KR101267561B1 - Method for providing context aware service using mobile terminal and system thereof - Google Patents

Abstract

A situation recognition system according to an embodiment of the disclosed technology includes a mobile terminal having a plurality of sensors and providing sensing data collected from the plurality of sensors; And a situation recognition server generating a situation model through learning based on the sensing data, inferring the situation of the mobile terminal based on the situation model, and providing recommendation information associated therewith.

Description

Context aware service providing method and system using mobile terminal {METHOD FOR PROVIDING CONTEXT AWARE SERVICE USING MOBILE TERMINAL AND SYSTEM THEREOF}

The disclosed technology relates to a situation recognition service technology, and in particular, collects information about a user using a mobile terminal including various sensors and provides a recommendation service to a user based on learning about situation recognition based on the collected information. It relates to a situation awareness service technology using a mobile terminal.

Various technologies for life-logs that record individual lives as digital information are being developed, and interest in context-aware services using them is increasing. These technologies store a variety of information that can characterize an individual's behavior and provide context-aware services based on that. However, in the related art, in order to collect information from an individual, there has been a burden of installing various sensor devices based on a domain. In the prior art using a mobile terminal, there is a problem in that it is difficult to provide a practically effective situation recognition service due to a limitation of information collected from the mobile terminal and lack of an effective learning process for such information.

Among the embodiments, the situation recognition system has a plurality of sensors, a mobile terminal for providing sensing data collected from the plurality of sensors and the situation model through the learning based on the sensing data, and generates the situation model And a situational awareness server for inferring the situation of the mobile terminal and providing recommendation information associated therewith. In one embodiment, the mobile terminal may collect and provide the sensing data every set time. In one embodiment, the situation recognition server may include an averaging module for averaging the sensing data collected at each set time by a sliding window method, and a clustering module for dividing the averaged sensing data into a plurality of groups. In one embodiment, the context recognition server may include a context learning module for extracting a topic using the LDA model or the Bigram topic model for the clustered sensing data, and generating a context model using the extracted topic. . In an embodiment, the context recognition server may include a context inference module that selects a context model corresponding to the clustered sensing data and generates recommendation information related to the selected context model.

Among the embodiments, the context aware service method is performed in a context aware server that receives a plurality of sensing data from a mobile terminal and generates a context model. The situation recognition service method includes (a) storing a plurality of sensing data received from the mobile terminal, (b) performing a preprocessing for performing a learning on the stored sensing data, and (c) the preprocessing. And extracting at least one topic with respect to sensing data and generating the situation model using the extracted topic. In an embodiment, the step (b) may include (b-1) averaging the stored sensing data for a set unit time by a sliding window method and (b-2) dividing the averaged sensing data into a plurality of groups. Grouping may be included. In an embodiment, the step (c) may include (c-1) extracting at least one topic for the preprocessed sensing data and (c-2) using a distribution of the extracted at least one topic. The method may include generating the situation model by dividing the preprocessed sensing data. In an embodiment, the step (c) may generate the situation model by extracting the at least one topic using an LDA model or a Bigram topic model.

Among the embodiments, the context aware service method is performed in a context aware server that performs context awareness of the mobile terminal based on sensing data received from the mobile terminal. The context aware server has the context model database generated by learning. The context recognition service method may include (a) performing preprocessing on sensing data received from the mobile terminal, (b) extracting at least one topic on the preprocessed sensing data, and corresponding to the at least one topic. Generating a first situation model, (c) inferring the first situation model to select the situation model from the situation model database, and (d) providing recommendation information for the selected situation model; Include. In an embodiment, the step (a) may include grouping the received sensing data into a plurality of groups. In an embodiment, the step (c) may include (c-1) calculating a post-distribution for the first situation model and (c-2) performing approximate inference based on a sampling based on the post-distribution model for the corresponding situation. Selecting a model from the context model database.

1 is a graph according to an embodiment of a latent rich allocation (LDA) model that may be applied to the disclosed technology.
2 is a graph that can compare the LDA and Bigram topic model to which the disclosed technology can be applied.
3 is a block diagram of an embodiment of a context aware system to which the disclosed technology may be applied.
4 is an exemplary view showing an embodiment of an interface of a user recording unit.
5 is a block diagram of an embodiment of a context aware server to which the disclosed technology may be applied.
6 is a block diagram of an embodiment of a context aware service method to which the disclosed invention can be applied.
FIG. 7 is a flowchart illustrating a context awareness service method of FIG. 6.
8 is a block diagram of an embodiment of a context aware service method to which the disclosed invention can be applied.
FIG. 9 is a flowchart illustrating a context awareness service method of FIG. 8.
10 is a topic distribution diagram illustrating the types of behaviors by day of the week.

The description of the disclosed technique is merely an example for structural or functional explanation and the scope of the disclosed technology should not be construed as being limited by the embodiments described in the text. That is, the embodiments may be variously modified and may have various forms, and thus the scope of the disclosed technology should be understood to include equivalents capable of realizing the technical idea.

Meanwhile, the meaning of the terms described in the present application should be understood as follows.

The terms "first "," second ", and the like are intended to distinguish one element from another, and the scope of the right should not be limited by these terms. For example, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

It should be understood that the term "and / or" includes all possible combinations from one or more related items. For example, the meaning of "first item, second item and / or third item" may be presented from two or more of the first, second or third items as well as the first, second or third item It means a combination of all the items that can be.

It is to be understood that when an element is referred to as being "connected" to another element, it may be directly connected to the other element, but there may be other elements in between. On the other hand, when an element is referred to as being "directly connected" to another element, it should be understood that there are no other elements in between. On the other hand, other expressions describing the relationship between the components, such as "between" and "immediately between" or "neighboring to" and "directly neighboring to", should be interpreted as well.

It should be understood that the singular " include "or" have "are to be construed as including a stated feature, number, step, operation, component, It is to be understood that the combination is intended to specify that it does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

In each step, the identification code (e.g., a, b, c, etc.) is used for convenience of explanation, the identification code does not describe the order of each step, Unless otherwise stated, it may occur differently from the stated order. That is, each step may occur in the same order as described, may be performed substantially concurrently, or may be performed in reverse order.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. Commonly used predefined terms should be interpreted to be consistent with the meanings in the context of the related art and can not be interpreted as having ideal or overly formal meaning unless explicitly defined in the present application.

Hereinafter, a technology for analyzing a user's behavior pattern using data collected from a mobile terminal including various sensors such as a smart phone, and providing a customized service using context awareness to the user based on such analysis.

The disclosed technique may perform learning and situational awareness using a late diallet allocation (LDA) model or a bigram topic model to perform situational awareness on various sensor data collected.

1 is a graph according to an embodiment of an LDA model that may be applied to the disclosed technology. Hereinafter, an LDA model applied in the disclosed technology will be described with reference to FIG. 1.

로 구성될 수 있다. 각 단어는 우리가 관측할 수 없는 주제 z로부터 생성된다고 가정할 수 있다. 예컨대, 하나의 문서는 다수의 주제로 구성되어 있으며, 문서의 주제 분포에 따라 생성되는 단어의 빈도가 결정될 수 있다. 여기에서 하나의 문서가 생성되는 과정을 정리하면 다음과 같다.The entire data consists of D documents, where each document d is N _d words

≪ / RTI > We can assume that each word comes from a subject z that we cannot observe. For example, one document is composed of a plurality of subjects, and the frequency of generated words may be determined according to the subject distribution of the documents. Here is a summary of the process of creating a single document:

선택One.

Selection

2. For the N words that make up a document:

선택A) subject

Selection

선택S) word

Selection

를 디리슐레(Dirichlet) 분포로 정의할 수 있다. 그 후, 문서 내의 모든 단어에 대해 상기 주제 선택 과정과 단어 선택 과정을 반복할 수 있다. 주제 분포

를 매개변수로 하는 다항(multinomial) 분포로부터 주제 인덱스를 선택하고, 선택된 주제 인덱스를 기반으로 다항분포

에서 단어 인덱스를 선택할 수 있다. If the entire document can be expressed in K subjects, first, the distribution of the subjects contained in one document

Can be defined as the Dirichlet distribution. Thereafter, the subject selection process and the word selection process can be repeated for all words in the document. Topic distribution

Select a subject index from a multinomial distribution whose parameters are, and multinomial distribution based on the selected subject index.

You can select the word index from.

The disclosed technique can be applied to the field recognition situation using the LDA described above, as well as document classification as well as learning the relationship between the subjects through learning. For example, LDA is a learning method designed for document modeling, but the disclosed technique can be variously applied to other kinds of discrete data. For example, a Bayesian probabilistic model may be applied to learn hidden variables in image data, video data, audio data, data collected from a gravitational acceleration sensor, or a location information sensor.

The disclosed technique can perform inference by calculating the probability for unobserved data given observation data in a Bayesian probability model. Using the document model as an example, it can be inferred by inferring the distribution of the subject of the document, the subject to which each word belongs, and the post-distribution of the distribution of the words of each subject. The disclosed technique can calculate the posterior distribution using Equation 1. Where H is a model and D is a random variable representing the data.

In order to calculate the posterior distribution according to [Equation 1], it is necessary to know the joint distribution and the marginal distribution for the data. As shown in the graphical representation of the LDA of FIG. 1, the post-distribution for one document can be summarized as in Equation 2 below.

를 은닉 변수

에 대해서 주변화(marginalize)한

를 계산-[수학식 3]-함으로써, 사후 분포를 계산할 수 있다. Here, the bond distribution

Conceal variables

Marginalized for

The post-distribution can be calculated by calculating-[Equation 3]-.

는

를 함께 고려하는 경우

개의 상태 공간(state space) 값에 대한 계산이 현실적으로 어려울 수 있으며, 이에 대하여 사후 분포를 계산하기 위해서는 근사 추론을 할 수 있다.Ambient Distributions for Hidden Variables

The

If you consider

It may be difficult to calculate the state space values of dogs, and approximation inference may be performed to calculate post-distribution.

Hereinafter, an approximation inference method that can be applied in the disclosed technology will be described. As an approximation reasoning method, a deterministic approximation based reasoning or the other sampling based reasoning may be used. Here, the Gibbs sampler will be described as one of sampling-based reasoning.

에 대한 사후 분포를 결정하기 위해서는 많은 계산을 필요로 하기 때문에, 가장 관심있는 확률 변수인 주제 z에 대한 사후 분포로 결정한다. 다른 은닉 변수인

에 대해 각각 적분한 결과로 얻을 수 있는 가장 관심 있는 은닉 주제 z에 대한 사후 분포를 이하의 [수학식 4]로서 정의할 수 있다.First all hidden variables

Since a lot of calculations are needed to determine the post-distribution for, we decide the post-distribution for the subject z, which is the most interesting random variable. Another hidden variable

The posterior distribution of the most interesting hidden subject z that can be obtained as a result of integrating for each can be defined as Equation 4 below.

For the Gibbs sampler, this may be more effective when there are several unobserved random variables. Assuming that one document is composed of N words, the posterior distribution for the N hidden subjects can be calculated as [Equation 5].

If the process of determining the subject of all N words is called the sample (topic) extraction process, it is possible to calculate the post-distribution by repeating the sampling process by a predetermined number of repetitions, and when the words are given, You can see what the topic is. As such, in order to extract a sample, it is necessary to define a full conditional distribution, and a full conditional distribution can be derived as shown in Equation 6 from the relationship between the Dirichlet-polynomial distribution. .

는 문서에서 n 번째 등장하는 단어의 주제가 k일 때, n 번째 단어 v를 제외한 나머지 모든 단어의 주제가 k인 경우의 빈도를 나타내며,

는 문서 d 의 n 번째 단어를 제외하고 문서에서 주제가 k인 경우의 빈도를 나타낸다.From here

Denotes the frequency when the subject of the nth word in the document is k, and the subject of all words except the nth v is k.

Denotes the frequency of the subject k in the document, except for the nth word of document d.

를 추정할 수 있으며, 이는 [수학식 7]과 같이 표현될 수 있다.Stochastic variables excluded from the sampling process using samples obtained by Gibbs sampler

It can be estimated, which can be expressed as Equation 7.

와

는 다항 분포이며,

와

는 디리슐레 분포이다. 그리고

를 전개해서 얻은 결과를 이용하면 확률변수

,

를 [수학식 8]을 이용하여 얻을 수 있다.From here,

Wow

Is a polynomial distribution,

Wow

Is the Dirichlet distribution. And

Using the result obtained by expanding the probability variable

,

Can be obtained using Equation 8.

에만 영향을 받으며, 단어들 사이에는 연관 관계가 없는 것으로 가정하고, BT 모델에서는

뿐만 아니라 n 번째 단어는 이전 단어인 n-1 번째 단어에 의해서도 영향을 받는다고 가정한다.2 is a graph that can compare the LDA and Bigram topic model to which the disclosed technology can be applied. In Figure 2 (a) is a graph of the LDA model, (b) is a graph of the Bigram topic model (hereinafter, BT model). Where LDA is the observation word

Is only affected, and it is assumed that there is no association between words.

In addition, the nth word is assumed to be affected by the n-1th word, the previous word.

가 주어지는 경우 단어

을 추론할 수 있다.If you need more than two higher order Markov chain assumptions between words, you can solve this by connecting contiguous bigrams. If you use a model that takes into account the dependencies between words,

If given word

Can be deduced.

The combined distribution of the words and subjects of the entire document in the BT model can be expressed as shown in [Equation 9].

In order to infer the hidden subjects in the BT model, the complete conditional distribution of the Gibbs sampler can be derived using Equation 10 from the defined binding distribution.

를 추정할 수 있다.Here, the BT model may perform the sampling process in consideration of the n-1th word v 'when inferring the subject of the nth word v. Based on the sample obtained by performing the Gibbs sampler as described above, By using Equation 11, the hidden random variable

Can be estimated.

Hereinafter, a technique for collecting sensing data and performing probabilistic situation modeling using the above-described topic model will be described with reference to FIG. 3.

3 is a block diagram of an embodiment of a context aware system to which the disclosed technology may be applied.

Referring to FIG. 3, the context aware system may include a mobile terminal 110 and a context aware server 120.

The mobile terminal 110 may include a plurality of sensors, collect various sensing data using the same, and provide the collected sensing data to the situation recognition server 120.

In one embodiment, the mobile terminal 110 is implemented as a smart phone including at least one of a Global Positioning System (GPS) module, a proximity sensor, a motion sensor, an illuminance sensor, a gyro sensor, a microphone module, a WiFi module, and a Bluetooth module. Can be. Here, the sensing data includes position and speed data, proximity information, motion variation, brightness per unit area, the direction of the mobile phone, the strength of the magnetic field, the size of the ambient sound, the remaining battery capacity and the charging cable connection, searchable wireless Internet information, It may include at least one of information about a nearby Bluetooth device and information about a running program. In one embodiment, the sensing data may be discrete data represented by a multidimensional value including the above-described information. Here, the sensing data may be heterogeneous in the values of each dimension, or the magnitudes of the values may be different from each other in the same shape (homoogeneous).

In one embodiment, mobile terminal 110 may include a user record. 4 is an exemplary view showing an embodiment of an interface of a user recording unit. The user recorder may receive information about a current action from the user and store the information in association with the above-described sensing data. For example, the mobile terminal 110 may receive the current behavior, the means of movement, the current location, the companion information from the user, and store the information in association with the sensing data.

In one embodiment, the mobile terminal 110 may collect the sensing data for each set periodic time. For example, the mobile terminal 110 may collect sensing data by using each sensor or the above-described module every 10 seconds.

5 is a block diagram of an embodiment of a context aware server to which the disclosed technology may be applied. Hereinafter, the situation recognition server 120 will be described with reference to FIG. 5.

The context recognition server 120 receives sensing data from the mobile terminal 110 and performs context recognition using the sensing data. In more detail, the context aware server 120 may include the sensing data database 510, the averaging module 520, the clustering module 530, the context learning module 540, and the context inference module 540. In one embodiment, the context aware server 120 may further include a recommendation information management module.

The sensing data database 510 may store sensing data received from the mobile terminal 110. Here, the sensing data includes data collected from various sensors, and the format of the data collected from each sensor may be different from each other, so that the averaging module 520 and the clustering module 530 may be used to convert them into the same type of data. ).

The averaging module 520 may average the sensing data collected every predetermined time. For example, the mobile terminal 110 may collect sensing data every 10 seconds, and since the periodically recorded information has a large amount of change, the averaging module 520 may perform averaging on the sensing data.

를 윈도우 W에 대하여 이동거리 s 만큼 윈도우를 이동시킴으로서 평균화를 수행할 수 있으며, 이는 [수학식 12]로 표현될 수 있다.In one embodiment, the averaging module 520 may perform smoothing in a sliding window manner. More specifically, sensing data

Can be averaged by moving the window by the moving distance s with respect to the window W, which can be expressed by Equation (12).

Here, when the number of collected data is N, the size of the window is | W | and the window moving distance is s, N / | W | averaged data can be obtained.

The clustering module 530 may classify the averaged sensing data into a plurality of groups and group them. In more detail, the averaged sensing data is still continuous and multidimensional, so it is not easy to apply the LDA model or the BT model. Accordingly, the clustering module 530 clusters the averaged sensing data and converts the sensing data into discontinuous data to facilitate learning and recognition.

가 있을 때, k-means 알고리즘을 통해 각 데이터를 거리가 가장 가까운 군집 중심점에 할당할 수 있고, 이는 [수학식 13]으로 표현될 수 있다. 여기에서 군집의 수 K는 LDA 모델이나 BT 모델에서 사용할 수 있는 고유한 단어의 수 V가 될 수 있다.In one embodiment, the clustering module 530 may cluster sensing data using a k-means algorithm. In more detail, the averaged data

When, the k-means algorithm can assign each data to the nearest cluster center point, which can be expressed by Equation (13). Here, the number K of clusters may be the number V of unique words that can be used in the LDA model or the BT model.

The context learning module 540 may generate a context model by using the averaging and clustered sensing data (hereinafter, referred to as preprocessed sensing data). In more detail, the contextual learning module 540 may generate a contextual model by extracting a topic on the preprocessed sensing data, performing sampling, and calculating a post-distribution. The calculation method of the post distribution was described above with reference to FIGS. Here, the generated situation model may be stored in the situation model database 560.

행렬 중

행렬을 이용하기 위해서 확률을 기준으로 내림차순 정렬을 수행할 수 있다. 여기에서, 다수의 주제에서 특정 단어(codeword)의 생성 확률이 높다면 주제별로 분류하거나 의미 분석을 수행할 때 비효율적일 수 있으므로, 상황 학습 모듈(540)은 다수의 주제에서 높은 확률을 갖는 단어들에 대하여 특정한 패널티를 부과할 수 있다. [수학식 14]는 이러한 단어 확률을 수정하기 위한 식이다.In one embodiment, contextual learning module 540 may use cast sampling for post-distribution learning. In more detail, the contextual learning module 540 is obtained through cast sampling.

In the procession

In order to use the matrix, the descending order may be performed based on the probability. Here, if the probability of generating a specific word (word) in a plurality of topics is high, it may be inefficient when sorting by topic or performing a semantic analysis, context learning module 540 is a word having a high probability in a plurality of topics You can impose specific penalties for. Equation (14) is for modifying such word probabilities.

The context model database 560 may store a context model generated by the context learning module 540. In one embodiment, the context model database 560 may further include recommendation information for the context model.

The context inference module 540 may infer the current situation of the corresponding terminal using the sensing data received from the mobile terminal 110 based on the learned situation model and provide recommendation information related thereto. In more detail, when sensing data is received from the mobile terminal 110, the situation recognition server 120 performs preprocessing using the averaging module 520 and the clustering module 530. The context inference module 540 may infer the current situation of the mobile terminal 110 by selecting the learned situation model corresponding to the preprocessed sensing data, and generate and provide recommendation information associated with the inferred situation model. A concrete method of performing situation inference has been described above with reference to FIGS. 1 and 2.

The recommendation information management module may generate the recommendation information by using the information received from the user recorder of the mobile terminal 110. For example, if the information input by the user is watching a movie, recommendation information for movie viewing (for example, nearby movie theater information, using a situation model related to location information on the sensing data and searchable Internet network information). Information about the screening movie, movie-related event information, etc.) may be generated.

6 is a block diagram of an embodiment of a context aware service method to which the disclosed invention can be applied. 6 illustrates a situation recognition service method of generating a context model by receiving sensing data from the mobile terminal 110 and performing training based on the sensing data.

Referring to FIG. 6, the context recognition service method includes storing a plurality of sensing data received from the mobile terminal 110, performing preprocessing for performing a learning on the stored sensing data, and preprocessing sensing data. Extracting at least one topic, and generating the situation model using the extracted topic.

FIG. 7 is a flowchart illustrating the context awareness service method of FIG. 6, and the context awareness service method will be described in more detail below with reference to FIG. 7.

When the situation recognition server 120 receives the sensing data from the mobile terminal 110 (step S710), the situation recognition server 120 may store the received plurality of sensing data (step S720). The context aware server 120 may average the sensing data stored for each set unit time (step S730). Here, the situation recognition server 120 may perform averaging using a sliding window method. Thereafter, the situation recognition server 120 may classify the averaged sensing data into a plurality of groups and group them (step S740) to perform preprocessing on the sensing data. The context recognition server 120 extracts at least one topic from the preprocessed sensing data (step S750), and generates the situation model by dividing the preprocessed sensing data using the extracted distribution of the at least one topic. (Step S760). Here, the situation recognition server 120 may generate a situation model by extracting a topic using an LDA model or a Bigram topic model.

8 is a block diagram of an embodiment of a context aware service method to which the disclosed invention can be applied. FIG. 8 discloses a situation recognition service method for recognizing a situation of the current mobile terminal 110 and providing recommendation information suitable for the situation based on the situation model generated by the method of FIG. 6.

Referring to FIG. 8, the context recognition service method includes performing preprocessing on sensing data received from the mobile terminal 110, extracting at least one topic from the preprocessed sensing data, and corresponding to at least one topic. Generating a model (hereinafter referred to as a first situation model), inferring the first situation model to select the situation model from the situation model database 560, and providing recommendation information for the selected situation model. do.

FIG. 9 is a flowchart illustrating the context recognition service method of FIG. 8, and the context awareness service method will be described in more detail with reference to FIG. 9.

When the situation recognition server 120 receives the sensing data from the mobile terminal 110 (step S910), the situation recognition server 120 may perform preprocessing on the received sensing data (step S920). In one embodiment, the situation recognition server 120 may divide the received sensing data into a plurality of groups and group the received sensing data. Thereafter, the context recognition server 120 may extract at least one topic with respect to the preprocessed sensing data (step S930), and generate a first context model corresponding to the extracted topic. Here, the first situation model is generated based on the received sensing data and may not exactly match the situation model stored in the situation model database 560. Therefore, the situation recognition server 120 performs inference with respect to the first situation model (the detailed method of inference is described in detail with reference to FIGS. 1 and 2), so that the situation model corresponding to the received sensing data can be selected from the situation model database. (Step S950). In more detail, the situation recognition server 120 calculates a post-distribution with respect to the first situation model, and performs approximation inference based on sampling based on the calculated post-distribution model to determine a situation model corresponding to the first situation model. Select from model database 560. Here, the selected situation model is a situation that is inferred as the situation of the current mobile terminal 110, the situation recognition server 120 may provide the mobile terminal 110 with recommendation information suitable for the inferred situation (step S960). ). In more detail, the context aware server 120 may include recommendation information for each context model. In one embodiment, the recommendation information may be included in the context model database 560 or may be stored as a separate storage module.

As one example to which the disclosed technology may be applied, it is possible to infer that the current user is on the way to work from the sensing data collected through the mobile terminal 110, and guide the fast way through the commute traffic information. You can provide a service that informs you about the movies being shown at movie theaters you use frequently.

Hereinafter, with reference to FIG. 10 will be described an experiment and results for an exemplary service that can be provided by the context aware service according to the disclosed technology.

In this case, an experiment is performed to show that the pattern can be found from the sensing data collected through the mobile terminal 110. One experiment was conducted using sensor data collected for two months, which was collected for about 60 days by one subject. Data collected for 35 days (approximately 600 hours) was used, except when the number of data collected during the day was too small.

10 is a topic distribution diagram illustrating the types of behaviors by day of the week.

The topic distribution diagram shown in FIG. 10 relates to the first data and shows a topic distribution map for 17 days of the week and 1 day of holidays selected from the 35 days of sensor data collected during daily life.

As shown in FIG. 10, five topics were calculated based on sensing data, and the distribution of the topics was found for each day of the week. Here, the Tuesday shows that the ratio of the topic 2 is 80% or more, and comparing the ratio with the rest of the topics shows that the Tuesday can be specified as the topic itself. In addition, in the case of Thursday, since the ratio of topic 3 is formed high and the ratio of the remaining topics is near, it can be confirmed that the topic distribution is Thursday. In this way, a situation model can be generated for each day of the week by using a topic distribution for the day of the week from the sensing data, the corresponding day is checked based on the collected sensing data, and the recommendation information suitable for the day is provided. can do. For example, if a user mainly watched a movie on a Thursday afternoon (which can be confirmed from information input by the user through the user log), and if Thursday is confirmed from the received sensing data, information related to the movie (recommended information) may be obtained. Can be provided to the user.

The disclosed technique may have the following effects. It is to be understood, however, that the scope of the disclosed technology is not to be construed as limited thereby, as it is not meant to imply that a particular embodiment should include all of the following effects or only the following effects.

The context recognition service providing method using the mobile terminal according to an embodiment may recognize current state of a user of the mobile terminal and provide appropriate information. This is because a plurality of sensor data of a mobile terminal can be used to recognize a user's situation model and provide appropriate recommendation information.

It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit and scope of the present invention as set forth in the following claims It can be understood that

Claims (12)

The mobile terminal comprising a plurality of sensors and providing the sensed data expressed in multi-dimensional value, which contains the data collected from the plurality of sensors; And
After averaging a plurality of data of different formats included in the sensing data at predetermined time intervals, and dividing them into a plurality of groups and discontinuously clustering the plurality of data of different formats to have the same format. Extract the topic for the preprocessed sensing data, perform sampling using LDA (Latent Dirichlet Allocaiont) model or Bigram Topic (BT) model to calculate post-distribution, create situation model, and based on the situation model And a context aware server for inferring the context of the context and providing recommendation information associated therewith.
The method of claim 1, wherein the mobile terminal
The situation recognition system, characterized in that for collecting the sensing data for each set time.
The method of claim 1, wherein the situation awareness server
An averaging module for averaging by using a sliding window method of moving and averaging a window having a size of W to a moving distance s for a plurality of data of different formats included in the sensing data; And
And a clustering module for classifying the plurality of data averaged by the averaging module into a plurality of groups in a K-means manner.
The number K of the clusters by the clustering module corresponds to the number V of unique words that can be used in the Latent Dirichlet Allocaiont (LDA) model or the Bigram Topic (BT) model.
The method of claim 3, wherein the situation awareness server
After the topic is extracted from the sensing data clustered by the clustering module, a post-learning is performed by imposing a penalty on words having a probability greater than or equal to a predetermined value in a plurality of topics by using Gibbs sampling on the extracted topic. And a context learning module for generating the latent rich allocaiont (LDA) model or the bigram topic model.
delete In the situation recognition service method of the situation recognition server that receives a plurality of sensing data from the mobile terminal to generate a situation model,
(a) storing a plurality of sensing data received from the mobile terminal;
(b) pre-processing the plurality of data of different formats into the same format by averaging the plurality of data of different formats included in the sensing data at predetermined time intervals and then grouping them into a plurality of groups discontinuously. Performing; And
(c) Extract the topic on preprocessed sensing data and perform sampling using Lateral Dirichlet Allocaiont (LDA) model or Bigram Topic (BT) model to calculate post-distribution, generate situation model, and based on the situation model Inferring a situation of the mobile terminal and providing recommendation information associated therewith;
Situational awareness service method comprising a.
7. The method of claim 6, wherein step (b)
Averaging a plurality of data of different types included in the sensing data by using a sliding window method of moving and averaging a window having a size of W to a moving distance s; And
A clustering step of dividing the plurality of data averaged by the averaging using the sliding window method into a plurality of groups by a K-means method;
The number K of clusters in the clustering step corresponds to the number V of unique words that can be used in the Latent Dirichlet Allocaiont (LDA) model or the Bigram Topic (BT) model.
delete delete A situation recognition service method performed in a situation recognition server that performs situation recognition of the mobile terminal based on sensing data received from a mobile terminal, wherein the situation recognition server includes the situation model database generated by learning.
(a) averaging the sensing data received from the mobile terminal at predetermined time intervals, and then performing preprocessing to divide the data into a plurality of groups and to discontinuously cluster the plurality of data having different formats in the same format. ;
(b) extracting at least one topic for the preprocessed sensing data and generating a first situation model using a latent rich allocaiont (LDA) model or a bigram topic (BT) model; And
(c) iteratively infers a posterior distribution by repeatedly performing a sampling process for a predetermined number of times using the Gibbs sampler with respect to the first situation model, and generates a first standard situation model corresponding to the first situation model by using a situation model database. Selecting from; And
(d) providing recommendation information for the selected first standard context model to the mobile terminal;
Situational awareness service method comprising a.
delete 11. The method of claim 10, wherein step (c)
(c-1) calculating a post distribution on the first situation model; And
and (c-2) performing approximation inference based on sampling on the posterior distribution to select a corresponding situation model from the situation model database.

Images