CN107172637B - Method and device for classifying calls - Google Patents

Abstract

The embodiment of the present application discloses a method and device for classifying calls, the method includes: acquiring a data packet, where the data packet includes N data sets corresponding to N calls respectively, and any call in the N calls Hi The corresponding data set includes all measurement reports MR sent by the communication terminal to the base station in turn during the call Hi process; each call in the N calls is divided into w preset time windows, and it is determined that each call is The corresponding feature set after each preset time window is split, the feature set includes the number of cells corresponding to the split call, the number of handovers, and the value of at least one self-defined feature; According to the corresponding feature set after classification, the N calls are classified according to the preset mobility state categories. In the embodiment of the present application, the type of the call can be determined according to the number of cells corresponding to the call, the number of handovers, and the self-defined feature, which is beneficial to improve the accuracy of classifying the call.

Description

A method and apparatus for classifying calls

technical field

The present application relates to the field of communication technologies, and in particular, to a method and apparatus for classifying calls.

Background technique

When performing network optimization, operators often need to use a processing device such as a server to classify the mobility status of each call in the acquired data packets. It should be noted that when using communication terminals such as mobile phones and tablet computers (portable android device, Pad) to make a call, we believe that one call process corresponds to two calls. Among them, the communication terminal that initiates the call corresponds to one call, and the communication terminal that receives the call also corresponds to a call. In addition, when portable communication devices such as mobile phones and pads access the Internet through the base station, one Internet access process also corresponds to one call. In the data packets obtained by the server, each call corresponds to a data set, and each data set includes all measurement reports (measurement reports, MRs) sent by the communication terminal to the base station during a call.

The inventor of the present application finds that when classifying calls, the prior art simply processes a data set corresponding to a call, and classifies them according to the acquired parameters of the number of cells and the number of handovers. Take a call corresponding to a scenario where a user uses a communication terminal to walk back and forth in a certain area as an example, the number of cells and handovers in the data set may be relatively large, and the existing technology may classify this call into the category of high-speed movement, which is obviously is inconsistent with the actual mobile state of the call. Therefore, the prior art is inaccurate in classifying calls.

SUMMARY OF THE INVENTION

The embodiments of the present application provide a method and apparatus for classifying calls, which are used to improve the accuracy of classifying calls.

In a first aspect, an embodiment of the present application provides a method for classifying calls, the method comprising:

Acquire a data packet, where the data packet includes N data sets corresponding to N calls respectively, and the data set corresponding to any call Hi in the N calls includes the communication terminal sending sequentially to the base station during the call Hi process All measurement reports MR of , wherein the N is an integer greater than 1, and the 1≤i≤N;

Each call in the N calls is segmented with w preset time windows, and a feature set corresponding to each call after being segmented by each preset time window is determined, and the feature set includes The number of cells corresponding to the divided call, the number of handovers, and the value of at least one self-defined characteristic, the self-defined characteristic is related to the movement state of the call;

According to the feature set corresponding to each call after being segmented by each preset time window, the N calls are classified according to preset mobility state categories.

In each embodiment of the present application, the at least one custom feature includes at least one of the following custom features: standard deviation of received signal strength, handover entropy, outdoor cell ratio, and inter-site distance speed; wherein,

The standard deviation of the received signal strength is the standard deviation of the received signal strength value of the call Hj;

the handover entropy, used to represent the uncertainty of the calling Hj accessing the cell;

The outdoor cell ratio is the percentage of the number of the call Hj accessing the outdoor type cells to the total number of all access cells;

The inter-station distance speed is the longest distance between the mean value of all the position information obtained by the call Hj and all the positions of the call.

In some possible implementations, the handover entropy is calculated according to the following formula:

Wherein, the entropy is the handover entropy corresponding to the call Hj, and the

where N represents the total number of cells accessed by the call Hj, i represents the ith cell accessed by the call Hj, #i represents the number of times the call Hj accesses the ith cell, and T represents the number of accesses to different cells. The total number of times, pi represents the probability of accessing the i-th cell during this period.

It can be seen that in the above technical solution, since a custom feature related to the mobility state of the call is set, when judging the mobility state, in addition to the number of cells corresponding to the call and the number of handovers, the custom feature can also be combined It is helpful to improve the accuracy when classifying calls.

In some possible implementations, depending on the nature of the access cell,

The received signal strength standard deviation in the preset feature set corresponding to the call Hj includes: the received signal strength standard deviation of the primary serving cell, the received signal strength standard deviation of neighboring cells, and the received signal strength standard deviation of all cells:

The handover entropy in the preset feature set corresponding to the call Hj includes: the entropy of the primary serving cell, the entropy of the adjacent cells, and the entropy of all cells.

In some possible implementations, when the data sets corresponding to N' calls in the N calls include precise geographic location information, the preset feature set corresponding to any one of the N' calls is further Including: average speed, wherein, the N'≥1.

In some possible implementations, the precise geographic location information includes location information obtained through AGPS or OTT positioning services.

In some possible implementations, when the data sets corresponding to N' calls in the N calls include precise geographic location information, the N calls are classified according to preset mobility state categories ,include:

Any call H _k in the N' calls determines the type of the call H _k according to its corresponding average speed and the average speed corresponding to the preset mobile state types respectively, wherein the 1≤K≤N ';

According to the category corresponding to each of the N' calls and the set of the feature sets corresponding to each of the N' calls, use a supervised learning algorithm to determine any two of the preset The limit of the category of the mobile state in the M-dimensional space, and according to the limit, the range of the distribution of any of the preset categories of the mobile state in the M-dimensional space is obtained, and the M is the number corresponding to the call H _k . Describe the number of features in the feature set;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

In some possible implementations, the N calls are classified according to preset mobility state categories, including:

When the N' calls include geographic information (Geography Information System, GIS) information,

Obtain the location of the feature information specified in the GIS information, and the specified feature information is the feature information strongly related to the preset movement state; Roads, intersections, highways or railways, etc.

Any call H _k among the N' calls determines the type of the call H _k according to its corresponding average speed and the speed corresponding to the preset mobile state type respectively, wherein the 1≤K≤N';

determining the feature information matched by any of the calls H _k ;

Determine the set J of the feature information corresponding to the N' calls;

Use a supervised learning algorithm according to the category corresponding to each of the N' calls, the set of feature sets corresponding to each of the N' calls, and the set of feature information J Determine the boundary of any two preset movement state categories in the M-dimensional space, and obtain the range of the distribution of any of the preset movement state categories in the M-dimensional space according to the boundary, and the M is the number of features in the feature set corresponding to the call H _k ;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

In some possible implementations, a set Ji of the N' calls in the call corresponding to any feature information in the set J is determined, if the set Ji includes N" calls, the N" calls The set of types of mobile states corresponding to calls is Ji', and the number of types of mobile states corresponding to the set Ji' is N"', if the number of calls corresponding to a certain type of mobile state in the set Ji' is less than

则复制该移动状态的类型对应的呼叫对应的向量F。通过这种方式可以提高概率较小的类别的分类精度。

Then copy the vector F corresponding to the call corresponding to the type of the mobility state. In this way, the classification accuracy of the less probable categories can be improved.

According to the category of the mobility state corresponding to each of the N' calls, the vector F, and the vector corresponding to the preset feature set corresponding to each call, use a supervised learning algorithm to determine any two of the presets The category of the mobile state is in the limit of the M-dimensional space, and according to the limit, the range of the distribution of any of the preset mobile state categories in the M-dimensional space is obtained, and the M is the corresponding value of the call H _k the number of features in the preset feature set;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on its corresponding preset feature set. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

In some possible implementation manners, when the N calls do not include precise geographic location information, classifying the N calls according to preset mobility state categories includes:

According to the set of the preset feature sets corresponding to the N calls, obtain N vectors corresponding to the preset feature sets corresponding to the N calls respectively;

According to the N described vectors and the unsupervised learning algorithm, the N calls are divided into M sets, and the M is greater than the preset number of mobile state categories;

According to the expert rule, the M sets are classified according to the preset mobility state classes, and the class of any call is the same as the class of the set to which it belongs.

In a second aspect, an embodiment of the present application provides an apparatus for classifying calls, the apparatus comprising:

an acquisition unit, configured to acquire a data packet, the data packet includes N data sets corresponding to N calls respectively, and the data set corresponding to any call Hi in the N calls includes the communication terminal during the call Hi process All measurement reports MR sent to the base station in sequence in the N, wherein the N is an integer greater than 1, and the 1≤i≤N;

a first processing unit, configured to divide each call in the N calls with w preset time windows, respectively, and determine a feature set corresponding to each call after being divided by each preset time window, The feature set includes the number of cells corresponding to the split call, the number of handovers, and the value of at least one self-defined characteristic, the self-defined characteristic being related to the movement state of the call;

The classification unit is configured to classify the N calls according to the preset mobility state categories according to the corresponding feature set of each call after being segmented by each preset time window.

In each embodiment of the present application, the at least one custom feature includes at least one of the following custom features: standard deviation of received signal strength, handover entropy, outdoor cell ratio, and inter-site distance speed; wherein,

The standard deviation of the received signal strength is the standard deviation of the received signal strength value of the call Hj;

the handover entropy, used to represent the uncertainty of the calling Hj accessing the cell;

The outdoor cell ratio is the percentage of the number of the call Hj accessing the outdoor type cells to the total number of all access cells;

The inter-station distance speed is the longest distance between the mean value of all the position information obtained by the call Hj and all the positions of the call.

In some possible implementations, the handover entropy can be calculated according to the following formula:

Wherein, the entropy is the handover entropy corresponding to the call Hj, and the

where N represents the total number of cells accessed by the call Hj, i represents the ith cell accessed by the call Hj, #i represents the number of times the call Hj accesses the ith cell, and T represents the number of accesses to different cells. The total number of times, pi represents the probability of accessing the i-th cell during this period.

It can be seen that in the above technical solution, since a custom feature related to the mobility state of the call is set, when judging the mobility state, in addition to the number of cells corresponding to the call and the number of handovers, the custom feature can also be combined It is helpful to improve the accuracy when classifying calls.

In some possible implementations, depending on the nature of the access cell,

The received signal strength standard deviation in the preset feature set corresponding to the call Hj includes: the received signal strength standard deviation of the primary serving cell, the received signal strength standard deviation of neighboring cells, and the received signal strength standard deviation of all cells:

The handover entropy in the preset feature set corresponding to the call Hj includes: the entropy of the primary serving cell, the entropy of the adjacent cells, and the entropy of all cells.

In some possible implementations, when the data sets corresponding to N' calls in the N calls include precise geographic location information, the preset feature set corresponding to any one of the N' calls is further Including: average speed, wherein, the N'≥1.

In some possible implementations, the precise geographic location information includes location information obtained through AGPS or OTT positioning services.

In some possible implementations, when the data sets corresponding to N' calls in the N calls include precise geographic location information, the classification unit is specifically configured to:

Any call H _k in the N' calls determines the type of the call H _k according to its corresponding average speed and the average speed corresponding to the preset mobile state types respectively, wherein the 1≤K≤N ';

According to the category corresponding to each of the N' calls and the set of the feature sets corresponding to each of the N' calls, use a supervised learning algorithm to determine any two of the preset The limit of the category of the mobile state in the M-dimensional space, and according to the limit, the range of the distribution of any of the preset categories of the mobile state in the M-dimensional space is obtained, and the M is the number corresponding to the call H _k . Describe the number of features in the feature set;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

In some possible implementations, the classification unit is specifically used to:

When the N' calls include GIS information,

Obtain the location of the specified feature information in the GIS information, where the specified feature information is the feature information strongly related to the preset movement state;

Any call H _k among the N' calls determines the type of the call H _k according to its corresponding average speed and the speed corresponding to the preset mobile state type respectively, wherein the 1≤K≤N';

determining the feature information matched by any of the calls H _k ;

Determine the set J of the feature information corresponding to the N' calls;

A supervised learning algorithm is used according to the category corresponding to each of the N' calls, the set of feature sets corresponding to each of the N' calls, and the set of feature information J Determine the boundary of any two preset movement state categories in the M-dimensional space, and obtain the range of the distribution of any of the preset movement state categories in the M-dimensional space according to the boundary, and the M is the number of features in the feature set corresponding to the call H _k ;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

In some possible implementations, when the N calls do not include precise geographic location information, the classification unit is specifically configured to:

According to the set of the preset feature sets corresponding to the N calls, obtain N vectors corresponding to the preset feature sets corresponding to the N calls respectively;

According to the N described vectors and the unsupervised learning algorithm, the N calls are divided into M sets, and the M is greater than the preset number of mobile state categories;

According to the expert rule, the M sets are classified according to the preset mobility state classes, and the class of any call is the same as the class of the set to which it belongs.

In a third aspect, an embodiment of the present application provides a storage medium, the storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each of which is The program includes instructions, and the instructions include instructions that can be executed by a device with a processor of some or all of the steps of any of the methods for classifying calls provided in the embodiments of the present application.

In a fourth aspect, an embodiment of the present application provides an apparatus for classifying calls, which is characterized by comprising:

A processor and a storage component coupled to each other; wherein the processor is configured to perform the method of any one of claims 1 to 9.

Description of drawings

In order to more clearly describe the technical solutions in the embodiments of the present application or the background technology, the accompanying drawings required in the embodiments or the background technology of the present application will be described below.

FIG. 1 is a schematic diagram of a scenario applied by an embodiment of the present application;

2 is a schematic flowchart of a method for classifying calls provided by an embodiment of the present application;

Fig. 3 is a schematic diagram of the actual movement track of each communication terminal in Fig. 1 and the distance speed between corresponding stations;

FIG. 4 is a schematic structural diagram of an apparatus for classifying calls provided by another embodiment of the present application.

Detailed ways

The embodiments of the present application will be described below with reference to the accompanying drawings in the embodiments of the present application.

Referring to FIG. 1 , FIG. 1 is a schematic diagram of a scenario where an embodiment of the present application is applied. As shown in FIG. 1 , the apparatus 101 for classifying calls obtains data packets from multiple base stations (BTS1, BTS2, BTS3, BTS4, . The corresponding data set includes all measurement reports (Measurement Report, MR) that the communication terminal sends to the base station in turn during the call process. MR is the information that the communication terminal feeds back to the apparatus 101 for classifying the call, and the MR includes the embodiments of the present application. Information such as serving cells, neighboring cells, and signal strengths corresponding to these cells received by the terminal to be used. The communication terminal A in FIG. 1 is located on a car and moves at high speed as the car moves. The communication terminal B is located indoors and is in a stationary state. The communication terminal C moves at a low speed along with the user's walking. It can be seen from FIG. 1 that at different times T1, T2 and T3, different communication terminals move at different distances, among which communication terminal A moves the farthest, communication terminal C is second, and communication terminal B does not move. In the prior art, when determining the mobile state of a communication terminal, it is usually determined according to the number of cells and the number of handovers corresponding to the communication terminal when calling. It should be noted that when the communication terminal moves rapidly in a small area, according to the prior art, due to the small number of cells and the number of handovers, the obtained moving state of the communication terminal may be slow or static, so Inaccurate when classifying.

In order to improve the accuracy of classifying the movement state of a communication terminal, a custom feature related to the movement state of a call is introduced in the embodiment of the present application, and the device 101 for classifying obtains data packets corresponding to multiple calls, and analyzes the movement state of the multiple calls. Status is classified.

Specifically, as shown in FIG. 2 , the method for classifying calls provided by an embodiment of the present application includes the following steps:

S201. Acquire a data packet, where the data packet includes N data sets corresponding to N calls respectively, and the data set corresponding to any call Hi in the N calls includes the communication terminal in the process of calling Hi. All measurement reports MR sent by the base station, wherein the N is an integer greater than 1, and the 1≤i≤N.

S202: Divide each of the N calls with w preset time windows, and determine a feature set corresponding to each call after being divided by each preset time window, where the feature set includes The number of cells corresponding to the split call, the number of handovers, and the value of at least one custom feature, where the custom feature is related to the mobility state of the call.

The number of cells is the total number of cells accessed within a period of time (for example, 30 seconds, or 1 minute, or 5 minutes, etc.).

Among them, the number of handovers is the total number of handovers that occur in one end of time (such as 30 seconds, or 1 minute, or 5 minutes, etc.).

Wherein, the at least one custom feature includes at least one of the following custom features: standard deviation of received signal strength, handover entropy, outdoor cell ratio, and inter-site distance speed; wherein,

The standard deviation of the received signal strength is the standard deviation of the received signal strength value of the call Hj. Specifically, it may be the standard deviation of the received signal strength values obtained by multiple measurements within a period of time (such as 30 seconds, or 1 minute, or 5 minutes, etc.). It can be understood that, according to the nature of the receiving end of the received signal, the standard deviation of the received signal strength can be divided into the standard deviation of the received signal strength of the primary serving cell, the standard deviation of the received signal strength of the neighboring cells, and the standard deviation of the received signal strength of all cells. In general, the faster the communication terminal moves, the greater the value of the standard deviation of the received signal strength.

The handover entropy is used to represent the uncertainty of the calling Hj accessing the cell. Specifically, it may be the uncertainty of the communication terminal accessing different cells within a period of time (for example, 30 seconds, or 1 minute, or 5 minutes, etc.), and the value is between 0 and 1. For example, if the communication terminal is in a stationary state, it accesses only one cell during this period, and its handover entropy is 0. It can be understood that, the faster the moving speed of the communication terminal, the greater the number of cells it accesses and the number of cell handovers within a period of time, and the closer its handover entropy is to 1. For example, if the handover entropy is 0.2, it can be inferred that the communication terminal is in a low-speed moving state. If the handover entropy is 0.9, it can be inferred that the communication terminal is in a high-speed moving state.

In some possible implementations, the switching entropy can be calculated according to the following formula:

Wherein, the entropy is the handover entropy corresponding to the call Hj, and the

where N represents the total number of cells accessed by the call Hj, i represents the ith cell accessed by the call Hj, #i represents the number of times the call Hj accesses the ith cell, and T represents the number of accesses to different cells. The total number of times, pi represents the probability of accessing the i-th cell during this period.

According to different properties of the access cells, the handover entropy includes: the entropy of the primary serving cell, the entropy of the adjacent cells, and the entropy of all cells.

The outdoor cell proportion is the percentage of the number of the calling Hj accessing the outdoor type cells to the total number of all access cells. It can be understood that if the proportion of outdoor cells is less than 50%, it can be considered that the communication terminal is located indoors and is in a static state.

The inter-station distance speed is the longest distance between the mean value of all the position information obtained by the call Hj and all the positions of the call. Specifically, it may be the longest distance between the mean value of all the location information (including MR positioning, AGPS positioning, and OTT positioning) obtained by the user within a period of time and all the positions. The larger the inter-station distance speed is, the larger the movement speed of the communication terminal is considered to be. For the situation that the communication terminal starts from a position and returns to the position within a certain period of time, the distance and speed between stations can better represent the movement speed of the user than the simple speed calculation model, that is, the ratio of the distance between the starting and ending points and the time difference. Referring to FIG. 3 , the actual movement trajectory of each communication terminal is shown as a dotted line in FIG. 3 , the starting position of the solid line segment with arrows in the figure indicates the mean value of the position information, and the position indicated by the arrows of the line segments with arrows is in the specified window. The point that is farthest from the mean value in time, the distance of the line segment with the arrow is the example speed between stations. It can be seen from Figure 3 that the inter-station distance velocity of the left communication terminal is the largest, the inter-station distance velocity of the middle communication terminal is second, and the inter-station distance velocity of the right communication terminal is the smallest.

S203: Classify the N calls according to preset mobility state categories according to the corresponding feature set of each call after being segmented by each preset time window.

It should be noted that the feature sets corresponding to the multiple calls obtained by dividing any preset time window are first calculated, and then the average value of the multiple feature sets is determined as the feature set corresponding to the preset time window. For example, a call H1 is divided into 6 small calls by a preset time window of 30 seconds, then the feature sets corresponding to the 6 small calls are calculated respectively, and then the average value of each feature in the obtained 6 feature sets is calculated. , and take the mean value as the feature set of the call H1.

It can be seen that in the above technical solution, since a custom feature related to the mobility state of the call is set, when judging the mobility state, in addition to the number of cells corresponding to the call and the number of handovers, the custom feature can also be combined It is helpful to improve the accuracy when classifying calls.

In some possible implementations, when the data sets corresponding to N' calls in the N calls include precise geographic location information, the N calls are classified according to preset mobility state categories ,include:

Any call Hk among the N' calls determines the type of the call Hk according to the average speed corresponding to the call Hk according to the average speed corresponding to the preset mobile state types respectively, where 1≤K≤N'. For example, if N is 1000 and N' is 120, the data set corresponding to 120 calls in the 1000 calls includes precise geographic location information. Wherein, the precise geographic location information includes obtaining through an assisted global positioning system (ssisted global positioning system, AGPS) or through an application service (Over The Top, OTT) of the Internet. According to the geographic location information of the 120 calls with precise geographic location information, the average speed of each call in the 120 calls can be obtained, and the mobility state of these calls can be determined according to the evaluation speed corresponding to the preset mobility state category. category.

According to the category corresponding to each of the N' calls and the set of the feature sets corresponding to each of the N' calls, use a supervised learning algorithm to determine any two of the preset The limit of the category of the mobile state in the M-dimensional space, and according to the limit, the range of the distribution of any of the preset categories of the mobile state in the M-dimensional space is obtained, and the M is the number corresponding to the call H _k . Describe the number of features in the feature set;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

In some possible implementations, when the N' calls include GIS information,

Obtain the location of the feature information specified in the GIS information, and the specified feature information is the feature information strongly related to the preset movement state; Roads, intersections, highways or railways, etc.

Any call H _k among the N' calls determines the type of the call H _k according to its corresponding average speed and the speed corresponding to the preset mobile state type respectively, wherein the 1≤K≤N';

determining the feature information matched by any of the calls H _k ;

Determine the set J of the feature information corresponding to the N' calls;

Use a supervised learning algorithm according to the category corresponding to each of the N' calls, the set of feature sets corresponding to each of the N' calls, and the set of feature information J Determine the boundary of any two preset movement state categories in the M-dimensional space, and obtain the range of the distribution of any of the preset movement state categories in the M-dimensional space according to the boundary, and the M is the number of features in the feature set corresponding to the call H _k . For example, in some possible implementations,

In some possible implementations, a set Ji of the N' calls in the call corresponding to any feature information in the set J is determined, if the set Ji includes N" calls, the N" calls The set of types of mobile states corresponding to calls is Ji', and the number of types of mobile states corresponding to the set Ji' is N"', if the number of calls corresponding to a certain type of mobile state in the set Ji' is less than

则复制该移动状态的类型对应的呼叫对应的向量F。通过这种方式可以提高概率较小的类别的分类精度。

Then copy the vector F corresponding to the call corresponding to the type of the mobility state. In this way, the classification accuracy of the less probable categories can be improved.

According to the category of the mobility state corresponding to each of the N' calls, the vector F, and the vector corresponding to the preset feature set corresponding to each call, use a supervised learning algorithm to determine any two of the presets The category of the mobile state is in the limit of the M-dimensional space, and according to the limit, the range of the distribution of any of the preset mobile state categories in the M-dimensional space is obtained, and the M is the corresponding value of the call H _k the number of features in the preset feature set;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on its corresponding preset feature set. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

For example, if the feature information corresponding to the GIS is a park, there are 150 calls including the precise geographic location passing through the park, of which 92 calls correspond to a slow moving state, and 51 calls correspond to a stationary state. The mobile state corresponding to the 7 calls is high-speed movement. Since the number of calls corresponding to high-speed motion is less than 150/3=50, the vector corresponding to the call of high-speed motion can be copied. After copying, the vector corresponding to high-speed motion can reach 50 or more. The vector corresponding to the preset feature set, as well as the 43 copies obtained (taking the number of calls corresponding to high-speed motion after copying as an example) correspond to 50 calls with high-speed motion, use the supervised learning algorithm to determine any two The boundary of the preset movement state category in the M-dimensional space, and the range of the distribution of any of the preset movement state categories in the M-dimensional space is obtained according to the boundary.

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

In some possible implementations, when the N calls do not include precise geographic location information, according to the set of the preset feature sets corresponding to the N calls, the corresponding N calls are obtained respectively. N vectors corresponding to the preset feature set;

According to the N vectors and an unsupervised learning algorithm, the N calls are divided into M sets, where M is greater than the number of preset mobility state categories; for example, if the preset mobility state includes 4 types: stationary, low-speed motion, medium-speed motion, and high-speed motion. If N is 1000, in some possible implementations of the present invention, 1000 calls can be divided into 20 sets according to an unsupervised learning algorithm, and then determined according to expert algorithms such as factor analysis, iterative, and principal component analysis. The state of movement for each of the 20 sets. For example, the moving state of the set whose switching entropy value is 0 is determined to be stationary. The mobility state of the call in the set where the handover entropy is 0.7 or more and the number of handovers is greater than 5 is determined to be high-speed mobility or the like. The class of any call is the same as the class of the set to which it belongs.

Referring to FIG. 4, an apparatus 400 for classifying calls provided in an embodiment of the present application. Specifically, the apparatus 400 for classifying calls shown in FIG. 4 may include: an obtaining unit 401, a first processing unit 402, and classification unit 403.

The obtaining unit 401 is configured to execute the method of step S201 in FIG. 2 of the method embodiment of the present invention. For the implementation of the obtaining unit 401, reference may be made to the description corresponding to step S201 in FIG. 2 of the method embodiment of the present invention, and details are not repeated here.

The first processing unit 402 is configured to execute the method of step S202 in FIG. 2 of the method embodiment of the present invention. For the implementation of the first processing unit 402, reference may be made to the description corresponding to step S202 in FIG. 2 of the method embodiment of the present invention, which will not be repeated here. .

The classification unit 403 is configured to execute the method of step S203 in FIG. 2 in the method embodiment of the present invention. For the implementation of the classification unit 403, reference may be made to the description corresponding to step S203 in FIG. 2 in the method embodiment of the present invention, which will not be repeated here.

Optionally, in some possible implementations of the present invention, the at least one custom feature includes at least one of the following custom features: standard deviation of received signal strength, handover entropy, outdoor cell ratio, and inter-station distance. speed; where,

The standard deviation of the received signal strength is the standard deviation of the received signal strength value of the call Hj;

the handover entropy, used to represent the uncertainty of the calling Hj accessing the cell;

The outdoor cell ratio is the percentage of the number of the call Hj accessing the outdoor type cells to the total number of all access cells;

The inter-station distance speed is the longest distance between the mean value of all the position information obtained by the call Hj and all the positions of the call.

Optionally, in some possible embodiments of the present invention, the handover entropy is calculated according to the following formula:

Wherein, the entropy is the handover entropy corresponding to the call Hj, and the

where N represents the total number of cells accessed by the call Hj, i represents the ith cell accessed by the call Hj, #i represents the number of times the call Hj accesses the ith cell, and T represents the number of accesses to different cells. The total number of times, pi represents the probability of accessing the i-th cell during this period.

Optionally, in some possible implementations of the present invention, according to different properties of the access cell,

The received signal strength standard deviation in the preset feature set corresponding to the call Hj includes: the received signal strength standard deviation of the primary serving cell, the received signal strength standard deviation of neighboring cells, and the received signal strength standard deviation of all cells:

The handover entropy in the preset feature set corresponding to the call Hj includes: the entropy of the primary serving cell, the entropy of the adjacent cells, and the entropy of all cells.

Optionally, in some possible implementations of the present invention, when the data sets corresponding to N' calls in the N calls include precise geographic location information, any call in the N' calls corresponds to The preset feature set also includes: average speed, where the N'≥1.

The precise geographic location information includes location information obtained through AGPS or OTT positioning services.

Optionally, in some possible implementations of the present invention, when the data sets corresponding to N' calls in the N calls include precise geographic location information, the classification unit is specifically configured to:

Any call H _k in the N' calls determines the type of the call H _k according to its corresponding average speed and the average speed corresponding to the preset mobile state types respectively, wherein the 1≤K≤N ';

According to the category corresponding to each of the N' calls and the set of the feature sets corresponding to each of the N' calls, use a supervised learning algorithm to determine any two of the preset The limit of the category of the mobile state in the M-dimensional space, and according to the limit, the range of the distribution of any of the preset categories of the mobile state in the M-dimensional space is obtained, and the M is the number corresponding to the call H _k . Describe the number of features in the feature set;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

Optionally, in some possible implementations of the present invention, the classification unit is specifically used to:

When the N' calls include GIS information,

Obtain the location of the specified feature information in the GIS information, where the specified feature information is the feature information strongly related to the preset movement state;

Any call H _k among the N' calls determines the type of the call H _k according to its corresponding average speed and the speed corresponding to the preset mobile state type respectively, wherein the 1≤K≤N';

determining the feature information matched by any of the calls H _k ;

Determine the set J of the feature information corresponding to the N' calls;

A supervised learning algorithm is used according to the category corresponding to each of the N' calls, the set of feature sets corresponding to each of the N' calls, and the set of feature information J Determine the boundary of any two preset movement state categories in the M-dimensional space, and obtain the range of the distribution of any of the preset movement state categories in the M-dimensional space according to the boundary, and the M is the number of features in the feature set corresponding to the call H _k ;

For the (N-N') calls that do not include precise geographic location information in the data set corresponding to the N calls, any of the (N-N') calls is based on the set of corresponding feature sets. Obtain the mapping position of the call in the M-dimensional space, and determine any of the (N- N') mobile states corresponding to calls.

Optionally, in some possible implementations of the present invention, when the N calls do not include precise geographic location information, the classification unit is specifically configured to:

According to the set of the preset feature sets corresponding to the N calls, obtain N vectors corresponding to the preset feature sets corresponding to the N calls respectively;

According to the N described vectors and the unsupervised learning algorithm, the N calls are divided into M sets, and the M is greater than the preset number of mobile state categories;

According to the expert rule, the M sets are classified according to the preset mobility state classes, and the class of any call is the same as the class of the set to which it belongs.

An embodiment of the present application further provides a computer storage medium, wherein the computer storage medium may store a program, and when the program is executed, the program includes part or all of the steps of any of the methods for classifying calls described in the above method embodiments .

An embodiment of the present application further provides an apparatus for classifying calls, including: a processor and a storage component coupled to each other; wherein the processor is configured to execute any one of the methods for classifying calls described in the foregoing method embodiments .

The steps in the method of the embodiment of the present application may be adjusted, combined and deleted in sequence according to actual needs.

Units in the apparatus of the embodiment of the present application may be combined, divided, and deleted according to actual needs.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented. The process can be completed by instructing the relevant hardware by a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed , which may include the processes of the foregoing method embodiments. The aforementioned storage medium includes: a read-only memory (Read-Only Memory, ROM) or a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk and other media that can store program codes.

Claims (20)

1. A method of classifying a call, the method comprising:

acquiring a data packet, wherein the data packet comprises N data sets respectively corresponding to N calls, and the data set corresponding to any call Hi in the N calls comprises all measurement reports MR which are received by a base station and are sequentially sent to the base station by a communication terminal in the process of calling Hi, wherein N is an integer greater than 1, and i is greater than or equal to 1 and less than or equal to N;

segmenting each call in the N calls by w preset time windows respectively, and determining a feature set corresponding to each call after each preset time window is segmented, wherein the feature set comprises the number of cells, the switching number and the value of at least one custom feature corresponding to the segmented call, and the custom feature is related to the moving state of the call;

and classifying the N calls according to the classes of the preset moving state according to the feature set corresponding to each call after being segmented by each preset time window.

2. The method of claim 1,

the at least one custom feature comprises at least one of the following custom features: receiving signal strength standard deviation, switching entropy, outdoor cell ratio and inter-station distance speed; wherein,

the received signal strength standard deviation is the standard deviation of the received signal strength value of the call Hi;

the switching entropy is used for representing the uncertainty of the Hi access cell of the call;

the outdoor cell occupation ratio is the percentage of the number of the call Hi access outdoor type cells in the total number of all the access cells;

and the inter-station distance speed is the farthest distance between the average value of all the position information obtained by the call Hi and all the positions of the call.

3. The method of claim 2,

the switching entropy is calculated according to the following formula:

wherein the entry is a handover entropy corresponding to the call Hi, and the entry is a handover entropy corresponding to the call Hi

Where N denotes the total number of cells accessed by the call Hi, i denotes the ith cell accessed by the call Hi, # i denotes the number of times the call Hi accesses the ith cell, T denotes the total number of times different cells are accessed, and pi denotes the probability of accessing the ith cell during this time.

4. The method of claim 3,

depending on the nature of the access cell,

the received signal strength standard deviation in the preset feature set corresponding to the call Hi includes: the received signal strength standard deviation of the main serving cell, the received signal strength standard deviations of the neighboring cells, and the received signal strength standard deviations of all cells:

the switching entropy in the preset feature set corresponding to the call Hi includes: entropy of the main serving cell, entropy of the neighboring cells, and entropy of all cells.

5. The method of claim 4,

when the data set corresponding to N 'calls in the N calls includes accurate geographical location information, the preset feature set corresponding to any one of the N' calls further includes: average speed, wherein N' is ≧ 1.

6. The method of claim 5,

the precise geographical location information includes location information obtained through AGPS or OTT location services.

7. The method of claim 6,

when the data sets corresponding to N' calls in the N calls include accurate geographical location information, classifying the N calls according to a preset category of a mobile state includes:

any one of the N' calls H_kDetermining the call H according to the corresponding average speeds and the corresponding average speeds of the types of the preset moving states_kWherein 1. ltoreq. K.ltoreq.N';

determining the boundary of any two preset mobile state categories in an M-dimensional space by using a supervised learning algorithm according to the category corresponding to each call in the N 'calls and the feature set corresponding to each call in the N' calls, and obtaining the region range of any one preset mobile state category in the M-dimensional space according to the boundary, wherein M is the call H_kThe number of the corresponding features in the feature set;

for the (N-N ') calls in the N calls, calling the (N-N') calls which do not include the accurate geographic position information in the corresponding data set, obtaining the mapping position of the call in the M-dimensional space according to the (N-N ') calls and determining the moving state corresponding to the (N-N') calls according to the mapping position and the regional range of the M-dimensional space distribution of any preset moving state.

8. The method of claim 6, wherein classifying the N calls according to a predetermined category of mobility state comprises:

when GIS information is included in the N' calls,

acquiring the position of the specified ground feature information in the GIS information, wherein the specified ground feature information is the ground feature information which is strongly related to the preset moving state;

any one of the N' calls H_kDetermining the call H according to the corresponding average speed and the speed corresponding to the preset type of the moving state_kWherein 1. ltoreq. K.ltoreq.N';

determining said any call H_kMatching ground feature information;

determining a set J of feature information corresponding to the N' calls;

determining the boundary of any two preset moving state categories in an M-dimensional space according to the category corresponding to each call in the N 'calls, the feature set corresponding to each call in the N' calls and the feature information set J by using a supervised learning algorithm, obtaining the region range of any one preset moving state category in the M-dimensional space according to the boundary, wherein M is the call H_kThe number of the corresponding features in the feature set;

for the (N-N ') calls in the N calls, calling the (N-N') calls which do not include the accurate geographic position information in the corresponding data set, obtaining the mapping position of the call in the M-dimensional space according to the (N-N ') calls and determining the moving state corresponding to the (N-N') calls according to the mapping position and the regional range of the M-dimensional space distribution of any preset moving state.

9. The method of claim 2,

when the accurate geographical location information is not included in the N calls, classifying the N calls according to a preset category of a moving state includes:

obtaining N vectors corresponding to the preset feature set and respectively corresponding to the N calls according to the set of the preset feature set corresponding to the N calls;

dividing the N calls into M sets according to the N vectors and an unsupervised learning algorithm, wherein M is larger than the number of the categories of the preset moving state;

and classifying the M sets according to the preset moving state category according to expert rules, wherein the category of any call is the same as that of the set to which the call belongs.

10. An apparatus for classifying a call, the apparatus comprising:

an obtaining unit, configured to obtain a data packet, where the data packet includes N data sets corresponding to N calls, and a data set corresponding to any call Hi in the N calls includes all measurement reports MR received by a base station and sent by a communication terminal to the base station in sequence in the process of calling Hi, where N is an integer greater than 1, and i is greater than or equal to 1 and is less than or equal to N;

the first processing unit is used for segmenting each call in the N calls by w preset time windows respectively, and determining a feature set corresponding to each call after each preset time window is segmented, wherein the feature set comprises the number of cells, the switching number and at least one value of a user-defined feature corresponding to the segmented call, and the user-defined feature is related to the moving state of the call;

and the classification unit is used for classifying the N calls according to the classes of the preset moving states according to the feature set corresponding to each call after being segmented by each preset time window.

11. The apparatus of claim 10,

the at least one custom feature comprises at least one of the following custom features: receiving signal strength standard deviation, switching entropy, outdoor cell ratio and inter-station distance speed; wherein,

the received signal strength standard deviation is the standard deviation of the received signal strength value of the call Hi;

the switching entropy is used for representing the uncertainty of the Hi access cell of the call;

the outdoor cell occupation ratio is the percentage of the number of the call Hi access outdoor type cells in the total number of all the access cells;

and the inter-station distance speed is the farthest distance between the average value of all the position information obtained by the call Hi and all the positions of the call.

12. The apparatus of claim 11,

the switching entropy is calculated according to the following formula:

wherein the entry is a handover entropy corresponding to the call Hi, and the entry is a handover entropy corresponding to the call Hi

Where N denotes the total number of cells accessed by the call Hi, i denotes the ith cell accessed by the call Hi, # i denotes the number of times the call Hi accesses the ith cell, T denotes the total number of times different cells are accessed, and pi denotes the probability of accessing the ith cell during this time.

13. The apparatus of claim 12,

depending on the nature of the access cell,

the received signal strength standard deviation in the preset feature set corresponding to the call Hi includes: the received signal strength standard deviation of the main serving cell, the received signal strength standard deviations of the neighboring cells, and the received signal strength standard deviations of all cells:

the switching entropy in the preset feature set corresponding to the call Hi includes: entropy of the main serving cell, entropy of the neighboring cells, and entropy of all cells.

14. The apparatus of claim 13,

when the data set corresponding to N 'calls in the N calls includes accurate geographical location information, the preset feature set corresponding to any one of the N' calls further includes: average speed, wherein N' is ≧ 1.

15. The apparatus of claim 14,

the precise geographical location information includes location information obtained through AGPS or OTT location services.

16. The apparatus of claim 15,

when the data sets corresponding to N' of the N calls include accurate geographical location information, the classifying unit is specifically configured to,

any one of the N' calls H_kDetermining the call H according to the corresponding average speeds and the corresponding average speeds of the types of the preset moving states_kWherein 1. ltoreq. K.ltoreq.N';

determining the boundary of any two classes of the preset moving state in an M-dimensional space by using a supervised learning algorithm according to the class corresponding to each of the N 'calls and the set of the feature set corresponding to each of the N' calls, and obtaining any one of the preset moving states according to the boundaryThe category of the state is in the region range of the M-dimensional space distribution, wherein M is the call H_kThe number of the corresponding features in the feature set;

for the (N-N ') calls in the N calls, calling the (N-N') calls which do not include the accurate geographic position information in the corresponding data set, obtaining the mapping position of the call in the M-dimensional space according to the (N-N ') calls and determining the moving state corresponding to the (N-N') calls according to the mapping position and the regional range of the M-dimensional space distribution of any preset moving state.

17. The apparatus according to claim 15, wherein the classification unit is specifically configured to,

when GIS information is included in the N' calls,

acquiring the position of the specified ground feature information in the GIS information, wherein the specified ground feature information is the ground feature information which is strongly related to the preset moving state;

any one of the N' calls H_kDetermining the call H according to the corresponding average speed and the speed corresponding to the preset type of the moving state_kWherein 1. ltoreq. K.ltoreq.N';

determining said any call H_kMatching ground feature information;

determining a set J of feature information corresponding to the N' calls;

determining the boundary of any two preset moving state categories in an M-dimensional space according to the category corresponding to each call in the N 'calls, the feature set corresponding to each call in the N' calls and the feature information set J by using a supervised learning algorithm, obtaining the region range of any one preset moving state category in the M-dimensional space according to the boundary, wherein M is the call H_kThe number of the corresponding features in the feature set;

for the (N-N ') calls in the N calls, calling the (N-N') calls which do not include the accurate geographic position information in the corresponding data set, obtaining the mapping position of the call in the M-dimensional space according to the (N-N ') calls and determining the moving state corresponding to the (N-N') calls according to the mapping position and the regional range of the M-dimensional space distribution of any preset moving state.

18. The apparatus of claim 11,

when the precise geographical location information is not included in the N calls, the classification unit is specifically configured to,

obtaining N vectors corresponding to the preset feature set and respectively corresponding to the N calls according to the set of the preset feature set corresponding to the N calls;

dividing the N calls into M sets according to the N vectors and an unsupervised learning algorithm, wherein M is larger than the number of the categories of the preset moving state;

and classifying the M sets according to the preset moving state category according to expert rules, wherein the category of any call is the same as that of the set to which the call belongs.

19. A storage medium, characterized in that it is a non-volatile computer-readable storage medium storing at least one program, each of said programs comprising instructions which, when executed by an apparatus having a processor, cause the apparatus to carry out the method of classifying a call according to any one of claims 1-9.

20. An apparatus for classifying a call, comprising:

a processor and a memory component coupled to each other; wherein the processor is configured to perform the method of any one of claims 1 to 9.

Images