated as a combination of various machine learning taxa It is possible to find a method that goes beyond this previously unconceived knowledge.) Through this, it is possible to more effectively find a classifier device that infers emotions and controls the environment.

The method of the present invention for achieving the above object is a device control method based on self-verification type sentiment analysis using artificial intelligence, and data that can affect the user's emotional state and data that allows the user's emotion to be estimated. A data collection step of collecting structured or unstructured data including at least one of; A data processing step of generating a combination of two or more emotional inference classifiers and environmental variable control classifiers by a learning model generation step through analysis of the structured or unstructured data using artificial intelligence; A learning model verification step of finally selecting an optimal combination of the emotional inference classifier and the environmental variable control classifier through evaluation and verification using artificial intelligence from the generated combination of the two or more emotional inference classifiers and the environmental variable control classifier; And a classifier driving step of driving the environmental variable control classifier according to the emotion classification value generated by driving the emotion inference classifier of the optimal combination, and outputting a device operation command for the device to be controlled. To do.

Here, in the data processing step, a data pre-processing step of performing pre- processing by two or more pre-processing modules on the unstructured data prior to generation of the combination of the two or more emotional reasoning classifiers and the environmental variable control classifier is further provided, and the learning model In the verification step, if the evaluation score by the evaluation function is more than the standard value, the optimal combination of the emotional reasoning classifier and the environmental variable control classifier is finally selected, and if the evaluation score by the evaluation function is less than the standard value, the emotional reasoning classifier and It is preferable that a variant of the parameter of the combination of the environmental variable control classifier is generated, and after evaluation by the directional predictive classifier for the variant, the variant is fed back to the data preprocessing step when the evaluation result is not impossible to improve.

In addition, the directional predictive classifier may be generated from a combination of the emotional inference classifier and the environmental variable control classifier, parameters, and evaluation scores by a directional machine learning engine.

In addition, the unstructured data collected in the data collection step may be of a plurality of types , and in the data processing step, the emotional state may be inferred through the correlation between the heterogeneous unstructured data.

And, in the data processing step, prior to the generation of the combination of the two or more emotional reasoning classifiers and the environmental variable control classifier, a data pre-processing step of pre- processing the unstructured data by two or more pre-processing modules is further provided, and the two or more The combination of the emotional reasoning classifier and the environmental variable control classifier is each formed through pre-processing using the two or more pre-processing modules, and when the optimal combination is finally selected in the learning model verification step, the optimal pre-processing module used for the pre-processing of the optimal combination It may be made to be selected.

In addition, in the classifier driving step, a data preprocessing driving step in which the unstructured data is preprocessed by the optimal preprocessing module may be further provided.

In addition, the output value generated by the operation of the target device may be fed back to the data collecting step and collected together with the unstructured data.

On the other hand, the device of the present invention for achieving the above task is a device control device based on self-verification type sentiment analysis using artificial intelligence, so that data that can affect the user's emotional state and the user's emotion can be estimated. A data collection module configured to collect structured or unstructured data including at least one of the data ; A data processing module for generating an emotional inference classifier and an environmental variable control classifier by a learning model generation step through an analysis of the unstructured data using artificial intelligence; The environmental variable control classifier is driven according to the emotion classification value generated by driving the emotion inference classifier, and a classifier driving module configured to output a device operation command for a device to be controlled.

In addition, the program of the present invention for achieving the above object is an information processing device in which a program for performing the steps constituting a device control method based on self-verification type sentiment analysis using artificial intelligence described above is recorded in the information processing device. It is characterized by consisting of a program recorded on a readable recording medium.

According to the present invention, a device control method based on self-verification type sentiment analysis using artificial intelligence, capable of inferring more precise emotions by collecting various data surrounding emotions and grasping relationships that humans have not thought of through machine learning. And an apparatus is provided.

In addition, for self-verification type sentiment analysis using artificial intelligence, which controls the target device based on unstructured data such as user's emotion or body data, local and global environmental data that can affect the user, and text data related to the user. A device control method and apparatus based on it are provided.

In addition, a device control method and apparatus based on self-verification type sentiment analysis using artificial intelligence, which generates an emotional reasoning classifier and an environmental variable control classifier through an analysis of such unstructured data using artificial intelligence, are provided.

In addition, a device control method and apparatus based on self-verifying emotion analysis using artificial intelligence, capable of inferring an emotional state through correlation of heterogeneous data, is provided.

In addition, a device control method and apparatus based on self-verification type sentiment analysis using artificial intelligence is provided to control a device to be controlled by driving a finally selected emotional reasoning classifier and environmental variable control classifier.

And according to the present invention, the classifier generated by artificial intelligence is additionally verified using a self-validation type analysis that borrows a genetic algorithm, and a self-validation using artificial intelligence is used to analyze the emotion more accurately. A device control method and apparatus based on type sentiment analysis are provided.

Since the criterion for how many more pre-processing models and result values can be created is as important as how artificial intelligence effectively constructs a classifier using the pre-processed values of data, according to the present invention, according to the present invention, By using a verification-type analysis device, various preprocessing models that process big data that affect emotions and the resulting values are randomly generated (therefore, it is possible to find a method beyond knowledge that humans have not previously thought of) Through this, it is possible to more effectively find a classifier device that infers emotions and controls the environment.

1 is an exemplary block diagram of an apparatus in which a device control method based on self-verification type sentiment analysis using artificial intelligence according to an embodiment of the present invention can be implemented.
2 is a block diagram of a learning module verification unit and related result sets.
3 is a flowchart of a genetic algorithm applied to the present invention.

Hereinafter, a device control method and apparatus based on self-verification type sentiment analysis using artificial intelligence according to the present invention will be described in detail with reference to the accompanying drawings. However, for members having the same function by the same configuration, the same reference numerals are maintained even if the drawings are different, and thus detailed description thereof may be omitted.

In addition, the relationship in which other members are arranged or connected to the front, rear, left and right sides, top and bottom of a certain member includes a case where a separate member is inserted in the middle. Conversely, when a member is said to be'right' before, left, or above and below the other member, it means that there is no separate member in the middle. And, when a part is said to'include' other components, it means that other components may be further included, rather than excluding other components unless otherwise stated.

In addition, the classification of the names of the configurations into first, second, etc. is for classifying the configurations in the same relationship, and is not necessarily limited to the order. In addition, terms such as'unit','means','unit','member', and'module' described in the specification mean a unit of a comprehensive structure that performs at least one function or operation. In addition, information processing devices such as terminals and servers described in the specification basically mean hard wiring, which means hardware in which a specific function or operation is implemented, but should not be interpreted to be limited to specific hardware. It does not exclude soft wiring made up of software that is driven to enable specific functions or operations to be implemented. That is, the terminal or the server may be a device or software installed on a device, such as an app.

In addition, the size and thickness of each component shown in the drawings are arbitrarily shown for convenience of description, so the present invention is not necessarily limited to those shown in the drawings, and in order to clearly express various parts and regions such as layers and regions In some cases, the thickness is exaggerated and enlarged or reduced.

<Basic configuration>

The device control method based on self-verification type sentiment analysis using artificial intelligence of the present invention performs emotional inference by artificial intelligence, such as machine learning, based on various unstructured data, and a control command for controlling environmental variables according to the result value. This is a technology for outputting to the target device 301. The method of the present invention includes data collection steps (110, 210) ; A data processing step 120 ; A learning model verification step 124 ; It characterized in that it comprises a classifier driving step 220 .

The data collection step (110, 210) is structured or unstructured data including at least one of data that may affect the user's emotional state and data that allows the user's emotion to be estimated, that is, At least one of body data, environmental data that may affect the user, and text data related to the user is collected. The collected data serves as a basis for inferring the emotional state of the user. The data collection step may be executed by the data collection units 111 and 211 constituting the data collection modules 110 and 210.

Data affecting the user's emotions include weather, outdoor environment, indoor environment, device operation data, activity data, and body data.

-Meteorological climate includes, for example, ASOS measurement air pressure, temperature, relative humidity, wind direction, wind speed, precipitation, snowfall, ground conditions, solar radiation, sunlight, evaporation, and the like.

-Outdoor environment data includes, for example, temperature, humidity, illuminance, fine dust, carbon dioxide concentration, noise, etc. outside the target device.

-Indoor environment data includes, for example, temperature, humidity, illuminance, fine dust, carbon dioxide concentration, noise, etc. inside the target device.

-Device operation data includes, for example, the number of app clicks and the number of visits to the app.

-Activity data includes, for example, financial status, purchase receipts, news, etc.

-Body data includes, for example, pulse, sleep time, caffeine/alcohol intake, number of steps, and body temperature.

The data that allows the user's emotions to be estimated include texts written by the user, facial expressions of the user, the user's pitch, and body data of the user.

-User-written posts include text data, such as SNS (Facebook, Twitter, Instagram, etc.) posts, texts (short messages), comments, and online posts.

-The user's facial expression includes, for example, facial expressions displayed in photos, videos, etc.

-The user's pitch includes, for example, a voice.

-The user's body data includes, for example, pulse rate, brain wave, respiratory rate, and the like.

The unstructured data may be composed of local data and global data according to the density of spatial/temporal/content related to the user. The local data is data that is spatially and spatially close to or related to the user. The global data is data that is spatially and spatially distant from the user or not related to the content. However, depending on the embodiment, processing may be limited to only local data or only global data, and the present invention should be interpreted as including such an embodiment.

Further, according to the configuration of the data collection device, the global data may be collected by the server 100 and the local data may be collected by the terminal 200 . However, in a system consisting of only the server 100 or only the terminal 200, both global data and local data may be configured to be collected by one type of device, and the present invention should be interpreted as including such an embodiment. do.

The data collection step 110 of the global data can be executed by the global data collection unit 111, and various data (eg, written texts) that can measure the emotional state of humans and affecting the emotional state of humans. It can play a role of collecting various data (eg, environmental information). Since the collection method is different according to the type of data, it may be performed in a module unit according to the data type and collection method. The data and modules to be collected include, but are not limited to, a “text collection module” and a “weather information collection module”, and various other modules may be further added.

For example, data collection may be performed by a text collection module.

Text data refers to all data consisting of'texts' containing emotions left by people, and may include personal data such as SNS posts, text messages, emails, and news and news comments reflecting multiple opinions. .

In the case of SNS posts, you can develop and collect a dedicated crawler that collects SNS posts. At this time, by analyzing the summary of the posted article, it is possible to collect only the articles posted by individuals not related to the advertisement.

In the case of news, for example, Naver News generates an average of about 10,000 to 15,000 articles per day and an average of 4 to 5 million articles per year, and an average of 9.8 comments per article. For example, assuming that 50% of articles with comments are among articles published in a year, and calculating that there are an average of 10 comments, 25 million comments can be confirmed. For this, it is possible to collect both the original text and comments of articles for at least 10 years or longer. Since it is assumed that the emotional state shown in the comment will be affected by, for example, personal sensitivity, the subject of the news, and the climate, a method of comparing and analyzing time information and emotional information can be sufficiently used.

Additionally, when the terminal 200 requests necessary information based on the personal information registered in the corresponding terminal, it may perform a search function and transmit data. At this time, if data is not in the corresponding condition, data can be transmitted after data collection is performed by actually updating the data on the web (processing with the highest priority).

For example, the weather information collection module may collect data about outdoor environment information, for example.

At this time, in order to utilize the local weather information provided by the Meteorological Agency, data can be collected in two ways, for example.

(1) A proposal to develop and utilize a web crawler that directly collects AWS data from the Weather Nuri website provided by the Meteorological Agency for real-time data collection.

(2) A plan to download and utilize past data provided collectively by the Meteorological Administration

This can be.

The original data of the former and the latter are the same, but there is a slight difference in the degree of processing, so the data can be used as data after correction work.

In addition, data provided by the Ministry of Environment, such as fine dust concentration and partial pressure information of gas (CO, CO2, etc.) can be managed and used as a database.

Additionally, when the terminal 200 requests necessary information based on the personal information registered in the corresponding terminal, the data may be transmitted by performing a search function for this. At this time, if data is not in the corresponding condition, data can be transmitted after data collection is performed by actually updating the data on the web (processing with the highest priority).

Collection of local data may be performed by the local data collection unit 211. The local data collection unit 211 is a part in charge of collecting surrounding environment information in a place where the present invention is installed (eg, a living room in a house). For example, data may be read from a sensor included in the invention, or data may be collected through a method such as wireless communication in a wearable device.

For example, the indoor environment sensor module may perform a function of collecting data in real time from an air temperature and humidity sensor, an atmospheric pressure sensor, a light amount sensor, a microphone (noise measurement), a CO2 partial pressure measurement sensor, etc. to collect indoor environment information.

For example, the wearable sensor module is a module that collects, in real time, information on heart rate, body temperature, movement speed, and location that can be obtained from various wearable devices such as a smart watch. This part collects personal state information of the user who uses the present invention, infers the emotional state through personal reactions in the current environmental state, and is used as basic data for giving environmental changes that can be transferred to a better emotional state. I can.

The data processing step 120 is a step of generating a combination of two or more emotional reasoning classifiers and environmental variable control classifiers by the learning model generation step 122 through analysis of the structured or unstructured data using artificial intelligence.

The learning model generation unit 122 that implements the learning model generation step 122, from the data preprocessing unit 121, includes'data preprocessing module','data preprocessing module result','emotional reasoning classifier parameter', and'environment variable control. 'Emotional Reasoning Classifier' and'Environmental Variable Control', which are learning models that receive'classifier parameters', infer people's'emotional state' by making the most of the machine learning engine, and derive the optimal environment from them. It can play a role of deriving'classifier'.

There are many different methods of machine learning, and each method has appropriate parameters. As a parameter, a plan that continues evolving from the learning model verification unit 124 may be applied. (Received through the data preprocessor 121) In the first step before passing through the learning model verification unit 124, the value of the initial value setting unit 126 may be applied.

There are several methods of machine learning, and they can be largely divided into Supervised Learning and Unsupervised Learning. The former refers to configuring a classifier based on learning data, and the latter refers to classifying data in the absence of learning data, that is, data whose results have been previously determined.

The Supervised Learning includes a linear regression, a decision tree, a support vector machine, and a neural network technique, and the Unsupervised Learning includes various cluster methods, For example, there are methods such as k-means clustering, hierarchical clustering, t-SNE clustering, and DBSCAN clustering.

One of the advantages of machine learning is that compared to the method of classifying and analyzing data based on previously known prior knowledge, it is possible to find new relationships that were not known before. This is the ability to infer emotional states from various data. It is a very suitable way to find out many factors that you do not know about.

In the present invention, both of the above two methods are utilized, and a better method is utilized at any time to derive an optimal result. In addition, by combining various methods, factors, and input data values, various classifiers are created, tested by themselves, and a system environment is provided to compare the results.

Linear regression analysis is a statistical method that explains or predicts the relationship between the independent variable and the dependent variable while the basic assumption of linearity is satisfied.It is a simple regression analysis for one independent variable and multiplex for multiple cases. It is called regression analysis. That is, in regression analysis, the value of the dependent variable changes in a certain pattern according to the independent variable, and the association relationship to the input value can be extracted by inferring a regression line representing the relationship between these variables.

Decision Tree is one of the most frequently used methods in data analysis. The classification value of the data to be analyzed is a node, and the data value that explains or connects the data is a branch. Has a structure that is assigned to When a continuous value for an analysis object is received from a decision tree, it is classified as a regression tree and used. This method is advantageous in that it is possible to track the process in which the analysis result is derived, and in the present invention, it can also be used as a method of confirming the relationship between environmental information and emotional state.

The support vector machine is a method suitable for classifying various and large amounts of data, and is a method that can be mainly used in the present invention. This method is applicable to non-linear classification, and shows excellent performance when appropriate extraction of features entered as input values is performed.

Neural Network (Deep Learning) is based on an artificial neural network (ANN) to enable computers to learn on their own like humans using multiple data. This deep learning mimics the information processing method that distinguishes objects after the human brain discovers patterns in a number of data, and learns machines so that computers can identify objects. When the deep learning technology is applied, the computer can recognize/infer/determine by itself, even if a person does not set all criteria for determination, and in the present invention, the state can be used to infer a new state or not be determined.

Hereinafter, an emotional reasoning classifier will be described. The emotional state refers not only to the states that humans generally think of (e.g., comfort, discomfort, joy, sadness), but the state of machine learning outcomes (emotional status). This new definition of the'emotional state' is not intended to infer the'emotional state' previously recognized by humans, but to define a new type of'emotional state' through machine learning, and This is because it aims to infer the optimal environmental change strategy. Some of the data to be derived through this method are expected to result in similar results to the currently known emotional state, but when learning under various conditions and checking the results, it is expected that emotional states that humans cannot interpret will be sufficiently produced. do.

The correlation between sensitivity and other factors aims to find all possible cases, even if it is not intuitive as in a previously known relationship (eg, a high discomfort index makes you feel a lot of anger or irritation). The emotional state is inferred using correlations between various types of collected unstructured data. Therefore, it is possible to proceed by borrowing a machine learning method that can analyze relationships even in the absence of existing prior knowledge.

Hereinafter, an environmental variable control classifier will be described. Based on the value of the emotional state obtained from the emotional inference classifier, an optimized emotional state is estimated than the current emotional state, and an optimal environmental change strategy for transitioning to the optimized emotional state is constructed.

Using the'current emotional state' which is the result of the emotional reasoning classifier and the'pre-processed data' that is the result of the data preprocessor 121 as input values, inferring the emotional state more optimized than the current emotional state, and changing the surrounding environment Create a classifier that derives the strategy.

The reasoning result serves to find a better emotional state based on the'emotional state' defined in the emotional reasoning classifier, and simultaneously derives strategies for transitioning to this state (e.g., the rate of temperature change). do.

At this time, the emotional change felt when transitioning from a certain state (state 1) to another state (state 2) based on the newly defined'emotional state' is a negative emotional state (sadness, annoyance) that is simply defined in the past. , Anger, etc.) to a positive emotional state (pleasure, satisfaction, etc.) may not be expected to result. This is because the emotional state may not always be the best state, and the results from the present invention are expected to lead to many new proposals that are not known before.

The emotional reasoning classifier and the environmental variable control classifier may be periodically transmitted to the terminal 200 as a learning model combination by the learning model engine management unit 123.

Two or more combinations of the emotional reasoning classifier and the environment variable control classifier generated by the data processing step 120 may be generated according to the type or environment of artificial intelligence, reasoning assumptions, and data types. When a combination of a plurality of emotional reasoning classifiers and environmental variable control classifiers is generated in this way, it is possible to promote self-verification type evolution in classifier construction.

In the learning model verification step 124, the combination of the two or more emotional inference classifiers and environmental variable control classifiers generated by the learning model generation unit 122, the emotional inference classifier parameter, the environmental variable control classifier parameter, and each classifier are derived. This is a step in which the optimal combination of the emotional reasoning classifier and the environmental variable control classifier is finally selected through evaluation and verification using artificial intelligence for the preprocessing modules in the previous stage. In other words, for the combination of two or more emotional inference classifiers and environmental variable control classifiers already created through various methods of artificial intelligence based on the result of the data preprocessing module, artificial intelligence, evaluation function, and genetic It evaluates through the method of the algorithm, adjusts and transforms the underlying'preprocessing module' to derive the optimal solution of the classifier. Through such a process of deriving a plurality of solutions and calculating an optimal solution among them, the control method of the present invention can have an organic characteristic that seeks an evolutionary change by itself so that an optimal solution can be sought according to the given data.

Specifically, in the learning model verification step 124, a combination of the emotional reasoning classifier and the environmental variable control classifier is evaluated through the learning model verification unit 124 based on artificial intelligence. As a method of evaluation, in addition to the existing machine learning, for example, an evaluation function using a unique set of test information and a genetic algorithm method of AI may be used.

That is, the reason for this transformation is that heterogeneous big data (environmental information, text, biometric data, feedback data from a terminal, etc.) is used in this patent. In general, when the amount of input data (dimension) increases when using machine learning, that is, when a certain amount of dimensional data is entered, the machine learning efficiency decreases rapidly due to the Vapnik-Chervonenkis dimension theory. to be. Accordingly, it becomes very important to actively utilize the data preprocessor to preprocess big data. However, there are many types of arithmetic units used in this pre-processing process, and if combinations thereof are considered, an almost infinite number of combinations can be created. In this way, finding the optimal solution in a state in which infinite combinations are possible can be solved by using a genetic algorithm.

Genetic algorithms are parallel and global search algorithms based on Darwin's theory of survival of the fittest, a fundamental concept in genetics. Genetic algorithms aim to find better and better solutions by expressing the possible solutions to the problem to be solved in a data structure in a fixed form, and then transforming them over generations. Here, the data structure representing the solutions can be expressed as'genes', and the process of creating better and better solutions by transforming them can be expressed as'evolution'.

This algorithm is a search algorithm to find the optimal value of x in an unknown function y=f(x) by borrowing the concept of evolution in biology.

In other words. Genetic algorithms make various attempts through evolutionary processes (selection, intersection, mutation, and imputation processes) from the starting form to find the local optima. Accordingly, in this patent, the advantage of overcoming the part where machine learning is not suitable for big data and efficiently solving the combination of near-infinity preprocessors was adopted.

(1) Performance verification and evaluation of the classifier:

Evaluation of each set of learning model combinations is conducted based on the amount of input values, the efficiency of the prediction work (time and resource usage, etc.), and the rationality of the results.

The evaluation function evaluates N result sets of the learning model generation unit.

The result set of the learning model generation unit is: 1 emotional reasoning classifier, 1 environment variable control classifier, 1 emotional reasoning classifier parameter, 1 environment variable control classifier parameter, and data preprocessing module 1 that derives the corresponding environment variable control classifier. Means dog.

To this end, an evaluation method for each item is set. For example, as shown in FIG. 2, the evaluation result can be quantified using the evaluation function 131 basically.

As an example, the evaluation function 131 may be defined as follows, for example.

f(x, y, z) = 0.1x + 0.3y + 0.6z

here,

x: amount of input value (relative value);

y: The efficiency of the prediction operation (relative value, calculated based on the one that requires the most resources in the combination of the training model to be tested);

z: Rationality of the result (the distance from the target value that can achieve the object of the present invention is calculated. The distance calculation is calculated as a multi-dimensional Euclidean distance taking into account the weights of several factors)

In addition, in the case of the rationality of the results, it is possible to infer the result of the change of emotions due to changes in the surrounding environment from big data, and evaluate the change of emotions according to the change of the environment based on this. In addition, it is possible to improve accuracy by comparing and evaluating the expected scenario of the learning model generation step and the effect that was actually operated on humans (local feedback data).

In addition, it is determined whether the result (evaluation score) of the evaluation function 131 is, for example, 90 points or more, and whether the current number of repetitions is the same as the maximum number of repetitions, for example, any of the subsequent selection module 132 and the final selection module 135 Decide whether to move on to the step.

(2) Transformation of learning model combinations:

When the evaluation is completed, the selection module 132 may perform a decision to leave only the upper 50% and discard the lower 50%, for example, with respect to the set evaluated by the evaluation function 131. At this time, the number determining existence or existence may change depending on the situation. This is similar to the process corresponding to'selection' of a genetic algorithm.

The set of the top 50% and the evaluation score of the set are transmitted to the algorithm module 133 and the storage unit 125. At this time, the transmission set is named'selection module transmission set', and consists of 1 emotional reasoning classifier parameter, 1 environmental variable control classifier parameter, and 1 data preprocessing module.

That is, the part that can be changed (e.g., change in preprocessing method, method for generating emotional inference classifier, factor value adjustment, etc.) for the top 50% of the learning model is automatically performed, and'for selecting the optimal learning model. An evaluation candidate group' may be generated and transmitted to the data preprocessor 121. Through this, the data preprocessing unit 121 may analyze and apply a new combination method, and the learning model generation unit 122 may perform a new classifier generation operation. Through this method, it is possible to repeat the task of testing and generating a large number of combinations of learning models, so that the combination of the classifier that produces the best results most efficiently can be selected.

For example, the present method can be described as a method of solving a problem in which the sum of three of the variables {1, 5, 6, 8, 3, 7, 3, 5, 9, 0} is 20.

In the solution method, first, four combinations of three random numbers are created. For example, (1,5,3) (8,0,9) (9,9,8) (3,7,5).

Thereafter, when determining how close the sum of the numbers of these four combinations is to the target value of 20, it can be seen that they are 11, 3, 6, and 5, respectively.

Among them, after selecting the combination (8,0,9) (3,7,5) that has the closest value to the target value, the combination of the selected numbers is randomly changed, and the sum is closer to 20. It goes through the process of checking if it comes out. Through this, it is possible to find a combination of numbers whose sum is 20. In this way, it is possible to find a combination of learning models that most efficiently derives optimal results.

The advantage of this method in the present invention is that numerous pre-processing processes can be considered by combining various types of calculation units. Through this, it is possible to quickly perform a test suitable for the purpose to be achieved by the present invention, and even if it does not necessarily meet the purpose, various methods can be considered by considering unexpected cases such as slight mutations. The reason why the top 50% is selected in the above method is also because it can play a role in guiding the direction of achieving the goal while making the width of the variation sufficient.

This method, in combination with the advantages of machine learning, can efficiently and continuously solve the complex problems proposed by the present invention. Particularly in the production of personalized classifiers, this method is expected to be more useful.

<Storage unit 125>

Definition: The storage unit 125 is responsible for storing various [data pre-processing-classifier combinations] that have emerged as an intermediate step in the process of deriving an optimal data pre-processor-classifier combination.

Role: In order to derive the optimal [data preprocessing-classifier combination], it is necessary to store various intermediate stages of [data processing-classifier combination] that come out through an iterative process, and keep the past records with each step. You can overcome the limitations. This means that this process can be implemented with fewer resources.

<Algorithm module 133>

To this end, parameter values (emotional reasoning classifier parameter, environmental variable control classifier parameter, and parameters inside the preprocessing module) are determined based on the'selection module transmission set' selected by the selection module 132 by the algorithm module 133 for this purpose. By adjusting, a new evolved variant is created.

Specifically, the parameters of the data preprocessing module randomly perform an operation of adding or subtracting terms from an existing arithmetic unit combination equation.

In addition, parameters of the emotional reasoning classifier and the environmental variable control classifier are evolved and transformed as follows.

If the parameter is of continuous type, the value is randomly changed based on the initial value entered within the parameter range. In addition, when the parameter is a selection type, an option other than the current option is selected to provide direction to machine learning.

For example, in the case of the machine learning method of a support vector machine, it has the following optional or continuous parameter values.

-S <int> [optional]

Set type of SVM (default: 0)

0 = C-SVC

1 = nu-SVC

2 = one-data preprocessing operation SVM

3 = epsilon-SVR

4 = nu-SVR

-K <int> [optional]

Set type of kernel function (default: 2)

0 = linear: u'*v

1 = polynomial: (gamma*u'*v + coef0)^degree

2 = radial basis function: exp(-gamma*|u-v|^2)

3 = sigmoid: tanh(gamma*u'*v + coef0)

-D <int> [continuous]

Set degree in kernel function (default: 3)

-G <double> [continuous]

Set gamma in kernel function (default: 1/k)

-R <double> [continuous]

Set coef 0 in kernel function (default: 0)

-C <double> [continuous]

Set the parameter C of C-SVC, epsilon-SVR, and nu-SVR

(default: 1)

-N <double>[continuous]

Set the parameter nu of nu-SVC, one-data preprocessing driving SVM, and nu-SVR

(default: 0.5)

-Z [optional]

Turns on normalization of input data (default: off)

For example, for changing parameters, -S, which is an optional type among the options, can have one of 0, 1, 2, 3, 4 as an optional type, and out of 4 cases excluding the option used just before. Choose at random.

<Directional Machine Learning Engine (134)>

The directional machine learning engine 134 plays a role of predicting and determining whether an evolutionary variant of the set generated through the algorithm module 133 is efficient or not efficient through machine learning.

The directional machine learning engine 134 is needed because not all evolutionary variants of the set generated through the algorithm module 133 are more advanced. The directional machine learning engine 134 includes a directional prediction classifier.

-Directional prediction classifier:

It is a machine learning classifier that evaluates evolutionary variants of a set randomly generated in the algorithm module 133. As a method of generating this machine learning classifier, linear regression, decision trees, support vector machines, and neural networks can be used.

The directional predictive classifier may be generated from a combination of the sentiment inference classifier and the environmental variable control classifier, parameters, and evaluation scores by a directional machine learning engine.

-Learning data required to build a classifier: The set information of the top 50% of the evaluation result selected by the selection module 132 during the loop being repeated from the storage unit 125 so far and the evaluation score are included.

-Purpose of learning: The purpose of this study is to derive the relationship between the final result value of the modified formula and classify the score into positive and negative cases.

-Input data to be entered into the classifier: an evolutionary variant of the set generated by the algorithm module 133 is included.

-Evaluation output value: Can be improved (Class 1) and cannot be improved (Class 2).

Able to improve (Class 1) is a group of possibilities where a variant of the set can show better results than the existing set.

Impossible to improve (Class 2) is a group with a high probability that a variant of the set will not produce better results than the existing set.

-Machine learning engine result: The model entering Class 2 is discarded, and the model entering Class 1 is transmitted to the data preprocessor 121.

(3) The final selection module 135 of the learning model:

For the various results selected through the above process (2) ((2) transformation of the learning model combination), the'data preprocessing module, emotional reasoning classifier, and environment control classifier' with the highest scores are finally selected.

That is, when the set evaluation by the evaluation function 131 is completed, the set and the evaluation result score are transmitted. At this time, not all information of the set (i.e., the result set of the learning model generation unit) performed in the evaluation function 131 is imported, but only the data preprocessing module, the emotional reasoning classifier, and the environmental variable control classifier are imported. It is called'the final selection module transmission set'. Among them, one'final selection module transmission set' that received the highest evaluation result score is transmitted to the learning model engine management unit 123.

In the classifier driving step 220, the environmental variable control classifier is driven according to the emotion classification value generated by driving the emotion inference classifier of the optimal combination, so that a device operation command for the target device 301 of control is This is the step to print.

The classifier driving step 220 may be executed by the learning model driving unit 222. The learning model driving unit 222 receives the data input from the data preprocessing driving unit 221 from the learning model engine management unit 123 of the server 100, a'emotional reasoning classifier' and a'environment variable control classifier' (learning model combination ), the final result is the device operation plan (control value). The learning model driving unit 222 may include an emotional reasoning classifier driving device and an environmental variable control classifier driving device.

The optimal learning model combination and data preprocessing module finally selected by the learning model verification unit 124 may be configured to be periodically transmitted to the terminal 200.
The emotional reasoning classifier driving device plays a role of driving the optimal emotional reasoning classifier received from the'learning model engine management unit' 123 of the server 100. The data received from the data preprocessing driver 221 is input to the emotional inference classifier, and a result is calculated.

The environment variable control classifier driving device is based on the optimal environment variable control classifier received from the'learning model engine management unit' 123 of the server 100, based on the result from the emotional reasoning classifier as an input value. Derive a strategy (control value) that can create an environment of

Strategy information that can change the environment of the place where the present invention is installed becomes the final output value of the learning model driving unit 222. For example, it can be an output value that says,'I want to lower the current temperature by 2 degrees in 10 minutes and increase the humidity by 20%'.

The output of the control value may be performed by the controller 223. The control unit 223 plays a role of deriving a strategy that can be actually implemented based on the environment change strategy obtained from the learning model driving unit 222.

The control unit 223 may include an environment variable analysis module. The environmental variable analysis module includes all other devices in addition to home appliances such as air conditioners, fans, heaters, humidifiers, dehumidifiers, smart lights, smart diffusers, and smart speakers that can control environmental variable change strategies. Considering the characteristics, a specific strategy is composed of which equipment to use with a certain amount of output, that is, whether to use an air conditioner or a humidifier.

In addition, by configuring a control signal in each device, a method to drive the actual device is configured.

For the final selection of the device driving method, the optimal conditions can be extracted using a cost evaluation function. This function calculates the cost (including energy usage) of running each device. Based on the determined strategy, various possibilities of operating the device are evaluated by the cost evaluation function to determine and execute the optimal device operation plan.

At this time, the cost evaluation function is composed of a function that considers the electricity consumption of devices (eg, air conditioners, dehumidifiers) that can change the indoor environment, and the degree of consumption of consumables used by each device. It is configured to consider dividing electricity consumption and electricity consumption under continuous use.

For example, it is assumed that a plan to change the indoor environment from the current 28°C and 70% relative humidity to 24°C and 40% relative humidity within 30 minutes is previously derived. Strategies to do this include, for example

1) Start the air conditioner at a desired temperature of 18 degrees for 30 minutes,

2) Operate the air conditioner at a desired temperature of 24 degrees for 15 minutes, operate at the maximum dehumidifier for 30 minutes,

3) The air conditioner operates at a desired temperature of 22 degrees for 10 minutes, the fan operates for 10 minutes, and the dehumidifier operates for 15 minutes at the maximum.

And other strategies can be inferred. In order to choose one of these three strategies, a cost evaluation function can be used, for example, strategy 2 with the minimum value can be selected and executed.

The configuration of the cost evaluation function takes as the first data to input the financial resources related to the device 301 from the outside, and is configured in a form in which this value is continuously modified based on continuous feedback information. Through this process, wear and efficiency of each device are calculated, and it may be possible to propose replacement and inspection of devices.

The final output value of the control unit 223 is a control signal for achieving a driving method (duration time, intensity, etc.) for various currently registered environmental variable control devices (eg, air conditioners, warmers, humidifiers, etc.) .

The control target device 301 means a device (eg, an air conditioner, a warmer, a humidifier, a wearable device, etc.) that can contribute to adjusting the state of emotion.

As another example, the output value of the learning model driver 222 is, for example, in order to transition from'<currently'emotional state 7'to'emotional state 12', the room temperature is lowered by 3 degrees within 30 minutes and the humidity is adjusted to 10%. And, the output value of the learning model driving unit 222 is determined.

Accordingly, the control unit 223 and the environment variable analysis module find a way to lower the indoor temperature by 3 degrees within 30 minutes, adjust the humidity to 10%, and create an environment with yellow lighting. As candidates for a treatment method for this, for example, the following methods may be selected.

(1) Set the air conditioner to a desired temperature of 18 degrees and run it for 30 minutes. Turn on the yellow smart light.

(2) Operate the dehumidifier at maximum for 30 minutes. At the same time, the air conditioner is set to a desired temperature of 20 degrees and runs for 15 minutes. Turn on the yellow smart light.

(3) Operate the fan with strong wind for 30 minutes. Also, run the dehumidifier for 30 minutes. Turn on the yellow smart light.

At this time, for the three methods, an optimal condition is extracted using, for example, a cost evaluation function based on the amount of power. Each of the three methods uses, for example, 5 kWh, 3 kWh, and 4.5 kWh, so select No. 2 as the most optimal method.

The final output value of the control unit is, for example, that the dehumidifier is operated at the maximum for 30 minutes, the air conditioner is set to a desired temperature of 20 degrees and operated for 15 minutes, and the yellow smart light is turned on.

<Effect>

With this configuration, for example, hot summer environment, heat wave weather condition, beach congestion news due to heat wave, text data indicating that it is difficult due to tropical night in the user's Facebook column, indoor fan operation situation, high indoor temperature, fast heart rate, etc. When various heterogeneous unstructured data are collected, the emotional state is accurately inferred from such data by artificial intelligence, such as machine learning. This inferred emotional state can be the emotional stages newly classified by artificial intelligence, regardless of whether humans understand or not. In addition, a control value of the target device 301 and an air conditioner temperature control control value for transitioning to a good emotional state by improving the inferred current emotional state may be output. Accordingly, a decrease in the indoor temperature is promoted, so that the user's emotional state can be improved.

For example, the information collected from the global data collection unit 111 of the server 100 is time 15:26 outdoor temperature 35 degrees, relative humidity 80%, fine dust concentration 89 μg/m ³ , ultrafine dust concentration 20 μg/ m ³ , nitrogen dioxide 0.012ppm, ozone 0.028ppm, carbon monoxide 0.4ppm, it may be a sulfuric acid gas 0.006ppm.

And, the information collected from the global data collection unit 111 of the server 100 is, as the content of the sns posting,'Oh, it's hot and rainy and it's totally annoying! It can be all annoying_15:20','It's raining, the sky is cloudy, my heart is cloudy, and I don't want to go to work_07:40'.

On the other hand, the information collected from the local data collection unit 211 of the terminal 200, room temperature 30 degrees, relative humidity 80%, fine dust concentration 34 ㎍ / m ³ , ultra fine dust concentration 5 ㎍ / m ³ , nitrogen dioxide It may be 0.012 ppm, ozone 0.024 ppm, carbon monoxide 0.2 ppm, and sulfuric acid gas 0.003 ppm.

In addition, information collected from the local data collection unit 211 of the terminal 200 may be 80 heart rate, 36.5 degrees body temperature, and 4,000 steps.

In the present invention, the preprocessing module, that is, the data preprocessing unit 121 preprocesses them.

As a result, the pretreatment result may be as follows: temperature and humidity index: 78 (strong stress) / SNS text emotional state: anger (15:20), irritation-sadness-depression (07:40).

And, the average temperature of the last 7 days may be 35 degrees.

In this case, when the driving device of the emotional reasoning classifier is operated, it can be said that the driving result is currently estimated to be, for example,'Emotional State No. 7'(generally 75% angry, 2% happy, 5% sad).

And when the driving device of the environmental variable control classifier is activated, the optimal sensibility transition direction in the current ``Emotional State No. 7'' is ``Emotional State No.12'' (generally calm 40%, comfort 50%, joy 1%, cool 9 %).

In order to transition to the emotional state No. 12 in this way, it can be determined that it is necessary to lower the room temperature by 3 degrees within 30 minutes, adjust the humidity to 10%, and create an environment with yellow lighting.

Accordingly, since it is possible to estimate the emotional state from various unstructured data and output a signal for controlling the target device for converting it to a desirable emotional state, there is an effect of inducing a more accurate emotional transition.

In addition, the present invention may have the following functional advantages.

(1) Diversification of preprocessing modules

-Conventionally, by applying the method of combining calculation units fixed by human intuition, there was a limit to the level of preprocessing, but various preprocessing modules (combination method of calculation units) can be automatically generated, thus reducing time and improving efficiency. do.

-Due to the nature of the'genetic algorithm' among the implementation methods of the learning model verification unit 124, since numerous pre-processing processes are performed by combining various types of computational units, combination results that humans may not expect can be derived. It is possible to create a learning model.

(2) Automated learning model creation

-Through the learning model verification unit 124, it is possible to automatically generate a large number of learning models, breaking away from the limited function of creating a learning model in a single line.

-Whenever a learning model is improved, it is possible to produce various classifiers more efficiently by automatically adjusting parameters and machine learning methods, instead of changing parameters and machine learning methods by making direct judgments by humans. do.

-Since a plurality of learning models are automatically generated by automatically calculating and processing input data processing method, learning model type, parameter value setting, etc., it is possible to efficiently create and test a classifier to find an optimized learning model. .

-For example, Tae-eum-in, Tae-eum-in, So-yang-in, and So-e-in are divided into four constituents. It is possible to create a classifier that reflects scenarios such as increasing sensitivity to

(3) Optimization of learning model selection

-High-level selection can be made by utilizing the function of verifying and selecting the classifier created by artificial intelligence by other artificial intelligences outside.

-Various combinations of learning models can be evaluated

Through the learning model verification unit 124, it is possible to create a learning model simply, and to avoid the limited possibility of manually optimizing it, to repeatedly generate various types of preprocessing modules, and to set various parameters for machine learning. Make adjustments possible. In particular, it is possible to generate a combination of a data preprocessing module and a classifier that enables more efficient emotional inference by controlling the part that generates random offspring, which is a disadvantage of the genetic algorithm, in the learning model verification unit 124. It is expected to be there.

In this case, in the learning model verification step 124, if the evaluation score by the evaluation function is more than the reference value, the optimal combination of the emotional reasoning classifier and the environmental variable control classifier is finally selected, and the evaluation score by the evaluation function is less than the reference value. In this case, a variant of the parameter of the combination of the emotional inference classifier and the environmental variable control classifier is generated by a predetermined algorithm, and after the evaluation by the directional predictive classifier for the variant, the variant is pre-processed with data if the evaluation result is not impossible to improve. It can be made to feed back to the stage. In other words, it is possible to find an optimal combination by creating a variant and testing various parameters. The AI genetic algorithm can be applied to the generation of the variant.

-Automated learning model creation and improvement possible

The problem (Vapnik-Cherovenkis dimension), which was difficult to apply to machine learning by directly using big data, can be sufficiently overcome, and by automatically producing and evaluating various combinations, the best combination of learning models can be derived. Is expected.

<Initial value setting unit 126>

In order to repeatedly proceed with the'data preprocessing unit' 121, the'learning model generation unit' 122, and the'learning model verification unit' 124, initial data preprocessing module, emotional reasoning classifier, and environmental variable control classifier The process of setting the value is necessary.

The initial value setting unit 126 sets the initial value of the parameter of the data preprocessing unit 121, the parameter of the emotional reasoning classifier, and the parameter initial value of the environmental variable control classifier, and transmits it to the data preprocessing unit 121. In charge.

<The correlation of heterogeneous unstructured data>

The unstructured data collected in the data collection steps 110 and 210 may be configured to be of a plurality of types . In this case, in the data processing step 120, it is preferable that the emotional state is inferred through the correlation between the heterogeneous unstructured data.

The data that affects human emotions and allows the emotional state to be inferred are not limited to any one type, and if possible, the more accurate the emotional state inference will be the more various types of heterogeneous data are considered and analyzed. I can. In the present invention, such heterogeneous unstructured data is collected and the emotional state is inferred by analyzing the correlation between these heterogeneous data rather than simple analysis between them. Therefore, more accurate inference of emotional state becomes possible.

<Data preprocessing>

In the data processing step 120, a data pre-processing step 121 is further provided for pre- processing the unstructured data by two or more pre-processing modules prior to generation of the combination of the two or more emotional inference classifiers and the environmental variable control classifier. This is desirable.

This data pre-processing step 121 is responsible for pre-processing the data by pulling the data collected in the data collecting steps 110 and 210. The preprocessed result value is transmitted to the learning model generation unit 122 and is used to generate an emotional inference classifier and an environment variable control classifier.

In this case, the combination of the two or more emotional reasoning classifiers and the environmental variable control classifier may be configured to be formed respectively through pre-processing using the two or more pre-processing modules. With this configuration, the combination of the emotional reasoning classifier and the environmental variable control classifier generated according to the preprocessing scenario may be different. Therefore, testing of various preprocessing modules becomes possible.

In addition, when the optimal combination is finally selected in the learning model verification step 124, the optimal preprocessing module used for the preprocessing of the optimal combination may be selected. Accordingly, since the preprocessing module used to derive the optimum combination of the emotional reasoning classifier and the environmental variable control classifier is selected as the optimum preprocessing module, the optimum preprocessing module can be used for preprocessing of unstructured data later.

The data preprocessing unit 121 is composed of one or more data preprocessing modules. The data preprocessing module implementing this step can be configured or expressed as a combination of arithmetic elements, and is composed of parameters and change records.

The parameter is a combination formula of these calculation units, and the parameter value continuously evolves by the learning model verification unit 124. The change record means a record of evolution until one preprocessing module has the current parameter value.

The operation unit is a unit that receives one or more input values, calculates the average, maximum, median, and special index (eg, discomfort index, heat index), and returns the result. A method of combining computational units, that is, a parameter, has a characteristic that is continuously evolving received from the learning model verification unit 124, and data is read from the data collection units 111 and 211 according to the parameter (combination method), and is automatically To perform the operation. The initial operation unit (parameter) may be started with a default value set by a user in the data preprocessing module through the initial value setting unit 126.

This preprocessing module can interpret and apply the combining method according to the generated learning model, that is, the verification result of the classifier. The derived result value is transmitted to the learning model generation unit 122 again. The units of operation are pre-combined in one fixed way. The data preprocessing unit 121 receives the'evaluation candidate group (emotional reasoning classifier and environmental variable control classifier) for selecting an optimal learning model' from the learning model verification unit 124, and interprets and applies the combination method. The derived result value is transmitted to the learning model generation unit 122.

As an example of the calculation unit, the calculation unit may request the AWS minute-by-minute temperature data of the Meteorological Administration for January 2017 as an input value to calculate the monthly average temperature, and calculate a value excluding the missing value as an average.

As an example of the calculation unit, the calculation unit for calculating the daily average discomfort index is 9/5*T-0.55(1-RH)(9/5*T-26)+32 formula (T: temperature, Meteorological Administration AWS It can be calculated by using the daily average calculated value based on the minute unit data of, RH: condition humidity, and the daily average calculated value based on minute unit data of the Meteorological Administration AWS.

As an example of the method of combining the calculation units, in order to calculate the daily average temperature and precipitation, the weekly average temperature and precipitation, first, by reading the AWS minute data of the Meteorological Administration at a specific point in time and point, all temperature and precipitation information ( The operation unit that calculates the average value) is driven. Among these results, the weekly average temperature and precipitation information can be obtained by driving the weekly average calculation unit with the desired temperature and precipitation data of a specific week (eg, the first week of May 2017) as input values.

In the case of simultaneously driving various calculation units as described above, a combination method such as using the result value of the preceding calculation unit as an input value of a later calculation unit is also possible. As the number of types of operations increases, such a combination method may become more diverse.

<Transmission of the preprocessing module to the terminal 200>

In the pre-processing module for processing the data pre-processing step 121, the pre-processing module finally selected by the learning model verification unit 124 is the terminal 200 by the learning model engine management unit 123 in the data processing step 120. ) Can be made to be transmitted periodically.

The emotional inference classifier and the environmental variable control classifier made by the learning model generation unit 122 are finally selected by the learning model verification unit 124, and the selected emotional inference classifier and the environmental variable control classifier are the learning model engine management unit 123 By this, it may be periodically transmitted to the terminal 200.

<Pre-processing of data just before driving the learning model>

In the classifier driving step 220, a data pre-processing driving step 221 in which the unstructured data is pre-processed by the optimal pre-processing module is preferably further provided.

The data preprocessing driving step may be performed by a data preprocessing driver. The data preprocessing driving unit refers to a part that receives a data processing policy based on data obtained from the data collection unit, for example, from the'learning model engine management unit' 123, and preprocesses the data according to the corresponding method.

The data preprocessing driving unit may be executed by the data preprocessing module driving apparatus. The data preprocessing module driving device, for example, periodically downloads a policy for the optimal data preprocessing method from the'learning model engine management unit' 123, and preprocesses the data collected by the data collection units 111 and 211 based on this. .

In addition, when specific data (eg, text data related to personal information registered in the terminal, weather information under specific conditions, etc.) is required, the data is requested from the server and downloaded and processed.

<Feedback configuration of output value>

It is preferable that the output value by the operation of the target device is fed back to the data collection step and collected together with the unstructured data.

The state after driving each device is measured and collected by the local data collection unit 211, which is defined as feedback local data. The feedback local data can be actively used to learn the emotional reasoning classifier and the environmental variable control classifier. In addition, it can be used as essential data in the production of a personalized classifier in the future.

For example, all data (text, wearable device_heart rate, body temperature, etc.) for 30 minutes from the moment the controller output value is input to the device and operated are labeled and transmitted to the data collection unit 211.

<Description of the operation in the configuration including the data preprocessing unit 121>

Hereinafter, the operation of the learning model verification unit 124 in the configuration including the data preprocessor 121 will be described.

N learning model generation unit result sets (1 data pre-processing module, 1 emotional reasoning classifier, 1 environmental variable control classifier, 1 emotional reasoning classifier parameter, 1 environmental variable control classifier parameter) to the learning model generation unit 122 ), the learning model verification unit 124 performs an evaluation using the'evaluation function' for the currently configured set (S11). If the evaluation score of the evaluation function is a reference value, for example, 90 points or more (S12: Y), the N final selection model transmission set and the evaluation score are transmitted, and selected as the set of the highest score by the final selection module 135 (S14), which is transmitted to the learning model engine management unit 123 (S15). This set of emotional reasoning classifiers and environmental variable control classifiers are driven by the classifier driving module 220 and used to drive and control the target device 301.

And, even when the evaluation score of the evaluation function is less than the reference value, for example, 90 points (S12: N), if the current number of iterations reaches the maximum number of iterations (S13: Y), N final selection models for processing cost efficiency The transmission set and its evaluation score are transmitted and selected as the set of the highest score by the final selection module 135 (S14), which is transmitted to the learning model engine management unit 123 (S15), and the target device 301 It should be used for drive control.

On the other hand, if the evaluation score of the evaluation function is less than 90 points (S12: N) and the current number of repetitions has not reached the maximum number of repetitions (S13: N), various modifications to the current set can be searched. There is a need. To this end, in order to reduce the target range, for example, only the top 50% set is selected by the selection module 132 (S21), and the algorithm module 133 for the transmission set of N selection modules and the evaluation score While generating a variant (S22) while adjusting, the set information and evaluation score are transmitted to the directional machine learning engine, and a directional predictive classifier is constructed (S23) from the set log record and evaluation score, and the set is set using the directional predictive classifier. Evaluation (S24) is made for the modified proposal of.

The variant that is determined to be impossible to improve (Class 2) by this classifier is removed (S25), and the selected variant is N learning model verification unit result sets (1 emotional inference classifier parameter, 1 environmental variable control classifier parameter, As one data preprocessing module), it is transmitted to the data preprocessing unit 121 (S26). The set variant transmitted to the data preprocessing unit 121 is used for preprocessing of newly incoming data from the data collection units 111 and 211, and is used for the preprocessing of the N data preprocessing unit result sets (1 data preprocessing module, 1 data preprocessing module. Calculation result value, one emotional reasoning classifier parameter, and one environmental variable control classifier parameter), and the applied data again affects the learning model generation unit 122 and results in N learning model generation unit result sets (1 data). It becomes a preprocessing module, 1 emotional reasoning classifier, 1 environmental variable control classifier, 1 emotional reasoning classifier parameter, 1 environmental variable control classifier parameter).

Accordingly, a set that satisfies the reference value is finally selected, and a set that does not satisfy the reference value is modified by an appropriate number of repetitions and is reevaluated, and there is room for satisfying the reference value again. In addition to the AI algorithm applied to the learning model generation unit, a modified genetic algorithm using an algorithm module and a directional machine learning engine can be applied to the generation of the variant.

<device>

The device control apparatus based on self-verification-type sentiment analysis using artificial intelligence according to an embodiment of the present invention substantially corresponds to the method of the present invention described above, and each step of the method is formed of a hardware module. Including the case of earrings. The apparatus of the present invention includes a data collection module 110 ; A data processing module 120 ; Learning model verification module 124 ; It characterized in that it comprises a classifier driving module 220 .

The data collection module 110 is a module for collecting unstructured data including at least one of a user's body data, environmental data that may affect the user, and text data related to the user.

The data processing module 120 is a module for generating a combination of two or more emotional reasoning classifiers and environmental variable control classifiers by a learning model generation module through analysis of the unstructured data using artificial intelligence.

The learning model verification module 124 allows the optimal combination of the emotional inference classifier and the environmental variable control classifier to be finally selected through evaluation and verification using artificial intelligence from the generated combination of the two or more emotional inference classifiers and the environmental variable control classifier. It is a module.

The classifier driving module 220 drives the environmental variable control classifier according to the emotion classification value generated by driving the emotion inference classifier of the optimal combination, so that a device operation command for the target device of control is output. It is a module.

<Program>

The present invention is a program recorded on a recording medium that can be read by an information processing device in which a program for performing the steps of configuring a device control method based on self-verification type sentiment analysis using artificial intelligence described above is recorded in an information processing device. It may be implemented as

Although the preferred embodiments of the present invention have been described above, the present invention is not limited to the disclosed embodiments, but is implemented by modifying in various forms different from within the scope of the claims, the detailed description of the invention, and the accompanying drawings. It can be practiced, and other equivalent embodiments are possible, and it is natural to those of ordinary skill in the art that this also falls within the scope of the present invention, and only the embodiments are intended to make the disclosure of the present invention complete, and the present invention It is provided to completely inform the scope of the invention to those of ordinary skill in the art to which it belongs, and the invention is only defined by the scope of the claims.

The present invention can be used in the industry of a device control method and apparatus based on self-verification type sentiment analysis using artificial intelligence.

100: server
110: data collection module
111: data collection unit
120: data processing module
121: data preprocessor
122: learning model generation unit
123: Learning model engine management unit
200: terminal
210: data collection module
211: data collection unit
220: classifier drive module
221: data preprocessing driver
222: learning model driving unit
223: control unit
300: target device
301: target device

Claims (9)

As a device control method based on self-verification type sentiment analysis using artificial intelligence,
A data collection step in which structured or unstructured data including at least one of data that may affect a user's emotional state and data that allows a user to infer a user's emotional state is collected by the data collection module;
Two or more emotional reasoning classifiers and artificial intelligence machines for inferring the emotional state, which is the state of the machine learning result, by the learning model generation step through the analysis of the structured or unstructured data using artificial intelligence machine learning of the data processing module Estimating a more optimized emotional state than the current emotional state based on the emotional classification value obtained from the emotional inference classifier using running, and controlling the environmental variables to change to the optimal environment for transition to the optimized emotional state. A data processing step of generating two or more combinations of environmental variable control classifiers according to the type or environment of the artificial intelligence, inference assumption, and data type;
Artificial intelligence and evaluation of the learning model verification module for the generated combination of the two or more emotional inference classifiers and environmental variable control classifiers, the emotional inference classifier parameter, the environmental variable control classifier parameter, and the preprocessing module at the front of each classifier. A learning model verification step in which an optimal combination is finally selected by adjusting and transforming the underlying preprocessing module through evaluation and verification through the method of function and genetic algorithm to derive the optimal solution of the emotional reasoning classifier and the environmental variable control classifier;
The driving method comprising the sorter is controlled environment variable categorizer driven in accordance with the emotional classification value generated by the driving of the emotional inference classifier of the optimum combination by the classifier drive module, such that the unit operation command for the destination device in the control output
Device control method based on self-verification type sentiment analysis using artificial intelligence, characterized in that it comprises a. As a device control method based on self-verification type sentiment analysis using artificial intelligence,
A data collection step in which structured or unstructured data including at least one of data that may affect a user's emotional state and data that allows a user to infer a user's emotional state is collected by the data collection module;
A data processing step of generating a combination of two or more emotional inference classifiers and environmental variable control classifiers by a learning model generation step through analysis of the structured or unstructured data using artificial intelligence of a data processing module;
A learning model verification step in which the optimal combination of the emotional inference classifier and the environmental variable control classifier is finally selected through evaluation and verification using the artificial intelligence of the learning model verification module from the generated combination of the two or more emotional inference classifiers and the environmental variable control classifier. ;
The driving method comprising the sorter is controlled environment variable categorizer driven in accordance with the emotional classification value generated by the driving of the emotional inference classifier of the optimum combination by the classifier drive module, such that the unit operation command for the destination device in the control output
It is made including,
The data processing step further includes a data pre-processing step of pre- processing the unstructured data by two or more pre-processing modules prior to the generation of the combination of the two or more emotional reasoning classifiers and the environmental variable control classifier,
In the learning model verification step,
If the evaluation score by the evaluation function is more than the standard value, the optimal combination of the emotional reasoning classifier and the environmental variable control classifier is finally selected,
If the evaluation score by the evaluation function is less than the reference value, a modification plan of the parameter of the combination of the emotional inference classifier and the environmental variable control classifier is generated by a predetermined algorithm, and after the evaluation by the directional predictive classifier for the modified plan, the evaluation result cannot be improved. If not, the modified proposal is fed back to the data preprocessing step.
A device control method based on self-verification type sentiment analysis using artificial intelligence, characterized in that. As a device control method based on self-verification type sentiment analysis using artificial intelligence,
A data collection step in which structured or unstructured data including at least one of data that may affect a user's emotional state and data that allows a user to infer a user's emotional state is collected by the data collection module;
A data processing step of generating a combination of two or more emotional inference classifiers and environmental variable control classifiers by a learning model generation step through analysis of the structured or unstructured data using artificial intelligence of a data processing module;
A learning model verification step in which the optimal combination of the emotional inference classifier and the environmental variable control classifier is finally selected through evaluation and verification using the artificial intelligence of the learning model verification module from the generated combination of the two or more emotional inference classifiers and the environmental variable control classifier. ;
The driving method comprising the sorter is controlled environment variable categorizer driven in accordance with the emotional classification value generated by the driving of the emotional inference classifier of the optimum combination by the classifier drive module, such that the unit operation command for the destination device in the control output
It is made including,
The data processing step further includes a data pre-processing step of pre- processing the unstructured data by two or more pre-processing modules prior to the generation of the combination of the two or more emotional reasoning classifiers and the environmental variable control classifier,
The combination of the two or more emotional reasoning classifiers and the environmental variable control classifier is each formed through pre-processing using the two or more pre-processing modules,
When the optimal combination is finally selected in the learning model verification step, the optimal preprocessing module used for preprocessing of the optimal combination is selected.
A device control method based on self-verification type sentiment analysis using artificial intelligence, characterized in that. As a device control device based on self-verification type sentiment analysis using artificial intelligence,
A data collection module for collecting structured or unstructured data including at least one of data that may affect a user's emotional state and data that allows a user's emotion to be estimated;
Using two or more emotional reasoning classifiers for inferring the emotional state, which is the state of the machine learning result, by a learning model generation module through analysis using artificial intelligence machine learning on the structured or unstructured data and machine learning of artificial intelligence. Environmental variable control for estimating a more optimized emotional state than the current emotional state based on the emotional classification value obtained from the emotional inference classifier, and controlling the environmental variable to change to the optimal environment for transition to the optimized emotional state. A data processing module for generating two or more combinations of classifiers according to the type or environment of the artificial intelligence, inference assumptions, and data types;
For the generated combination of the two or more emotional inference classifiers and environmental variable control classifiers, the emotional inference classifier parameter, the environmental variable control classifier parameter, and the preprocessing module at the front of each classifier, artificial intelligence, evaluation function, and genetic algorithm A learning model verification module that adjusts and transforms the underlying preprocessing module through evaluation and verification through the method to derive the optimal solution of the emotional reasoning classifier and the environmental variable control classifier so that the optimal combination is finally selected;
Classifier driving module for driving the environmental variable control classifier according to the emotion classification value generated by the driving of the emotion inference classifier of the optimal combination, and outputting a device operation command for a control target device
Device control device based on self-verification type sentiment analysis using artificial intelligence, characterized in that it comprises a. As a device control device based on self-verification type sentiment analysis using artificial intelligence,
A data collection module for collecting structured or unstructured data including at least one of data that may affect a user's emotional state and data that allows a user's emotion to be estimated;
A data processing module for generating a combination of two or more emotional reasoning classifiers and environmental variable control classifiers by a learning model generation module through analysis of the unstructured data using artificial intelligence;
A learning model verification module for finally selecting an optimal combination of the emotional inference classifier and the environmental variable control classifier through evaluation and verification using artificial intelligence from the generated combination of the two or more emotional inference classifiers and the environmental variable control classifier;
Classifier driving module for driving the environmental variable control classifier according to the emotion classification value generated by the driving of the emotion inference classifier of the optimal combination, and outputting a device operation command for a control target device
It is made including,
The data processing module further includes a data preprocessing module that performs preprocessing by at least two preprocessing modules on the unstructured data prior to generation of the combination of the at least two emotional reasoning classifiers and the environment variable control classifier,
In the learning model verification module,
If the evaluation score by the evaluation function is more than the standard value, the optimal combination of the emotional reasoning classifier and the environmental variable control classifier is finally selected,
If the evaluation score by the evaluation function is less than the reference value, a modification plan of the parameter of the combination of the emotional inference classifier and the environmental variable control classifier is generated by a predetermined algorithm, and after the evaluation by the directional predictive classifier for the modified plan, the evaluation result cannot be improved. If not, the modified proposal is fed back to the data preprocessing module.
A device control device based on self-verification type sentiment analysis using artificial intelligence, characterized in that. As a device control device based on self-verification type sentiment analysis using artificial intelligence,
A data collection module for collecting structured or unstructured data including at least one of data that may affect a user's emotional state and data that allows a user's emotion to be estimated;
A data processing module for generating a combination of two or more emotional reasoning classifiers and environmental variable control classifiers by a learning model generation module through analysis of the unstructured data using artificial intelligence;
A learning model verification module for finally selecting an optimal combination of the emotional inference classifier and the environmental variable control classifier through evaluation and verification using artificial intelligence from the generated combination of the two or more emotional inference classifiers and the environmental variable control classifier;
Classifier driving module for driving the environmental variable control classifier according to the emotion classification value generated by the driving of the emotion inference classifier of the optimal combination, and outputting a device operation command for a control target device
It is made including,
The data processing module further includes a data preprocessing module that performs preprocessing by at least two preprocessing modules on the unstructured data prior to generation of the combination of the at least two emotional reasoning classifiers and the environment variable control classifier,
The combination of the two or more emotional reasoning classifiers and the environmental variable control classifier is each formed through pre-processing using the two or more pre-processing modules,
When the optimal combination is finally selected in the learning model verification module, the optimal preprocessing module used for preprocessing of the optimal combination is selected.
A device control device based on self-verification type sentiment analysis using artificial intelligence, characterized in that. A recording medium readable by an information processing device in which a program for performing the steps of a device control method based on self-verification type sentiment analysis using artificial intelligence according to any one of claims 1 to 3 is recorded in the information processing device Recorded in the program. delete delete

Images