Training method and device for identifying a model for identifying a permanent location of a user on a different network

Abstract

This application discloses a training method and apparatus for a model for identifying the permanent location of users on a different network, aiming to address the problem of inaccurate identification of the permanent locations of users on a different network. The solution provided in this application includes: obtaining communication behavior information of multiple users on the same network; generating a complete graph of the social circles corresponding to each user on the same network based on the communication behavior information of the multiple users on the same network; generating multiple training samples and corresponding training labels based on the complete graph of the social circles corresponding to the multiple users on the same network and the corresponding communication behavior information; and training a pre-built time-series-based recurrent neural network model based on the multiple training samples and corresponding training labels.

Description

Technical Field

[0001] The present application relates to the field of communications, and in particular to a method and device for training a permanent location identification model for users on a different network. Background Art

[0002] In the field of communications, different operators can often only obtain the communication behavior information of users on their own networks, but cannot obtain the communication behavior information of users on other networks.

[0003] In some application scenarios, operators can identify the location trajectory of communication users on their own network through call records, and thus predict the actual residential cells of users on their own network. However, users on other networks cannot obtain communication data under the base station, and cannot fundamentally obtain the mapping relationship between the cell and the permanent residence. Therefore, the permanent residence information of customers on other networks is lost, resulting in the inability to effectively poach customers on other networks.

[0004] How to improve the accuracy of identifying the permanent location of users communicating on different networks is the technical problem to be solved by this application. Summary of the Invention

[0005] The purpose of the embodiments of the present application is to provide a method and device for training a model for identifying the permanent location of users on a different network, so as to solve the problem of inaccurate identification of the permanent locations of users communicating with different networks.

[0006] In a first aspect, a method for training a model for identifying a permanent location of a user on a different network is provided, comprising:

[0007] Acquire communication behavior information of multiple local network communication users, wherein the communication behavior information includes geographical location information and time information of the communication behaviors performed by the local network communication users and the communication users of the other networks;

[0008] Generating a complete graph of social circles corresponding to the communication users of the local network to which each communication behavior information belongs based on the communication behavior information of the multiple communication users of the local network, wherein the complete graph of social circles includes vertices representing the communication users and edges connecting different vertices, wherein the edges represent communication behaviors performed between the communication users corresponding to the connected vertices;

[0009] Generate multiple training samples and corresponding training labels based on the complete graphs of social circles corresponding to the multiple local network communication users and the corresponding communication behavior information, wherein any of the training samples includes a time-series communication relationship feature matrix between communication users belonging to the same complete graph of social circles, and the training labels include the geographical locations of the permanent residences where the multiple local network communication users in the complete graph of social circles to which the alien network communication users belong perform communication behaviors;

[0010] A pre-built time-series-based recurrent neural network model is trained based on the multiple training samples and corresponding training labels.

[0011] In a second aspect, a training device for a permanent location identification model for users on a different network is provided, comprising:

[0012] An acquisition module is configured to acquire communication behavior information of multiple local network communication users, wherein the communication behavior information includes geographical location information and time information of the communication behaviors performed by the local network communication users and the communication users of the other networks;

[0013] A first generating module generates, based on the communication behavior information of the plurality of local network communication users, a complete graph of the social circles corresponding to the local network communication users to which each communication behavior information belongs, wherein the complete graph of the social circles includes vertices representing the communication users and edges connecting different vertices, wherein the edges represent communication behaviors performed between the communication users corresponding to the connected vertices;

[0014] A second generation module generates a plurality of training samples and corresponding training labels based on the complete graphs of social circles corresponding to the plurality of local network communication users and the corresponding communication behavior information, wherein any of the training samples includes a time-series communication relationship feature matrix between communication users belonging to the same complete graph of social circles, and the training labels include the geographic locations of the permanent locations where the plurality of local network communication users in the complete graph of social circles to which the alien network communication users belong perform communication behaviors;

[0015] A training module trains a pre-built time-series-based recurrent neural network model based on the multiple training samples and corresponding training labels.

[0016] In a third aspect, an electronic device is provided, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the method of the first aspect when executed by the processor.

[0017] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method of the first aspect are implemented.

[0018] In an embodiment of the present application, communication behavior information of multiple local network communication users is first obtained. The communication behavior information includes geographic location information and time information of the communication behaviors performed by the local network communication users and the communication behaviors performed by the users of the other network. Then, a complete social circle graph corresponding to the local network communication users to which each communication behavior information belongs is generated based on the communication behavior information of the multiple local network communication users. The complete social circle graph includes vertices representing the communication users and edges connecting different vertices, where the edges represent the communication behaviors performed between the communication users corresponding to the connected vertices. Subsequently, multiple training samples and corresponding training labels are generated based on the complete social circle graphs and the corresponding communication behavior information of the multiple local network communication users. Each training sample includes a time-series communication relationship feature matrix between the communication users belonging to the same complete social circle graph. The training labels include the geographic locations of the permanent locations where multiple local network communication users in the complete social circle graph to which the other network communication users belong perform communication behaviors. Finally, a pre-built time-series-based recurrent neural network model is trained based on the multiple training samples and corresponding training labels. The complete social circle graph generated based on the communication behavior information of the local network communication users can represent the communication relationships between the multiple communication users. Because multiple communication users with similar behaviors often share the same geographic location—for example, colleagues share the same office, or family members share the same residence—this solution identifies the permanent locations of inter-network communication users with similar communication behaviors based on their communication behaviors and the geographic location of the users on the same network. This allows accurate identification of the permanent locations of inter-network communication users without obtaining their communication behavior information. Furthermore, by generating training samples based on time series and training using a time-series-based model, this solution can effectively identify the permanent locations of inter-network users whose geographic locations change over time, making it widely applicable to a variety of practical application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 This is one of the flow charts of a method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0021] Figure 2a This is the second flow chart of a method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0022] Figure 2b This is one of the schematic diagrams of a complete social circle diagram of a method for training a permanent location identification model for users on a different network in one embodiment of the present application;

[0023] Figure 3a This is the third flow chart of a method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0024] Figure 3b This is the second schematic diagram of a complete social circle diagram of a method for training a permanent location identification model for users on a different network in accordance with an embodiment of the present application;

[0025] Figure 4 This is the fourth flow chart of a method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0026] Figure 5 This is the fifth flow chart of the method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0027] Figure 6 This is the sixth flow chart of the method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0028] Figure 7 This is the seventh flow chart of a method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0029] Figure 8 This is the eighth flow chart of a method for training a permanent location identification model for users on a different network according to an embodiment of the present application.

[0030] Figure 9 It is a structural diagram of an apparatus for executing a training method for a permanent location identification model of a user on a different network according to an embodiment of the present application.

[0031] Figure 10 It is a structural diagram of a training device for a permanent location identification model for users on a different network according to an embodiment of the present application. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The figure numbers in this application are only used to distinguish the various steps in the scheme and are not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.

[0033] In the communications sector, operators can directly obtain the communication behavior information of users on their own network, such as their historical communication behaviors and their locations at the time of communication. For users on other networks, communication behavior information can be obtained through manual visits, questionnaires, and other methods. However, this manually collected information contains subjective factors and may be inaccurate.

[0034] In addition, communication users are often mobile, and their locations are often different on different dates and at different times. The constant changes in the locations of communication users also make it difficult to determine the permanent locations of communication users on different networks.

[0035] In order to solve the problems existing in the prior art, the present invention provides a method for training a model for identifying the permanent location of users in a different network. Figure 1 Shown, including:

[0036] S11: Acquire communication behavior information of multiple communication users of the local network, wherein the communication behavior information includes geographical location information and time information of communication behaviors performed by communication users of the local network and communication users of different networks.

[0037] Communication activities may include voice calls, video calls, text messaging, multimedia messaging, or other forms of communication between users. Communication activity information may specifically include a user's call logs, text messaging logs, and other information recorded over a historical period, representing the user's communication activities during that period. Preprocessing may be performed on the communication records of users on this network over that historical period to obtain communication activity information. Preprocessing may include optimizing communication records, converting formats, and removing or repairing defective records, thereby obtaining valid communication activity information for users on this network.

[0038] The geographic location information in the communication behavior information can specifically be the communication permanent location data of the communication user of the network, and can specifically include information about the communication base station where the communication user of the network resides. Optionally, more types of geographic location information of the communication user of the network can be obtained, which can be used to improve the accuracy of data processing in subsequent steps. For example, the geographic location information of the communication behavior performed by the communication user of the network and other communication users of the network can be obtained, or the permanent location of the communication user of the network over a period of time, such as the location information of the top three permanent locations of the network within the past six months.

[0039] The time information in communication behavior information can include the time when the user on the network starts and ends the communication behavior. For example, the time information can specifically include the time when the user on the network receives or sends a text message. Alternatively, the time information can specifically include the time when the user on the network receives or makes a phone call. It can also include information such as the time when the call is hung up and the duration of the call.

[0040] S12: Generate a complete graph of the social circles corresponding to the communication users of the network to which each communication behavior information belongs based on the communication behavior information of the multiple communication users of the network, wherein the complete graph of the social circles includes vertices representing the communication users and edges connecting different vertices, wherein the edges represent the communication behaviors performed between the communication users corresponding to the connected vertices.

[0041] The complete graph of the communication circle in this scheme is used to express the corresponding communication circle of the communication users of this network. The complete graph of the communication circle contains multiple vertices and edges connecting different vertices. The vertex represents the communication user who has directly or indirectly performed communication behavior with the communication user of this network, and the edge represents that there is a communication relationship between the communication users corresponding to the two connected vertices.

[0042] In practical applications, the rules for generating a complete graph of social circles can be pre-set. For example, for vertices in the complete graph of social circles, users with whom the network's users directly engage in communication activities can be generated as vertices in the complete graph of social circles. Alternatively, users with whom the network's users directly and indirectly engage in communication activities can be generated as vertices in the complete graph of social circles. For edges in the complete graph of social circles, edges can be generated between vertices corresponding to users with one-way or two-way conversations, or only between vertices corresponding to users with two-way conversations. Furthermore, the communication form for generating edges can be set, for example, edges can be generated between vertices corresponding to users who have both engaged in conversations and sent and received text messages. It should be understood that the rules for generating vertices and edges can be pre-set based on actual needs and are not limited in this application.

[0043] The complete social circle graph generated in this step can represent the communication relationships between communication users. Communication users who have communicated with each other may know each other, for example, they may be colleagues, friends, or family members. The complete social circle graph generated in this solution can facilitate the identification and analysis of interpersonal relationships between different communication users.

[0044] In this example, the generated complete graph of social circles can be an undirected graph, that is, the edges connecting the vertices in the graph are used to represent the relationship between the corresponding communication users of the vertices, and there is no arrow pointing. For example, the rules for generating edges are as follows:

[0045] 1) If numbers A and B have a call for more than a certain period of time within a certain period of time, then numbers A and B are considered to have an edge in the graph. Because this is an undirected graph, the communication between numbers A and B can only be considered as an edge between A and B, not a directed edge from A to B or B to A in the graph.

[0046] 2) If number A and number B have sent text messages to each other within a period of time, then there is an edge between number A and number B in the graph. The edges in the graph are also undirected.

[0047] It should be noted that in the embodiments of this application, if a communication relationship exists between two vertices, an edge is used to represent it; if no relationship exists, no edge is connected. If at least one of the above two rules is met, an edge is generated between the two vertices to express the existence of a relationship between the connected vertices. Multiple edges are not required to connect the vertices.

[0048] S13: Generate multiple training samples and corresponding training labels based on the complete graph of the social circles corresponding to the multiple communication users of the local network and the corresponding communication behavior information, any of the training samples includes a time-series-based communication relationship feature matrix between the communication users belonging to the same complete graph of the social circles, and the training labels include the geographical locations of the permanent residences where the multiple communication users of the local network perform communication behaviors in the complete graph of the social circles to which the communication users of the different network belong.

[0049] The complete social circle graph can represent the communication relationships between communication users, and the communication behavior information can represent the specific communication behaviors performed by the communication users. In this step, the complete social circle graph and the communication behavior information are combined to generate training samples and training labels for model training. Training samples can be generated based on the communication relationships within the complete social circle graph. The relationship feature matrix within a training sample is used to represent the communication relationships between communication users belonging to the same complete social circle graph. In other words, this communication relationship feature matrix can represent the communication relationships between communication users within a social circle, including both local and remote network communication users. Furthermore, the communication relationship feature matrix is ​​generated based on a time series, representing the changes in the communication relationships between multiple communication users within a social circle over time.

[0050] The training labels corresponding to the training samples include the geographic locations of the permanent residences of multiple users on the same network who belong to the same complete graph of social circles. Because users within the same group often have similar behaviors, such as colleagues working in the same office building or family members living in the same residence, this solution generates training labels based on the permanent residences of users within the same social circle. These training labels represent the permanent residences of all users within the corresponding social circle.

[0051] In practical applications, if a user on the same network has multiple permanent locations within a unified social circle, the closeness of the relationships between the users can be determined based on the complete social circle graph. This closeness can then be used to generate corresponding weights. This weighted data can then be used to determine a geographic location representing the permanent location of each user within the social circle from the multiple permanent locations of the users on the same network within the same social circle. Furthermore, the training label can include a single geographic location or multiple locations with corresponding weights.

[0052] Optionally, the training labels generated by this solution correspond to the aforementioned training samples. The geographic locations in the training labels can be generated based on a time series, meaning that different time periods can correspond to different geographic locations. The training labels generated based on a time series can include multiple geographic locations corresponding to time periods. Specifically, the geographic location corresponding to any time period can be multiple geographic locations with corresponding weights. Optionally, the training labels can be expressed as vectors to reflect the temporal nature of the geographic locations.

[0053] S14: Training a pre-built time-series-based recurrent neural network model based on the multiple training samples and corresponding training labels.

[0054] A recurrent neural network (RNN) is a type of recurrent neural network that takes sequence data as input, recursively operates in the direction of the sequence's evolution, and has all nodes connected in a chain-like fashion. In this solution, the RNN model is trained using time-series-based training samples and corresponding training labels. This allows the RNN model to identify the permanent locations of inter-network communication users based on time. Since communication users often move based on time, and some users' locations are time-dependent, such as being at their office during work hours and at their residence during rest periods, this solution's time-series-based training samples and training labels can effectively train the model to accurately identify the permanent locations of inter-network communication users based on time.

[0055] Through the solution provided by the embodiment of the present application, the complete graph of the social circle generated based on the communication behavior information of the communication users of this network can represent the communication relationship between multiple communication users. Since the geographical locations of multiple communication users with similar behaviors are often the same, for example, multiple communication users with a colleague relationship are located in the same office location, or multiple communication users with a family relationship are located at the same address, etc., this solution identifies the permanent residence of the communication users of the other network with similar communication behaviors based on the communication behavior of the communication users of the other network and the communication users of this network, as well as the geographical location information of the communication users of this network. It can accurately identify the permanent residence of the communication users of the other network without obtaining the communication behavior information of the communication users of the other network. Moreover, this solution generates training samples based on time series and uses a time series-based model for training. It can effectively identify the permanent residence of the users of the other network whose geographical location changes over time. It can be widely applied to a variety of practical application scenarios, which is beneficial for operators to effectively carry out the poaching of customers from other networks.

[0056] Based on the solution provided in the above embodiment, optionally, Figure 2a As shown, the above step S12 includes:

[0057] S21: Generate a complete graph of the social circles corresponding to the communication users of the network to which each communication behavior information belongs in multiple time periods based on the communication behavior information of the multiple communication users of the network, wherein the vertices in the complete graph of social circles represent the communication users who have performed communication behaviors in the corresponding time period, the edges in the complete graph of social circles represent the communication behaviors performed between the communication users corresponding to the connected vertices in the corresponding time period, and the geographical location information of the communication users of the network corresponding to the multiple vertices in the complete graph of social circles performing communication behaviors in the corresponding time period is the same.

[0058] Because the location of communication users may change over time, the solution provided in the embodiment of the present application generates corresponding complete social circle graphs based on time periods. In actual applications, the above time periods can be pre-set according to actual needs. For example, 24 corresponding complete social circle graphs can be generated for 24 hours a day at one-hour intervals. Alternatively, the time periods can be set according to other rules.

[0059] For example, corresponding complete graphs of the social circles are generated for the working period from 9:00 (first day) to 21:00 (first day) and the rest period from 21:00 (first day) to 9:00 (second day), thereby generating two complete graphs of the social circles corresponding to different time periods, which is conducive to analyzing the communication relationships between communication users in these two time periods.

[0060] The following uses family relationships as an example to illustrate this. In this example, communication number A is the number of a user on a different network with an unknown permanent location:

[0061] If the communication number A in the communication circle has a complete graph relationship of communication circle of order 3 or above with other communication numbers, it is considered that the communication number A has an acquaintance relationship with other communication numbers.

[0062] If communication number A and the other numbers have the same permanent base station during the rest period (22:00-6:00), there is a possibility that communication number A and the other communication numbers have a family relationship.

[0063] For a complete subgraph with two or three resident cells with different rest times, the relationship can be determined by considering the historical resident cell information. For example, Figure 2b shows a complete graph of the interaction circle. Figure 2b If the communication number of the permanent base station C or the permanent base station B in the family relationship model B appears in the same permanent cell as the above-mentioned communication number A within a specific time period, the relationship between the communication users corresponding to the vertices in the graph can be determined as a non-strict family relationship.

[0064] Furthermore, communication relationships in practical applications are often complex. Since the complete graph of social circles and the residential communities may cover both daytime and nighttime working hours, for example, some communication numbers may reside in the same community. To further effectively distinguish different social circles, in addition to determining relationships in the above example, it is also possible to first remove the remaining social circles containing the same numbers based on the principle of prioritizing the largest social circle. The remaining social circles that contain the complete graph of social circles and do not overlap with the previously identified circles can be labeled as social circle 1, social circle 2, and so on.

[0065] Because the communication relationships between communication users often differ during different time periods, the complete graph of social circles corresponding to the time periods generated by the solution provided by the embodiments of this application can help more accurately identify the permanent locations of communication users on different networks during different time periods. In this solution, the geographical location information of the communication users on the same network performing communication behaviors during the corresponding time periods corresponding to the multiple vertices in the complete graph of social circles is the same, so that the geographical locations of the permanent locations of the communication users corresponding to the vertices in the generated complete graph of social circles are consistent, which is conducive to improving the accuracy of model recognition and determining the permanent locations of communication users on different networks based on the permanent locations of communication users on the same network. Moreover, the solution provided by the embodiments of this application can ensure the recall rate and the accuracy of the model.

[0066] Based on the solution provided in the above embodiment, optionally, Figure 3a As shown, the above step S21 includes:

[0067] S31: Generate a first complete graph of the interaction circle corresponding to the target local network communication user in the target time period based on the target communication behavior information of the target local network communication user, wherein the local network communication users corresponding to the vertices in the first complete graph of the interaction circle reside in the same target base station in the target time period.

[0068] In this step, the vertices in the first complete graph of social circles may represent users who had direct communication relationships with the target local network user during the target period. Furthermore, the users corresponding to the vertices in the first complete graph of social circles resided at the same target base station during the target period. In other words, these users were geographically close to each other during the target period.

[0069] S32: Obtain at least one second complete graph of social circles to which the communication users corresponding to the vertices in the first complete graph of social circles belong, wherein the communication users of the local network corresponding to the vertices in the second complete graph of social circles reside in the target base station during the target time period.

[0070] In practical applications, the communication relationships between communication users are often complex, and a single communication user may belong to multiple different complete graphs of social circles. In this step, a second complete graph of social circles is further obtained based on the communication users corresponding to the vertices in the first complete graph of social circles. Since the local network communication users corresponding to the vertices in the second complete graph of social circles also resided at the target base station during the target period, it can be determined that the local network communication users corresponding to the vertices in the second complete graph of social circles, while not in direct communication with the target local network communication user during the target period, are geographically close to each other. Alternatively, the communication users corresponding to the vertices in the second complete graph of social circles can be considered to have an indirect communication relationship with the target communication user.

[0071] S33: Merge the first complete graph of social circles and the at least one second complete graph of social circles to obtain the complete graph of social circles corresponding to the target local network communication user in the target time period.

[0072] Based on the first and second complete graphs of social circles, since the communication users corresponding to the nodes in these two complete graphs resided at the target base station during the target period, and therefore these communication users were geographically close during the target period, the two complete graphs can be merged to obtain the complete graph of social circles corresponding to the target local network communication user during the target period. The merged complete graph includes multiple communication users who resided at the target base station during the target period, and these communication users have direct or indirect communication relationships with the target communication user.

[0073] The following description will be made by taking a colleague relationship as an example. In this example, it is assumed that the communication number A is the number of a communication user on an alien network whose permanent location is unknown.

[0074] If communication number A has a complete graph of social circles of order 3 or higher with other communication numbers, then number A is considered to have an acquaintance relationship with the other numbers. The size of the complete graph of social circles can be accurately assessed by porting in and out numbers to obtain a more accurate size of the social circle.

[0075] If communication number A and other communication numbers have the same resident base station during working hours (8:00-17:00), then communication number A and other communication numbers can be artificially regarded as colleagues, such as Figure 3b As shown, the left side of the figure shows a colleague relationship model A, in which multiple local network communication numbers reside in the same base station.

[0076] Figure 3b The figure also shows the colleague relationship merging model. Because departments within an organization often communicate closely, communication with the outside world is often limited. To merge these discrete groups, in the colleague relationship merging model, if the preferred permanent base station A complete graph A and the preferred permanent base station A complete graph B have two 3-complete graphs or one or more 4-complete graphs in communication, since both permanent base stations are A, the permanent base station A complete graph A and the permanent base station A complete graph B are merged.

[0077] Figure 3b The figure also shows a colleague relationship discarding mode. If there is a communication relationship but some vertices correspond to communication users with different permanent base stations, the vertices with different permanent base stations can be discarded and the vertices with the same permanent base stations can be merged.

[0078] Furthermore, for communication relationships represented by the colleague-discarding model, if the communication user corresponding to a vertex in the complete graph B of resident base station B has a close communication relationship with the communication user corresponding to a vertex in the complete graph A of resident base station A, the vertices corresponding to communication users at different resident base stations can be merged based on the closeness of the communication relationship. Accordingly, weights can be assigned to the vertices corresponding to different resident base stations in the merged complete graph to comprehensively represent the permanent locations of the communication users corresponding to the vertices in the complete graph.

[0079] The complete graph of the communication circle obtained by combining the solutions provided in the embodiments of the present application can more comprehensively express the direct and indirect communication relationships of the target communication user within the target time period, thereby optimizing the training effect of the model and enabling the model to identify the permanent residence of users of different network communications based on direct and indirect communication relationships.

[0080] Based on the solution provided in the above embodiment, optionally, Figure 4 As shown, the above step S13 includes:

[0081] S41: Generate a corresponding communication relationship adjacency matrix based on the target social circle complete graph and target communication behavior information corresponding to the target network communication user, wherein the communication relationship adjacency matrix includes a call adjacency matrix and / or a text message adjacency matrix, and the communication relationship adjacency matrix represents the number of times communication behaviors are performed between communication users.

[0082] In this embodiment, a one-dimensional array is used to store all vertex data in the complete graph of social circles, and a two-dimensional array is used to store data on the relationships (edges) between vertices. This two-dimensional array is called an adjacency matrix. In this step, the communication relationship adjacency matrix can include two types: a call adjacency matrix and a text message adjacency matrix. The generation methods of these two adjacency matrices will be illustrated in the subsequent examples.

[0083] S42: Generate training samples corresponding to the target local network communication users according to the communication relationship adjacency matrix.

[0084] Using the adjacency matrix to represent the complete graph of the communication circle described in this example can intuitively indicate whether there is a communication relationship between the communication users corresponding to two vertices, so that the generated training samples have communication relationship features that are convenient for model learning.

[0085] S43: Generate a training label corresponding to the target local network communication user according to the geographical location information of the communication user performing the communication behavior corresponding to the vertex in the target social circle complete graph.

[0086] The generated training labels can contain multiple geographic locations, which can be expressed in the form of vectors. For example, if there are three geographic locations corresponding to the training sample, a three-dimensional vector training label can be generated.

[0087] Through the solution provided in the embodiment of the present application, the communication relationship between communication users can be intuitively and fully expressed through the adjacency matrix, so that the generated training samples and corresponding training labels are conducive to efficient model learning, thereby more accurately identifying the permanent residence of communication users in different networks.

[0088] Based on the solution provided in the above embodiment, optionally, Figure 5 As shown, the communication relationship adjacency matrix includes a call adjacency matrix;

[0089] The above step S41 includes:

[0090] S51: Generate a first vector based on the target complete graph of social circles and the target communication behavior information, wherein the values ​​of each dimension of the first vector respectively represent the duration of calls between the target local network communication user and the communication users corresponding to each vertex in the target complete graph of social circles;

[0091] S52: generating a plurality of second vectors based on the target communication behavior information, wherein each second vector represents the number of calls between the target local network communication user and the communication users corresponding to each vertex in the target social circle complete graph within a plurality of time periods based on time sequence;

[0092] S53: Generate the call adjacency matrix according to the first vector and the plurality of second vectors.

[0093] The following is an example to illustrate this solution. The solution provided in the embodiment of the present application is used to generate a call adjacency matrix A. v .

[0094] Since voice calls between communication numbers involve both incoming and outgoing calls, and these calls can be divided into two types: number of calls and number of minutes. To accurately describe the relationship between voice calls, we can use minutes and number of calls, respectively, with the number of minutes combining both incoming and outgoing calls. For example, for number A, the first vector [0, 10, 25, …, 0] can be used to describe the number of 10-minute and 25-minute calls between number A, number B, and number C over the past month. An adjacency matrix is ​​also constructed based on the number of calls. For example, for number A, the second vector [0, 0, 5, 3, …, 0] can be used to represent a total of five incoming or outgoing calls with number C over a period of time. Based on these first and second vectors, a corresponding call adjacency matrix is ​​generated to fully describe the communication relationship between the users in terms of number of calls and call duration.

[0095] In addition, the call described in the embodiments of the present application may refer to a voice call, a video call, or other call forms.

[0096] Based on the solution provided in the above embodiment, optionally, Figure 6 As shown, the communication relationship adjacency matrix includes a short message adjacency matrix;

[0097] The above step S41 includes:

[0098] S61: generating a third vector based on the target complete graph of the social circle and the target communication behavior information, wherein the values ​​of each dimension of the third vector respectively represent the number of times the target local network communication user and the communication users corresponding to each vertex in the target complete graph of the social circle send text messages to each other;

[0099] S62: generating a plurality of fourth vectors based on the target communication behavior information, wherein each fourth vector represents the number of times the target local network communication user and the communication users corresponding to each vertex in the target social circle complete graph exchanged text messages within a plurality of time periods based on time sequence;

[0100] S63: Generate the short message adjacency matrix according to the third vector and the plurality of fourth vectors.

[0101] The following example illustrates this solution. The solution provided in this embodiment is used to generate the SMS adjacency matrix A. s The generation method is similar to the method of generating the call adjacency matrix in the above example. Since the construction rule of SMS edges is that only mutual communication is an edge, if number A and number B have 10 one-way SMS interactions within a period of time, but if number A and B have only two mutual SMS interactions, then the values ​​of A and B in the SMS adjacency matrix are 2.

[0102] Based on the solution provided by the above example, a text message adjacency matrix and a call adjacency matrix can be generated. In order to combine different forms of communication relationships, different forms of adjacency matrices can be combined through the following embodiments to comprehensively express various forms of communication relationships.

[0103] Based on the solution provided in the above embodiment, optionally, Figure 7 As shown, the communication relationship adjacency matrix includes a call adjacency matrix and a text message adjacency matrix;

[0104] The above step S42 includes:

[0105] S71: Normalizing the call adjacency matrix and the text message adjacency matrix respectively to obtain a normalized call adjacency matrix and a normalized text message adjacency matrix;

[0106] S72: Adding the normalized call adjacency matrix and the normalized SMS adjacency matrix to obtain a fusion feature matrix;

[0107] S73: Generate a training sample corresponding to the target local network communication user according to the inverse of the fusion feature matrix.

[0108] In the embodiment of the present application, first, the two adjacent matrices are normalized and preprocessed by preprocessing. The specific rules of the preprocessing can be set accordingly according to the actual need to merge the adjacency matrix. Then, for the two normalized adjacency matrices, they are fused into the adjacency matrix A by adding them together. Finally, the final fusion feature matrix is ​​obtained by processing them by 1 / A. Since the longer the communication behavior and the higher the frequency, it indicates that the users are at a certain distance, the inverse is used in the embodiment of the present application to process the adjacency matrix after the addition.

[0109] In addition, after obtaining the communication information of different time periods, since users have different locations on weekends and weekdays in certain specific scenarios, they have different behavioral characteristics. In order to effectively capture the communication behavior of users at different times and accurately extract the relationship information between users, the following example illustrates the weekly time series location recognition model based on different time periods. Figure 8 , the relationship circle in this example includes the complete graph of the communication circle in the above embodiment.

[0110] Step 1: Extract the top three permanent locations of different relationship circles of communication users in this network in different time periods, as well as the permanent locations during the daytime and nighttime on weekdays and weekends.

[0111] Step 2: Select the top 5 relationship circles according to the number of the same permanent locations in the relationship circles every day of the week. If there are none, fill them with the selected ones. -1 , using the summation method ∑A -1 Select the top 3 permanent locations in each relationship circle. Construct the permanent location vector w for each relationship circle h The above five relationship circles can be expressed as

[0112] Step 3: Use the RNN model to process the generated attribute features of the permanent location of the network in different relationship circles. To calculate the result after n+1 iterations, is the input for day T.

[0113] Step 4: In the time series model, combine the RNN strategy to extract the daily features. Through the trainable matrix h T and h′ T are weighted inputs for the Tth day and the T-1th day, so the time series feature calculation is obtained as follows: T =α(U T h T +w T h′ T )+β, the time series feature is used as the input of the next time series step. The result output at the Tth moment is expressed as

[0114] Step 5: Cross entropy is used as the loss function of the cyclic neural network for 7 days This loss function represents the degree of compliance for each different time period of the seven days of the week.

[0115] Step 6: Output the top three permanent locations in different time periods of the day or night in different spatiotemporal scenarios every day of the week.

[0116] The solution provided by the embodiments of the present application can extract features from the communication behavior of users on the same network, even when the location of users on other networks cannot be directly obtained. This allows the relationship between users on the same network and multiple permanent location candidates in their multiple relationship circles to be discovered based on the communication behavior of users on the same network. This solution also improves recognition accuracy by combining different time periods and spatiotemporal scenarios across seven days of the week. The solution provided by the embodiments of the present application can extract a complete graph of user call behavior data, and by combining rules such as permanent locations for different time periods, it facilitates further filtering and purification of the multiple relationship circles identified by the existing model on the same network. The solution provided by the embodiments of the present application utilizes a fused communication behavior matrix and pre-processed weights to select the top-ranked permanent locations. Furthermore, by training a model based on time periods (e.g., a week), it can extract user behavior features at different time points throughout the week, as well as the permanent location latent vector features of the user's communication circle. This allows the identification of permanent locations for users on the same network and users on other networks at different time periods across seven days of the week, effectively improving the accuracy of identifying the permanent locations of users on the other network.

[0117] The solution provided by the embodiment of the present application is further described below in conjunction with the device. Figure 9 As shown, the device can extract sufficient relationship circle sample data by extracting communication behaviors within the local network and using an effective relationship circle enhancement recognition model. It also constructs different weight matrices based on communication behaviors and extracts call patterns of inter-network relationship networks through time-series relationship circles based on the local network patterns.

[0118] First, we extract communication behavior data between users from the network's call logs, including information about voice and text message interactions between users. This information is then collected from multiple users on the network. We extract the minutes and times of voice calls, as well as the number of text messages, and store them in a database. To facilitate subsequent relationship enhancement, we extract signaling data from existing location databases or call records, including call base stations and power on / off locations, and store this as geographic location information.

[0119] Based on the preset complete graph rules of the social circle, the social circles extracted by the social circle model in different time periods, such as the relationship networks of family, friends, colleagues, etc., are further filtered according to the intersection method, and a more accurate relationship network obtained with the maximum recall rate is extracted to prepare sample data for subsequent social relationship extraction.

[0120] The adjacency matrix of call and text communication behaviors is obtained from the pre-stored communication behavior data, and the resident vector is filtered by fusion weighting.

[0121] Then, a mining algorithm for the time series relationship graph is constructed to update the permanent location vectors of the top five relationship networks at different times. The number of relationship networks used can be predefined. Simultaneously, the network weights at different times in the time series network are continuously updated, extracting the forward propagation of vector weights for different days of the week to ensure the continuity of the permanent locations of adjacent days.

[0122] Finally, the weekly time series model is used to identify and recommend the top three permanent locations of users from different networks at different times of the day and night.

[0123] The solution provided in the embodiments of the present application can realize the identification of the permanent location of users on different networks. This technical solution has the following technical advantages:

[0124] 1. It can optimize the relationship networks of communication users on this network, such as colleagues, family, friends, etc. It can combine the family circle mining model that has been mined on this network to correct and purify the results when the recall rate is maximized, providing accurate sample data for permanent residence identification.

[0125] 2. Comprehensively consider the communication behavior adjacency matrix and integrate different communication behavior data to construct the residence vector screening weights for different relationship networks. By using the inverse method, we can more objectively measure the proximity of users on different networks to the residence of different relationship circles.

[0126] 3. Build a weekly time series model for different time periods to accurately extract the relationship characteristics between users on different networks and the permanent locations of their different relationship networks, ensuring the accuracy of the model in identifying the permanent locations of users on different networks at different time periods and on different days of the week.

[0127] The solution provided in the embodiment of the present application can realize accurate identification of the permanent residence of users on different networks, is conducive to the mining and expansion of different network customers, is conducive to multi-touchpoint integrated marketing, and can effectively support marketing personnel in carrying out resident different network marketing.

[0128] In order to solve the problems existing in the prior art, the embodiment of the present application also provides a training device 100 for identifying a permanent location of a user in a different network. Figure 10 Shown, including:

[0129] Acquisition module 101, acquires communication behavior information of multiple local network communication users, wherein the communication behavior information includes geographical location information and time information of the communication behaviors performed by the local network communication users and the communication users of the other networks;

[0130] A first generating module 102 generates, based on the communication behavior information of the plurality of local network communication users, a complete graph of the social circles corresponding to the local network communication users to which each communication behavior information belongs, wherein the complete graph of the social circles includes vertices representing the communication users and edges connecting different vertices, wherein the edges represent communication behaviors performed between the communication users corresponding to the connected vertices;

[0131] A second generation module 103 generates a plurality of training samples and corresponding training labels based on the complete graphs of social circles corresponding to the plurality of local network communication users and the corresponding communication behavior information, wherein any of the training samples includes a time-series communication relationship feature matrix between communication users belonging to the same complete graph of social circles, and the training labels include the geographic locations of the permanent locations where the plurality of local network communication users in the complete graph of social circles to which the alien network communication users belong perform communication behaviors;

[0132] The training module 104 trains the pre-built time-series-based recurrent neural network model based on the multiple training samples and corresponding training labels.

[0133] Through the device provided in the embodiment of the present application, the complete graph of the social circle generated based on the communication behavior information of the communication users of this network can represent the communication relationship between multiple communication users. Since the geographical locations of multiple communication users with similar behaviors are often the same, for example, multiple communication users with a colleague relationship are located in the same office location, or multiple communication users with a family relationship are located at the same address, etc., this solution identifies the permanent residence of the communication users of the other network with similar communication behaviors based on the communication behavior of the communication users of the other network and the communication users of this network, as well as the geographical location information of the communication users of this network. It can accurately identify the permanent residence of the communication users of the other network without obtaining the communication behavior information of the communication users of the other network. Moreover, this solution generates training samples based on time series and uses a time series-based model for training. It can effectively identify the permanent residence of the communication users of the other network whose geographical location changes over time. It can be widely applicable to a variety of practical application scenarios, which is beneficial for operators to effectively carry out the poaching of customers from other networks.

[0134] The aforementioned modules in the apparatus provided in the embodiments of the present application may also implement the method steps provided in the aforementioned method embodiments. Alternatively, the apparatus provided in the embodiments of the present application may further include other modules in addition to the aforementioned modules to implement the method steps provided in the aforementioned method embodiments. Furthermore, the apparatus provided in the embodiments of the present application can achieve the technical effects achievable by the aforementioned method embodiments.

[0135] Preferably, an embodiment of the present application also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and runnable on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned training method embodiment for identifying a permanent location model for users on a different network are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0136] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the above-mentioned embodiment of the method for training a resident location recognition model of a different network user is implemented, and the same technical effect is achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0137] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1A step that specifies a function in one or more boxes.

[0141] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0142] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0143] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0144] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0145] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A training method for a model for identifying the permanent location of users on a different network, characterized in that: include: Acquire communication behavior information of multiple local network communication users, wherein the communication behavior information includes geographical location information and time information of the communication behaviors performed by the local network communication users and the communication users of the other networks; Generating a complete graph of social circles corresponding to the communication users of the local network to which each communication behavior information belongs based on the communication behavior information of the multiple communication users of the local network, wherein the complete graph of social circles includes vertices representing the communication users and edges connecting different vertices, wherein the edges represent communication behaviors performed between the communication users corresponding to the connected vertices; Generate multiple training samples and corresponding training labels based on the complete graphs of social circles corresponding to the multiple local network communication users and the corresponding communication behavior information, wherein any of the training samples includes a time-series communication relationship feature matrix between communication users belonging to the same complete graph of social circles, and the training labels include the geographical locations of the permanent residences where the multiple local network communication users in the complete graph of social circles to which the alien network communication users belong perform communication behaviors; A pre-built time-series-based recurrent neural network model is trained based on the multiple training samples and corresponding training labels.

2. The method according to claim 1, wherein Generating a complete graph of the social circles corresponding to the local network communication users to which each communication behavior information belongs based on the communication behavior information of the plurality of local network communication users includes: Based on the communication behavior information of the multiple communication users of the local network, a complete graph of the communication circles corresponding to the communication users of the local network to which the communication behavior information belongs in multiple time periods is generated, wherein the vertices in the complete graph of the communication circles represent the communication users who have performed communication behaviors in the corresponding time period, the edges in the complete graph of the communication circles represent the communication behaviors performed between the communication users corresponding to the connected vertices in the corresponding time period, and the geographical location information of the communication users of the local network corresponding to the multiple vertices in the complete graph of the communication circles when performing communication behaviors in the corresponding time period is the same.

3. The method according to claim 2, wherein Generating, based on the communication behavior information of the plurality of local network communication users, a complete graph of the social circles corresponding to the local network communication users to which each communication behavior information belongs in a plurality of time periods, including: generating a first complete graph of the communication circles corresponding to the target local network communication user during a target period according to the target communication behavior information of the target local network communication user, wherein the local network communication users corresponding to the vertices in the first complete graph of the communication circles reside in the same target base station during the target period; Acquire at least one second complete graph of social circles to which the communication users corresponding to the vertices in the first complete graph of social circles belong, wherein the communication users of the local network corresponding to the vertices in the second complete graph of social circles reside in the target base station during the target time period; The first complete graph of social circles and the at least one second complete graph of social circles are combined to obtain a complete graph of social circles corresponding to the target local network communication user in the target time period.

4. The method according to claim 1, wherein Generating a plurality of training samples and corresponding training labels according to the complete graphs of the social circles corresponding to the plurality of local network communication users and the corresponding communication behavior information includes: Generate a corresponding communication relationship adjacency matrix based on the target interaction circle complete graph and target communication behavior information corresponding to the target local network communication user, wherein the communication relationship adjacency matrix includes a call adjacency matrix and / or a text message adjacency matrix, and the communication relationship adjacency matrix represents the number of communication behaviors performed between the communication users; Generating a training sample corresponding to the target local network communication user according to the communication relationship adjacency matrix; The training labels corresponding to the target local network communication users are generated according to the geographical location information of the communication users performing communication behaviors corresponding to the vertices in the target social circle complete graph.

5. The method according to claim 4, wherein The communication relationship adjacency matrix includes a call adjacency matrix; The corresponding communication relationship adjacency matrix is ​​generated according to the target communication circle complete graph and target communication behavior information corresponding to the target local network communication user, including: generating a first vector according to the target complete graph of the social circle and the target communication behavior information, wherein the values ​​of each dimension of the first vector respectively represent the duration of the call between the target local network communication user and the communication user corresponding to each vertex in the target complete graph of the social circle; generating a plurality of second vectors based on the target communication behavior information, wherein each second vector represents the number of calls between the target local network communication user and the communication users corresponding to each vertex in the target social circle complete graph within a plurality of time periods based on time sequence; The call adjacency matrix is ​​generated according to the first vector and the plurality of second vectors.

6. The method according to claim 4, wherein The communication relationship adjacency matrix includes a short message adjacency matrix; The corresponding communication relationship adjacency matrix is ​​generated according to the target communication circle complete graph and target communication behavior information corresponding to the target local network communication user, including: generating a third vector based on the target complete graph of the social circle and the target communication behavior information, wherein the values ​​of each dimension of the third vector respectively represent the number of times the target local network communication user and the communication users corresponding to each vertex in the target complete graph of the social circle exchange text messages; generating a plurality of fourth vectors based on the target communication behavior information, wherein each fourth vector represents the number of times the target local network communication user and the communication users corresponding to each vertex in the target social circle complete graph exchanged text messages within a plurality of time periods based on time sequence; The short message adjacency matrix is ​​generated according to the third vector and the plurality of fourth vectors.

7. The method according to any one of claims 4 to 6, characterized in that The communication relationship adjacency matrix includes a call adjacency matrix and a text message adjacency matrix; The step of generating a training sample corresponding to the target local network communication user according to the communication relationship adjacency matrix includes: Normalizing the call adjacency matrix and the text message adjacency matrix respectively to obtain a normalized call adjacency matrix and a normalized text message adjacency matrix; Adding the normalized call adjacency matrix and the normalized text message adjacency matrix to obtain a fused feature matrix; A training sample corresponding to the target local network communication user is generated according to the inverse of the fusion feature matrix.

8. A training device for identifying a model for a permanent location of a user on a different network, characterized in that: include: An acquisition module is configured to acquire communication behavior information of multiple local network communication users, wherein the communication behavior information includes geographical location information and time information of the communication behaviors performed by the local network communication users and the communication users of the other networks; A first generating module generates, based on the communication behavior information of the plurality of local network communication users, a complete graph of the social circles corresponding to the local network communication users to which each communication behavior information belongs, wherein the complete graph of the social circles includes vertices representing the communication users and edges connecting different vertices, wherein the edges represent communication behaviors performed between the communication users corresponding to the connected vertices; A second generation module generates a plurality of training samples and corresponding training labels based on the complete graphs of social circles corresponding to the plurality of local network communication users and the corresponding communication behavior information, wherein any of the training samples includes a time-series communication relationship feature matrix between communication users belonging to the same complete graph of social circles, and the training labels include the geographic locations of the permanent locations where the plurality of local network communication users in the complete graph of social circles to which the alien network communication users belong perform communication behaviors; A training module trains a pre-built time-series-based recurrent neural network model based on the multiple training samples and corresponding training labels.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the method according to any one of claims 1 to 7 when executed by the processor.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images