KR20170059546A - Automatic analysis apparatus of IoT things and IoT services and method thereof - Google Patents
Automatic analysis apparatus of IoT things and IoT services and method thereof Download PDFInfo
- Publication number
- KR20170059546A KR20170059546A KR1020150163542A KR20150163542A KR20170059546A KR 20170059546 A KR20170059546 A KR 20170059546A KR 1020150163542 A KR1020150163542 A KR 1020150163542A KR 20150163542 A KR20150163542 A KR 20150163542A KR 20170059546 A KR20170059546 A KR 20170059546A
- Authority
- KR
- South Korea
- Prior art keywords
- iot
- data
- service
- objects
- class
- Prior art date
Links
Images
Classifications
-
- G06F17/30705—
-
- G06K9/6269—
-
- G06K9/628—
Abstract
Description
Embodiments of the present invention relate to the analysis of IoT objects and IoT services present in an IoT environment.
The Internet of Things (IoT) is a technology for connecting various physical and virtual objects to provide more convenient and advanced services. In an IoT environment, a large number of objects and services can be concurrently connected to various IoT platforms and released. Therefore, faster and more accurate search and analysis techniques for objects and services are essential skills in the IoT environment.
Most importantly, a platform must be built to collect, store and manage the data provided by IoT objects and IoT services and the data generated from them.
On the other hand, it is not possible to create value only by collecting and storing data generated in the IOT environment. It is necessary to analyze and process these data to create and utilize value. In recent years, attempts have been made to perform IoT data analysis and information extraction through the use of mathematical theoretical approaches such as machine learning, analysis tools, and algorithm refinement.
In addition, query data for searching and using IoT data generated in the IoT environment is expected to increase explosively.
Embodiments of the present invention provide a method for enabling quick and accurate retrieval of valid IoT objects and IoT services in an IoT environment in which IoT objects and IoT services are dynamically connected and released.
Embodiments of the present invention analyze usage patterns between IoT objects, IoT objects, IoT objects and IoT services, IoT services, and IoT services, and analyze the relationship between them, and provide them to users.
The method of analyzing IoT objects and IoT services according to an embodiment of the present invention includes collecting IoT (Internet of Things) data; Generating abbreviated data from the collected IoT data; And generating a multi-class classification model by applying a single-class support vector machine (SVM) to the condensed data.
In one embodiment, the reduced data may be data obtained by applying a weight to high frequency words extracted from the collected IoT data, or may be data obtained by normalizing the weighted data.
In one embodiment, the single class SVM may be SVDD (Support Vector Data Description).
In one embodiment, the IoT data may include IoT object data and IoT service data.
In one embodiment, the method may further include analyzing the similarity between the IoT object and the IoT service using the multi-class classification model and the condensed data.
In one embodiment, analyzing the similarity may include analyzing the similarity between the IoT object and the IoT service using a cosine similarity function.
In one embodiment, the method may further comprise visualizing and providing the analyzed similarity.
In one embodiment, the method includes collecting new IoT data and generating abbreviated new data; Generating a single class SVM corresponding to the abbreviated new data when the single class SVM corresponding to the abbreviated new data does not exist in the multiple class classification model; And reflecting the generated single class SVM to the multi-class classification model.
In one embodiment, the method may include determining the class having the largest relative distance on the feature space as the belonging class of the abbreviated new data.
The IoT object and IoT service analyzing apparatus according to an embodiment of the present invention includes an IoT data collecting unit for collecting IoT (Internet of Things) data; An IoT data extracting unit for generating condensed data from the collected IoT data; And an IoT object and IoT service classifier for applying a single class SVM (support vector machine) to the reduced data to generate a multi-class classification model.
In one embodiment, the apparatus may further include an IoT object and an IoT service analyzer for analyzing the similarity between the IoT object and the IoT service using the multi-class classification model and the reduced data.
In one embodiment, the IoT object and the IoT service analyzing unit may analyze the similarity between the IoT object and the IoT service using the cosine similarity function.
In one embodiment, the IoT object and the IoT service analyzing unit can visualize and provide the analyzed similarity.
In one embodiment, when the single class SVM corresponding to the abbreviated new data generated from the new IoT data does not exist in the multi-class classification model, the IoT object and the IoT service classifier classify the IoT object and the IoT service class corresponding to the abbreviated new data A single class SVM may be generated and the generated single class SVM may be reflected in the multi class classification model.
In one embodiment, the IoT object and the IoT service classifier may determine the class having the largest relative distance on the feature space as the belonging class of the reduced new data.
According to the embodiments of the present invention, when IoT data collected in an IoT environment in which IoT objects and IoT services are dynamically connected and released is not classified as a specific IoT object or IoT service, a new class (group) . Therefore, there is no need to re-learn the entire system to newly construct a classification model.
According to embodiments of the present invention, usage patterns between IoT objects existing in the IoT environment and services using the IoT objects can be analyzed and provided to the user.
FIG. 1 is a flowchart illustrating a multi-class classification model generation process according to an embodiment of the present invention. FIG.
FIG. 2 is a flowchart illustrating a multi-class classification model updating process according to an exemplary embodiment of the present invention. FIG.
3 is a block diagram for explaining an IoT object and an IoT service analyzing apparatus according to an embodiment of the present invention;
FIG. 4 is an exemplary diagram showing the results of analysis of the use status of IoT objects per IoT service,
FIG. 5 is an exemplary diagram showing a result of using pattern analysis of IoT objects based on a similarity matrix for each IoT service; FIG.
In the following description of the embodiments of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.
In an IoT environment, various IoT objects (including physical objects and virtual objects) can exchange information using various communication methods and various interfaces. IoT services can collect information from IoT objects or control IoT objects.
Embodiments of the present invention can automatically classify IoT objects and IoT services using SVM (support vector machine), a class of machine learning, and analyze the relationship between IoT objects and IoT services using them And the like.
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
1 is a flowchart illustrating a process of generating a multi-class classification model according to an embodiment of the present invention. Depending on the embodiment, at least one of the steps shown in Fig. 1 may be omitted.
In
The IoT data may include IoT object data and IoT service data.
The IoT object data may include, for example, at least one of specification data of the IoT object and status data of the IoT object. The specification data of the IoT object includes, for example, the identifier (MAC address or serial number) of the IoT object, the domain to which the IoT object belongs, the position of the IoT object, the keyword used to search for the IoT object and the actual measured value of the IoT object And may include at least one.
The IoT service data may include at least one of specification data of the IoT service and status data of the IoT service. The specification data of the IoT service may include at least one of, for example, a domain to which the IoT service belongs, a provider that provides the IoT service, and a keyword used to search for the IoT service. The status data of the IoT service may include the operating status information or the availability status information of the IoT service.
In
The abbreviated data may be, for example, data having a weight applied to the high frequency word extracted from the IoT data, or data obtained by normalizing the weighted data.
For example, the IoT object and the IoT service analyzing apparatus apply a TF-IDF (Term Frequency - Inverse Document Frequency) to IoT data expressed in a character (string) form, (Words with low weight) are removed, and high frequency words (words with high weight) can be extracted.
Here, N denotes the total number of objects or services provided in the IoT environment, and w (t i , d x ) denotes the weight of the index word (word) t i in the object or service d x .
The IoT object and the IoT service analyzing apparatus can perform normalization on the weight for each word. An example of obtaining the normalized data (level of satisfaction (v ')) is shown in Equation (2).
Min A is the minimum value among the weights of the words, max A is the maximum value among the weights of the words, new_min A is the minimum value of the range to be normalized (for example, 0.0), new_max A Means the maximum value of the range to be normalized (for example, 1.0).
In
That is, the IoT object and the IoT service analyzing apparatus can perform learning using the abbreviated data and generate a multi-class classification model for the IoT object and the IoT service classification.
Recently SVM, which is actively studied as classification and prediction theory in machine learning field, is based on statistical learning theory. The SVM shows very good performance because the solution to a given problem is obtained by using the convex quadratic problem (CQP), in which the global optimal solution is always guaranteed.
However, due to the functional limitations of the binary classifier, the binary class SVM can not be used directly if the given problem is a multi-class problem such as the classification of IoT objects and services to be dealt with in the embodiments of the present invention. Therefore, it is a common practice to design a multi-class SVM by organically combining several binary class SVMs. However, when a multi-class SVM is designed using a binary class SVM, it is highly likely that new data will be misclassified because a decision boundary including an unobserved area is generated. Therefore, when designing a multi-class SVM, it is advantageous to select a decision boundary with a single class SVM that represents only the corresponding class independently.
Therefore, in embodiments of the present invention, a multi-class SVM is designed based on Support Vector Data Description (SVDD), which is a typical algorithm of a single class SVM, so that IoT objects and IoT services can be independently classified.
The single class classification method using SVDD is as follows.
1, ..., N k , k = 1, ..., K) of K learning data residing on the d-dimensional input space (D k = {x i k ∈R d | (Where D k means the kth learning data set, x i k means the ith learning data of the kth set, and N k means the number of learning data of the kth set) The problem of classifying a class corresponding to each set is defined as a problem of obtaining a sphere that minimizes volume while including learning data of each class, and the problem can be formulated as Equation (3).
Here, a k is a center of a sphere expressing a k-th class, R k 2 is a square of a radius of a sphere, and ξ i k is a penalty point indicating a degree of deviation of the i-th learning data belonging to the k- C is a trade-off constant that adjusts relative importance.
Applying a Lagrange function L to obtain a dual problem on Equation (3) can be expressed as Equation (4).
<Equation 4> should satisfy R k 2, a k, ξ i have a minimum value for the k variables, and the variable α k, so to have the maximum value for η k, the condition of the following <Equation 5> do.
Therefore, by substituting the expression (5) into the Lagrangian function L, the double problem defined by (6) can be obtained.
Spheres defined on the input space can represent only very simple shapes. To overcome this limitation, a sphere defined on a high dimensional feature space F defined through a kernel function k can be used. Since independent classes of objects or services can express their boundaries more accurately in their respective feature spaces, the learning of the system can be expressed by Equation (7) in consideration of the independence of the feature space to which each class is mapped This can be done by getting the answer to the CQP problem in question.
In this case, when a Gaussian kernel is used, k (x, x) = 1 holds, so that Equation (7) can be simplified as Equation (8).
Therefore, in the application process after completion of the learning, the determination function of each object class or service class can be defined as Equation (9).
On the other hand, the value of the output f k (x) of a single SVM defined in different feature space means the absolute distance from the test data to the boundary of the feature space of each class. Therefore, It is not desirable to determine the class. Thus, as defined in Equation (10), the absolute distance f k (x) on the feature space is divided by the radius R k of the spherical sphere defined in the feature space,
And the class having the largest relative distance can be determined as the belonging class of the input data x.
In
For example, the IoT object and the IoT service analyzing apparatus use the IoT object data (which may be abbreviated data) of the IoT objects used by the IoT services as the input of the weight matrix, and use the cosine similarity function Can be used to generate and analyze service-specific similarity matrices.
Here, t i denotes the i-th object, and S x and S y denote the respective services.
In
The multi-class classification model described with reference to FIG. 1 may be updated based on new IoT data periodically collected in a situation where IoT objects and IoT services are dynamically connected and released. This will be described with reference to FIG.
2 is a flowchart illustrating a process of updating a multi-class classification model according to an exemplary embodiment of the present invention. Depending on the embodiment, at least one of the steps shown in Fig. 2 may be omitted.
In
In
In
If the abbreviated new data does not belong to the existing class, then in
In
According to the embodiment of the present invention, when the IoT objects and the IoT services are dynamically connected and released, if the IoT data not belonging to the existing class is collected, only the classification model for the new class is generated, It can be reflected in the model. Therefore, it is not necessary to reconstruct the entire multi-class classification model.
3 is a block diagram for explaining an IoT object and an IoT service analyzing apparatus according to an embodiment of the present invention.
3, an IoT object and IoT service analyzing apparatus according to an embodiment of the present invention includes an IoT
The IoT
The IoT
The IoT object and
The IoT object and IoT
FIG. 4 is a diagram illustrating an analysis result of the use status of IoT objects per IoT service.
FIG. 4 shows frequency of use of IoT objects used in each of the IoT services. Referring to FIG. 3, a service having a close relationship among the IoT services can be identified. For example, IoT service called Smart Home and IoT service called Energy & Power have similar frequency of use of IoT objects. This means that these services have a close relationship with each other.
FIG. 5 is a diagram illustrating an analysis result of a usage pattern of IoT objects based on a similarity matrix for each IoT service.
5, the size of a node (indicated by a circle) means the number of IoT objects using the corresponding IoT service or the amount of traffic generated by IoT objects using the IoT service, and the thickness of the link connecting each node is represented by IoT Service.
For example, IoT services such as Smart Home and IoT services such as Energy & Power have a large number of IoT objects to be used, or a large amount of traffic generated by the IoT objects they use and a relatively close relationship .
On the other hand, in the case of IoT service called Healthcare, since the number of IoT objects to be used or the amount of traffic generated by the IoT objects used is large, there is no connection node with other IoT services. Group.
As a result, the usage pattern analysis result of IoT service by IoT service can be used as meaningful data in the process of discovering and using service of IoT service developer and user and creating new service through fusion of services, It can be expected to be used as a scientific basis for the operation and management of IoT services.
The embodiments of the invention described above may be implemented in any of a variety of ways. For example, embodiments of the present invention may be implemented using hardware, software, or a combination thereof. When implemented in software, it may be implemented as software running on one or more processors using various operating systems or platforms. Additionally, such software may be written using any of a number of suitable programming languages, and may also be compiled into machine code or intermediate code executable in a framework or virtual machine.
Also, when embodiments of the present invention are implemented on one or more processors, one or more programs for carrying out the methods of implementing the various embodiments of the invention discussed above may be stored on a processor readable medium (e.g., memory, A floppy disk, a hard disk, a compact disk, an optical disk, a magnetic tape, or the like).
Claims (18)
Generating abbreviated data from the collected IoT data; And
Applying a single class SVM (support vector machine) to the reduced data to generate a multi-class classification model
And an IoT service analysis method.
The weighted data may be data obtained by applying a weight to high frequency words extracted from the collected IoT data, or data obtained by normalizing the weighted data
How to analyze IoT objects and IoT services.
SVDD (Support Vector Data Description)
How to analyze IoT objects and IoT services.
IoT object data and IoT service data
How to analyze IoT objects and IoT services.
Analyzing the similarity between the IoT object and the IoT service using the multi-class classification model and the reduced data
And an IoT object analyzing method.
Analyzing the similarity between the IoT object and the IoT service using a cosine similarity function
And an IoT service analysis method.
Visualizing and providing the analyzed similarity
And an IoT object analyzing method.
Collecting new IoT data and generating abbreviated new data;
Generating a single class SVM corresponding to the abbreviated new data when the single class SVM corresponding to the abbreviated new data does not exist in the multiple class classification model; And
Reflecting the generated single class SVM to the multi-class classification model
And an IoT object analyzing method.
Determining a class having the largest relative distance on the feature space as a class of belonging to the reduced new data
≪ / RTI >
An IoT data extracting unit for generating condensed data from the collected IoT data; And
An IoT object and an IoT service classifier for applying a single class SVM (support vector machine) to the reduced data to generate a multi-
And an IoT service analyzing device.
The weighted data may be data obtained by applying a weight to high frequency words extracted from the collected IoT data, or data obtained by normalizing the weighted data
How to analyze IoT objects and IoT services.
SVDD (Support Vector Data Description)
IoT objects and IoT service analysis devices.
IoT object data and IoT service data
IoT objects and IoT service analysis devices.
An IoT object analyzing unit for analyzing the similarity between the IoT object and the IoT service using the multi-class classification model and the reduced data,
The IoT object and the IoT service analyzing apparatus.
The degree of similarity between the IoT object and the IoT service is analyzed using the cosine similarity function
IoT objects and IoT service analysis devices.
The analyzed degree of similarity is visualized and provided
IoT objects and IoT service analysis devices.
If the single class SVM corresponding to the abbreviated new data generated from the new IoT data does not exist in the multi class classification model, generates a single class SVM corresponding to the abbreviated new data, The multi-class classification model
IoT objects and IoT service analysis devices.
The class having the largest relative distance on the feature space is determined as the belonging class of the reduced new data
IoT objects and IoT service analysis devices.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020150163542A KR20170059546A (en) | 2015-11-20 | 2015-11-20 | Automatic analysis apparatus of IoT things and IoT services and method thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020150163542A KR20170059546A (en) | 2015-11-20 | 2015-11-20 | Automatic analysis apparatus of IoT things and IoT services and method thereof |
Publications (1)
Publication Number | Publication Date |
---|---|
KR20170059546A true KR20170059546A (en) | 2017-05-31 |
Family
ID=59052460
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
KR1020150163542A KR20170059546A (en) | 2015-11-20 | 2015-11-20 | Automatic analysis apparatus of IoT things and IoT services and method thereof |
Country Status (1)
Country | Link |
---|---|
KR (1) | KR20170059546A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20190050519A (en) | 2017-11-03 | 2019-05-13 | 현대자동차주식회사 | Method and apparatus for controlling iot devices engaged with vehicle |
KR20200044200A (en) * | 2018-10-10 | 2020-04-29 | 전자부품연구원 | Method and system for distributed operation between cloud and edge in IoT comuting environment |
KR20210060830A (en) * | 2019-11-19 | 2021-05-27 | 주식회사 피씨엔 | Big data intelligent collecting method and device |
US11539529B2 (en) | 2020-05-27 | 2022-12-27 | Wipro Limited | System and method for facilitating of an internet of things infrastructure for an application |
-
2015
- 2015-11-20 KR KR1020150163542A patent/KR20170059546A/en unknown
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20190050519A (en) | 2017-11-03 | 2019-05-13 | 현대자동차주식회사 | Method and apparatus for controlling iot devices engaged with vehicle |
KR20200044200A (en) * | 2018-10-10 | 2020-04-29 | 전자부품연구원 | Method and system for distributed operation between cloud and edge in IoT comuting environment |
KR20210060830A (en) * | 2019-11-19 | 2021-05-27 | 주식회사 피씨엔 | Big data intelligent collecting method and device |
US11539529B2 (en) | 2020-05-27 | 2022-12-27 | Wipro Limited | System and method for facilitating of an internet of things infrastructure for an application |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110363449B (en) | Risk identification method, device and system | |
US10621284B2 (en) | Training data update | |
US10387430B2 (en) | Geometry-directed active question selection for question answering systems | |
CN113614748A (en) | Systems and methods for incremental learning for object detection | |
US10977445B2 (en) | Weighting features for an intent classification system | |
US11875253B2 (en) | Low-resource entity resolution with transfer learning | |
US20190236135A1 (en) | Cross-lingual text classification | |
US11373117B1 (en) | Artificial intelligence service for scalable classification using features of unlabeled data and class descriptors | |
CN108629358B (en) | Object class prediction method and device | |
KR20170059546A (en) | Automatic analysis apparatus of IoT things and IoT services and method thereof | |
US20200285636A1 (en) | Context-sensitive cross-lingual searches | |
US11748393B2 (en) | Creating compact example sets for intent classification | |
CN111062431A (en) | Image clustering method, image clustering device, electronic device, and storage medium | |
CN113822315A (en) | Attribute graph processing method and device, electronic equipment and readable storage medium | |
JP2023543150A (en) | Expressive machine learning methods, systems, and programs for product design | |
KR20220050915A (en) | Rare topic detection using hierarchical clustering | |
US20200380405A1 (en) | Data exposure for transparency in artificial intelligence | |
US20200019608A1 (en) | Linked data seeded multi-lingual lexicon extraction | |
US11556558B2 (en) | Insight expansion in smart data retention systems | |
US11762896B2 (en) | Relationship discovery and quantification | |
CN111259975B (en) | Method and device for generating classifier and method and device for classifying text | |
CN114638308A (en) | Method and device for acquiring object relationship, electronic equipment and storage medium | |
US20220156297A1 (en) | Efficient and compact text matching system for sentence pairs | |
US11074591B2 (en) | Recommendation system to support mapping between regulations and controls | |
Mukherjee et al. | Predictive framework for crime data analysis using a hybrid logistic regression—support vector machine based ensemble classifier powered by CART (LR-SVM CART) |