KR20140001266A - Method for recommending personalized contents based on naive bayesian filtering approach for ubiquitous health service of semantic environment and ubuquitous health service platform based on digital television using the same - Google Patents

Abstract

According to the embodiment of the present invention, a U-health service system capable of recommending U-health content includes: a health management center server which manages memberships, sends content, and provides a web service; and a U-health service platform device. The U-health service platform device is connected to the health management center server through a network, infers content for a user based on the status data including personal data or sensor data of the user, health management ontology, and a rule base, calculates the posterior probability that the user selects the inferred content through naïve bayesian filtering, and recommends the content with a relatively high posterior probability. [Reference numerals] (AA) User

Description

TECHNICAL FIELD [0001] The present invention relates to personalized content recommendation method based on Naive Bayesian filtering for providing a U-health service in a semantic environment, and a digital TV based U-health service platform using the same. BACKGROUND ART < RTI ID = 0.0 & UBUQUITOUS HEALTH SERVICE PLATFORM BASED ON DIGITAL TELEVISION USING THE SAME}

The present invention relates to a ubiquitous health service, and more particularly, to a ubiquitous health service based on a digital TV.

TV, which is the most important consumer electronics in everyday life, has developed into a digital TV with high-performance information processing capability and two-way communication capability, enabling users to provide functions or services that could not be expected in conventional analog TV. It has become possible to provide a ubiquitous health service, that is, a u-health service, which has been conventionally provided in a web environment, in a digital TV.

U-health services should provide personalized services as much as they provide services to individuals at home. For example, customized health content for individuals should be inferred based on information such as user body, region, location, weather, personal life pattern, disease, diet, etc. and recommended to the user.

To this end, intelligent u-health service was developed in ontology-based semantic environment. Such an ontology-based u-health service can analyze the meaning of the input data using the ontology, which is a rule-based filtering technique, and infer personalized contents to provide the contents to the user. However, There is a limit to providing recommended services.

A problem to be solved by the present invention is to provide a personalized content recommendation method based on Naïve Bayesian filtering and a digital TV using the same, in which a user can infer content preferred with high accuracy and recall in an ontology-based semantic environment .

In order to solve the above problems, a personalized content recommendation method based on Naïve Bayesian filtering and a digital TV based U-health service platform using the same provide a U-health service of a semantic environment, Providing the healthcare ontology and the rule base, and generating the situation data including the personal information or the sensor information on the user in the ontology language.

In addition, it is possible to infer content by applying the situation data to the health management ontology and the rule base, to perform the inference about whether or not to generate the contextual information about whether or not the inferred content is generated for each item, And constructing a learning transaction table including selection information on whether or not to be selected by the user from the information and the recommended contents.

Also, the method includes calculating a Bayesian posterior probability to be selected by the user from contents deduced through the Naive Bayesian filtering based on the context information, reasoning information, and selection information stored in the learning transaction table.

And selecting at least one content having a relatively large posterior probability as the recommended content.

In addition, the step of calculating the Bayesian posterior probability to be selected by the user may include calculating the content probabilities P (C | I) based on the context information C stored in the learning transaction table and the inference information I for the content inference information P (Ij) .

Also, for the preference reasoning information P (Sk), calculating preference probabilities P (C | S) based on the context information C and the selection information S stored in the learning transaction table and calculating the calculated content probabilities P S) to select a content of the inferred contents by the user using the probability P (C | I) and the preference probability P (C | S).

Furthermore, the present invention includes a situation data generation unit for generating situation data including personal information or sensor information about a user in an ontology language, an ontology management unit for managing a health management ontology and a rule base for reasoning of the u health service contents do.

In addition, ontology reasoning part for inferring contents by applying situation data to health management ontology and rule base, situation information on whether or not state data is generated for each item for Naïve Bayesian filtering, And a learning transaction DB for storing a learning transaction table constructed by including inference information and selection information on whether or not the recommended contents are selected by a user.

In addition, it is also possible to use a Naïve Bayesian filtering unit for calculating a Bayesian posterior probability to be selected by the user from the contents deduced through the Naïve Bayesian filtering based on the context information, reasoning information and selection information stored in the learning transaction table, And a content recommendation control unit for selecting one content as recommended content.

Furthermore, the present invention is networked with a health care center server and a health care center server that provides member management, content delivery, and web services, and the situation data and health care ontology including personal information or sensor information about a user. And a U-health service that infers content for a user based on a rule base, calculates a posterior probability of content selected by the user from the inferred content through naive Bayesian filtering, and recommends a content having a relatively high posterior probability to the user. It includes a platform device.

According to the personalized content recommendation method based on Naive Bayesian filtering of the present invention and the digital TV using the personalized content, it is possible to recommend personalized contents with higher accuracy and recall rate than existing ontology based u-health service, This improved effect can be obtained even if the rule base or situation data conversion process is used as it is.

According to the personalized content recommendation method based on the Naive Bayesian filtering of the present invention and the digital TV using the same, it is possible to provide contents that can be easily satisfied with the implementation and can satisfy the user's preference through learning without consuming a lot of resources, I can reason.

FIG. 1 is a conceptual diagram illustrating a digital TV-based U-health service system capable of recommending personalized content based on Naive Bayesian filtering according to an exemplary embodiment of the present invention.
FIG. 2 is a flowchart illustrating a personalized content recommendation method based on Naive Bayesian filtering according to an exemplary embodiment of the present invention.
3 is a diagram illustrating an ontology used in a personalized content recommendation method based on Naive Bayesian filtering according to an exemplary embodiment of the present invention.
FIG. 4 is a diagram illustrating a structure of a learning transaction table illustrated in a personalized content recommendation method based on Naive Bayesian filtering according to an exemplary embodiment of the present invention.
5 is a flowchart illustrating a Naïve Bayesian filtering procedure in a personalized content recommendation method based on Naïve Bayesian filtering according to an embodiment of the present invention.
FIG. 6 is a block diagram illustrating a digital TV apparatus implementing a u-health service platform capable of recommending personalized content based on Naive Bayesian filtering according to an embodiment.

For the embodiments of the invention disclosed herein, specific structural and functional descriptions are set forth for the purpose of describing an embodiment of the invention only, and it is to be understood that the embodiments of the invention may be practiced in various forms, The present invention should not be construed as limited to the embodiments described in Figs.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

FIG. 1 is a conceptual diagram illustrating a digital TV-based U-health service system capable of recommending personalized content based on Naive Bayesian filtering according to an exemplary embodiment of the present invention.

Referring to FIG. 1, a digital TV-based u-health service system 10 includes a digital TV-based u-health service platform 100, a sensor 200, a health care center 300, 400, and a content provider 500.

The U-health service platform 100 is operated, for example, in a digital TV set in a living room of a user's home and connected to the Internet, or in a set-top box for a digital TV.

The U-health service platform 100 receives personal data of a user from a user and collects health data of the user from the sensors 200 that are free of restraint and consciousness toward the user. The u-health service platform 100 generates situation data on the health status of the user based on the collected personal data and health data, deduces the personalized contents based on the situation data, the ontology and the rule base, It is possible to select the recommendable content that has been filtered through the Naive bezier according to the user's preference and to present it to the user. Since the U-health service platform 100 has all the functions of the digital TV, the recommended contents can be provided from the contents provider 500 and displayed on the screen.

The health care center 300 can provide technical services such as user member management, digital TV and content service, and web service.

The medical institution 400 may receive the disease data of the user from the health service platform 100 or the like according to a personal health record (PHR) standard to determine the prognosis and generate appropriate findings, And provide the configured content data to the user through the u-health service platform 100.

Through this collaboration relationship, the u-health system 10 can provide the u-health service systematically and effectively to the user.

FIG. 2 is a flowchart illustrating a personalized content recommendation method based on Naive Bayesian filtering according to an exemplary embodiment of the present invention.

Referring to FIG. 2, in step S21, a health management ontology and a rule base are constructed for reasoning service contents.

The health management ontology can be configured as shown in FIG. 3, for example. In the example of FIG. 3, there is a Thing class as a top class, and five classes next to each other, Device, Healthy, Disease_History, and Individual, are prepared. The Device class has the types and properties of health information measurement devices, and the instance of the Device class is measurement result values that each device can measure. The Healthy class consists of a state of user health and can have an instance of a health score. The Disease_History class consists of clinical outcomes measured at a medical institution such as a hospital and refers to data such as disease status, treatment, medication, and prescription. Finally, the Individual class has information such as the user's name, age, sex, lifestyle, smoking amount, and weight.

The rule base for content recommendation, in which the health care ontology is referred to, can be configured illustratively as shown in Table 1 below. The rule base in Table 1 is created by referring to the ACSM Guide Line 2010 edition. It is used to judge the user's body condition such as body mass index, hypertension, health level, Represents a rule.

division rule BMI (? In-body? Biosignal) ∩ (? Customer hasWeight? Weight) ∩
(? Customer hasHeight? Weight)
-> (? Customer hasBMI? BMI) High blood pressure (? Customer hasDisease? Hypertention) ∩ (? MaxBP <120 or MinBP <80)
-> (? Hypertention hasHypertention? Safety)
((? MaxBP> = 120 &&<139) or (MinBP> = 80 &&<89))
-> (? Hypertention hasHypertention? PreHypertention) Health level (? Customer hasBMI? BMI) ∩ (? BMI hasHealthLevel? Health_Level) ∩
(? Customer hasAge? Age) ∩ (? Age hasHealthLevel? Health_Level) ∩
(? Customer hasSex? Sex) ∩ (? Sex hasHealthLevel? Health_Level)
-> (? Customer hasRecommend_HealthLevel? Health_Level) Hypertension patient movement recommendation (? Customer hasHealthlevel? Health_Level5) ∩
(? Customer hasDisease_History? Hypertension)
-> (? ExcerciseRecommand hasExerciseRecommand? Exer5_Hypertension)
(? Customer hasHealthlevel? Health_Level1) ∩
(? Customer hasDisease_History? Hypertension)
-> (? ExcerciseRecommand hasExerciseRecommand? Exer1_Hypertension)

Next, in step S22, situation data is generated for a specific user. The situation data may include, for example, sensing data that can be detected by various unrestrained and non-greedy sensors, appropriate ontology language suitable for referencing the ontology such as personal data that is unsuitable for detection by the sensor, For example, OWL (Ontology Web Language) format.

Next, in step S23, the obtained content is applied to the ontology and the rule base to deduce the content.

In step S24, in order to perform the Naïve Bayesian filtering, context information about whether the situation data is generated for each item, inference information about whether or not the inferred content is generated for each item, and selection Build a learning transaction table with information.

FIG. 4 is a diagram illustrating a structure of a learning transaction table illustrated in a personalized content recommendation method based on Naive Bayesian filtering according to an embodiment of the present invention. Referring to FIG.

In FIG. 4, situation information C1, C2, C3,... Ci, regarding whether items of the situation data generated in step S22 are generated for each item, inference information I1, the content inferred in step S23, and the like. I2, I3, ..., Ij) and selection information (S1, S2, ..., Sk) about contents which the user selected or did not select among the contents recommended by the user in the later step S2xx. It is stored in the transaction table along with the index of each event occurrence sequence.

The context information, reasoning information, and selection information are written as 0 (false) or 1 (true), respectively. For example, among the total i kinds of situation data, the entry of the situation information item (C2, C3, Ci of the index 1 in the example of FIG. 4), in which the situation data is generated at an event, is 1, and the situation data is not generated. The entries of the contextual information item (C1 of index 1) are recorded in the learning transaction table as 0, respectively.

Similarly, when inferring through the ontology and rule base on the input context data, some of the total j total service contents controlled by the service manager will be inferred, so that the recommended information item of the recommended contents (for example, , The entry of I2) of index 1 is 1, and the recommendation information items of the contents that are not are 0 respectively.

Similarly, entries of the selection information items of the contents actually selected by the user among the k contents recommended to the user among the inferred contents are recorded in the learning transaction table as 1, otherwise as 0.

Returning to FIG. 2, in step S25, among the contents deduced through the Naive Bayesian filtering based on the context information C, reasoning information I, and selection information S stored in the learning transaction table, Calculates the probability P (I | S), and selects at least one content whose posterior probability is relatively large, as the recommended content.

5 is a flowchart illustrating a Naïve Bayesian filtering procedure in a personalized content recommendation method based on Naïve Bayesian filtering according to an exemplary embodiment of the present invention. Referring to FIG.

In Fig. 5, the Naïve Bayesian filtering proceeds to the following three steps.

First, content probabilities P (C | I) based on context information C and inference information I stored in the learning transaction table are calculated for content inference information P (Ij).

Next, for the preference reasoning information P (Sk), the preference probabilities P (C | S) based on the context information C and the selection information S stored in the learning transaction table are calculated.

Lastly, the posterior probabilities P (C | I) and the preference probability P (C | S) are calculated using the calculated content probability P (C | I) (I | S) is calculated.

Inferred contents having a relatively large posterior probability value among the computed posterior probabilities can be selected as contents to be recommended.

In detail, first, content inference information P (I _j ) is a probability that content is inferred from given context information, that is, a probability that context data is included in inferred content information, and thus P (C | I) as shown in Equation 1 below. Can be obtained by computing

Similarly, the preference inference information P (S _k ) has a probability of becoming a specific content preferred by the user in the given situation information, that is, the probability that the situation data will be included in the preference inference information, and thus P (C | S) Can be obtained by computing

In Equations 1 and 2, C is context data, I _j is inferred content, S _k is selected content, and Context means type of context data input.

After obtaining the content inference information P (C | I _j ) and the preference reasoning information P (C | S _k ), the Bayesian posterior probability P (I | S) can be obtained.

Referring to Equation 3, the // l-th content I _l (1≤l≤j) // deduced through the ontology based on the information accumulated in the training transaction table so far will be the content selected by the user. The posterior probability as to whether or not is calculated.

After the posterior probabilities have been calculated for all the contents I, it is recommended to the user that some content with a relatively large posterior probability value, for example, one content with the largest posterior probability value.

In Equation (3), the Naïve Bayesian inference is such that if the conditional probability value for the information for inference is 0, the entire inference value is calculated as 0, so that appropriate inference can not be made for such an item. In this case, the m-estimation technique known as one of the methods for solving this problem can be applied.

In equation 4, n is the situation data y _i practice could, n _c is a number of training examples in the status data y _i with x _i values, m is a parameter called the equivalent sample size parameters, p is the user-defined pre-probability parameters of the variable to be. If there is no learning set (n = 0), P (x _i | y _j ) = p. The equivalent sample size determines the trade-off between the prior probability p and the observation probability n _c and n.

FIG. 6 is a block diagram illustrating a digital TV apparatus implementing a u-health service platform capable of recommending personalized content based on Naive Bayesian filtering according to an embodiment.

6, the digital TV apparatus 60 includes a content recommendation control unit 61, a situation data generation unit 62, an ontology management unit 63, an ontology reasoning unit 64, a learning transaction DB 65 ), A navy Bayesian filtering unit 66, a content reproducing unit 67, and a display 68. [

The content recommendation control unit 61 may control the operation of the other components 62 to 68 in order to infer the user's preference and display the content that the user prefers to the user.

The situation data generation unit 62 receives personal data through a user interface, receives sensor data from at least one sensor through a sensor interface, and generates personal data and sensor data based on an ontology language such as OWL And provides it to the ontology reasoning unit 64. The ontology deducing unit 64 may be provided with an ontology deduction unit 64,

The ontology management unit 63 manages the health management ontology and the rule base, and the ontology reasoning unit 64 generates the content inference data by inputting the situation data provided from the situation data generation unit 62 into the health management ontology and the rule base .

When the content inference data is generated, the content recommendation control unit 61 receives the context information C about whether to generate the situation data, the reasoning information I of the content deduced by the ontology reasoning unit 64, 61 cumulatively stores the selection information (S) related to the content selected by the user in the learning transaction DB 65 when the recommendation is made to the user.

Based on the context information C, inference information I, and selection information S stored in the learning transaction DB 65, the navy Bayesian filtering unit 66 performs the Naïve Bayesian filtering &lt; RTI ID = 0.0 &gt; (I | S) to be selected by the user from among the inferred contents.

The content recommendation control unit 61 selects at least one content whose posterior probability is relatively large calculated as the recommended content from the computed posterior probability P (I | S) values.

The selected recommended content is displayed on the display 68. When the user selects a desired content from the recommended content through the user interface, the content recommendation control unit 61 transmits the selection information S about the selected content to the learning transaction DB 65 ).

The content recommendation control section 61 also controls the content reproduction section 67 to receive the actual data of the content selected by the user from an external content provider and reproduce it on the display 68. [

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, Modification is possible. Accordingly, the spirit of the present invention should be understood only in accordance with the following claims, and all of the equivalent or equivalent variations will fall within the scope of the present invention.

Further, the apparatus according to the present invention can be implemented as a computer-readable code on a computer-readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer system is stored. Examples of the recording medium include a ROM, a RAM, an optical disk, a magnetic tape, a floppy disk, a hard disk, a nonvolatile memory, and the like, and a carrier wave (for example, transmission via the Internet).

10 Digital TV based U-health service system
100 Digital TV-based U-Health Service Platform
200 sensors 300 health care center
400 medical institutions 500 content providers
60 U health service platform digital TV device
61 Content recommendation control unit 62 Situation data generation unit
63 ontology management unit 64 ontology reasoning unit
65 learning transaction DB 66 Naive Bayesian filtering unit
67 Content playback unit 68 Display

Claims (9)

Providing a health management ontology and a rule base for reasoning of the u health service contents;
Generating situation data including personal information or sensor information on a user in an ontology language;
Applying the context data to the healthcare ontology and the rule base to infer content;
For Naïve Bayesian filtering, context information about whether the situation data is generated for each item, reasoning information about whether the inferred content is generated for each item, and selection information about whether or not the user selects the recommended content among the recommended contents Building a transaction table;
Calculating Bayesian posterior probabilities to be selected by the user from among the inferred contents through Naïve Bayesian filtering based on context information, reasoning information and selection information stored in the learning transaction table; And
And selecting at least one content having a calculated posterior probability as a recommended content. 2. The method of claim 1, wherein calculating a Bayesian posterior probability for the user to select comprises:
Calculating content probabilities P (C | I) based on context information (C) and inference information (I) stored in the learning transaction table, for content inference information P (Ij);
Computing preference probabilities P (C | S) based on context information (C) and selection information (S) stored in the learning transaction table for preference reasoning information P (Sk); And
Calculating a Bayesian posterior probability P (I | S) for the user to select which content from the inferred contents using the calculated content probability P (C | I) and the preference probability P (C | S) Wherein the recommendation method comprises the steps of: The method of claim 2, wherein calculating the content probabilities P (C | I) for the content speculation information P (Ij)

Using
The step of calculating the preference probabilities P (C | S) for the preference reasoning information P (Sk)

Using
The step of calculating the Bayesian posterior probability P (I | S)

Using
Context is a kind of context data, and the U-health service content recommendation method of // 1 ≦ l ≦ j //. A recording medium on which a program for implementing in a computer a method for recommending a u-health service content according to any one of claims 1 to 3 is recorded. A situation data generation unit for generating, in an ontology language, situation data including personal information on a user or sensor information;
An ontology management unit for managing a health management ontology and a rule base for reasoning of the u health service contents;
An ontology reasoning unit for applying the situation data to the health care ontology and the rule base to deduce contents;
For the naive Bayesian filtering, context information about whether the situation data is generated for each item, reasoning information about whether the inferred content is generated for each item, and selection information about whether or not the user selects the recommended content among the recommended contents A learning transaction DB storing a learning transaction table;
A Naïve Bayesian filtering unit for calculating a Bayesian posterior probability to be selected by the user from among the inferred contents through Naïve Bayesian filtering based on context information, reasoning information and selection information stored in the learning transaction table; And
And a content recommendation control unit for selecting at least one content having a relatively large posterior probability calculated as recommended content. 6. The apparatus of claim 5, wherein the Naive Bayesian filtering unit comprises:
Calculates content probabilities P (C | I) based on context information C and inference information I stored in the learning transaction table for content inference information P (Ij), and for contention inference information P (Sk) (C | I) and the likelihood probability P (C | S) calculated based on the context information (C) stored in the learning transaction table and the selection information (S) ) To calculate a Bayesian posterior probability P (I | S) for the user to select a content of the inferred contents. The u-health service platform apparatus according to claim 5, wherein the u-health service platform device is a digital TV or a digital TV set-top box. A health care center server providing member management, content transfer and web service; And
It is networked with the health care center server, infers content for the user based on situational data including personal information or sensor information about the user, health care ontology and rule base, and infers the content through naive Bayesian filtering. A u-health service system capable of recommending a u-health service content, comprising a u-health service platform device which calculates a posterior probability of the content to be selected by the user among the prepared contents and recommends a content having a relatively high posterior probability to the user. The apparatus of claim 8, wherein the u-health service platform device comprises:
For the naive Bayesian filtering, context information about whether the situation data is generated for each item, reasoning information about whether the inferred content is generated for each item, and selection information about whether or not the user selects the recommended content among the recommended contents A learning transaction DB storing a learning transaction table; And
And a navy Bayesian filtering unit for calculating Bayesian posterior probabilities to be selected by the user from among the inferred contents through the Naïve Bayesian filtering based on the context information, reasoning information and selection information stored in the learning transaction table Recommended u-health service system.

Images