A customer portrait method and device based on federated learning

Abstract

Embodiments of the present application provide a customer portrait method and device based on federated learning, and relate to the technical field of data mobile communication. The customer portrait method based on federated learning obtains DPI log data, analyzes the DPI log data, obtains ur1, mdn and userID, and establishes a mapping relationship between the userID and the mdn. POI position information is encoded, and the number of encoding bits is adjusted. The sample set is encrypted and sample alignment is performed, and joint modeling is performed in a federated learning manner. The userID is issued to ensure compliance of the cooperation content, the flexibility of adjusting the position accuracy is increased through geohash encoding, the complexity of position distance calculation is reduced, and data fusion is realized under the premise of ensuring the safety of the original data of both parties through longitudinal federated learning, so that the customer portrait is obtained. Compared with the prior art, the business is more in line with the compliance requirements, the safety is higher, the calculation method is more flexible, and the portrait accuracy is higher.

Description

Technical Field

[0001] This invention relates to the field of mobile data communication technology, and in particular to a customer profiling method and apparatus based on federated learning. Background Technology

[0002] With the successive promulgation and implementation of the Data Security Law, the Cybersecurity Law, and the Personal Information Protection Law, data cannot be directly bought or sold. Any organization or individual collecting data must do so through legal and legitimate means and must not steal or obtain data through other illegal means. Where laws and administrative regulations stipulate the purpose and scope of data collection and use, data must be collected and used within the purpose and scope stipulated by laws and administrative regulations. With the further development of big data, emphasizing data privacy and security has become a global trend. At the same time, most industry data exhibits the phenomenon of data silos. How to conduct cross-organizational data collaboration while meeting user privacy protection, data security, and government regulations is a major challenge for artificial intelligence practitioners. Traditional machine learning algorithms require users to upload source data to high-computing cloud servers for centralized training. This method leads to uncontrollable data flow and the leakage of sensitive data. Federated learning can collaboratively train the globally optimal model in scenarios where multiple data sources are aggregated. It transfers the data storage and model training stages of machine learning to local users, effectively ensuring user privacy and security by interacting only with the central server for model updates.

[0003] Federated learning can protect raw data, ensuring information security during big data exchange, protecting terminal data and personal data privacy, and guaranteeing legality and compliance. It enables efficient machine learning among multiple participants or computing nodes. However, cooperation authorization relies on contractual agreements and manual spot checks, increasing the workload of manual auditing and posing compliance risks. In terms of location computing, electronic fence technology requires a large amount of computation. Summary of the Invention

[0004] This application provides a customer profiling method and apparatus based on federated learning, which solves the problems of location calculation complexity, inflexible precision adjustment, and incomplete authorization links in the prior art, and realizes joint profiling to accurately complete user profiles.

[0005] This application provides a customer profiling method based on federated learning for operators, including the following steps:

[0006] S1. Obtain DPI log data, analyze DPI log data, obtain ur1, mdn, userID, and establish a mapping relationship between userID and mdn;

[0007] S2. Encode the POI location information and adjust the number of bits in the encoding;

[0008] S3. Encrypt and align the sample sets, and perform joint modeling through federated learning.

[0009] Furthermore, S1 specifically includes the following steps:

[0010] S11. Obtain DPI log data for the domain specified by the partner;

[0011] S12. Analyze the DPI log data of the specified domain name to obtain ur1 and mdn;

[0012] S13. According to the negotiation rules, analyze ur1 to obtain userID;

[0013] S14. Establish the mapping relationship between userID and mdn.

[0014] Furthermore, the domain name specified in S11 is abc.com, and the negotiation rule in S13 is mu&userid=.

[0015] Furthermore, S2 specifically includes the following steps:

[0016] S21. Perform geohash encoding on the specified POI location information, and adjust the encoding bit length according to the precision requirements;

[0017] S22. Obtain the MDN that has accessed the specified physical range through signaling data;

[0018] S23. Obtain the userID corresponding to S22mdn according to the mapping relationship of S1.

[0019] Furthermore, if the encoding in S22 is 6 bits, then the specified physical range is: grid width 1.22km, grid height 0.61km.

[0020] Furthermore, S3 specifically includes the following steps:

[0021] S31. Align the encrypted samples of userID in the free sample set and userIDs that have accessed the specified area.

[0022] S32. Perform longitudinal federated learning joint modeling on the sample set.

[0023] Furthermore, the sample set in S32 includes: features and tags provided by the partner, and features provided by the operator.

[0024] Furthermore, the features provided by the partner include: the duration of user app usage, the number of times the app is used, and the terminal brand. The tags provided by the partner include: whether the user is a high-net-worth, high-spending user.

[0025] Furthermore, the features provided by the operator include: user gender, age, network duration, package amount, stability of permanent residence, and internet access behavior.

[0026] A customer profiling device based on federated learning includes a mapping module, a location encoding module, and a federated learning module, wherein:

[0027] The mapping module is used to obtain DPI log data, analyze DPI log data, obtain ur1, mdn, and userID, and establish a mapping relationship between userID and mdn;

[0028] The location encoding module is used to encode POI location information and adjust the encoding bit length.

[0029] The federated learning module is used to perform encrypted sample alignment on the sample set and to perform joint modeling through federated learning.

[0030] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0031] By issuing user IDs to partners, we ensure that the users used in the modeling process are authorized by the partners, thus ensuring the compliance of the collaborative content and solving the problem of incomplete authorization links.

[0032] By using geohash encoding for position calculation, the problems of complexity and inflexibility in precision adjustment of existing position calculation technologies are solved, increasing the flexibility of adjusting position precision and reducing the complexity of position distance calculation.

[0033] By using vertical federated learning for data interaction, the security of raw data is ensured, data fusion is achieved, a secure model for data circulation is provided, and the accuracy of user profiles is improved. Compared with existing technologies, it is more compliant with requirements, more secure, has more flexible calculation methods, and more accurate profiles. Attached Figure Description

[0034] Figure 1 This is a flowchart of the customer profiling method based on federated learning in this invention;

[0035] Figure 2 A flowchart for establishing customer profile models for the partners and operators of this invention. Detailed Implementation

[0036] This invention, relying on operator DPI data and signaling data, proposes a customer profiling method based on federated learning. By having a partner issue a userID, the operator obtains DPI log data from a specified domain of the partner. Geohash encoding is used to increase the flexibility of adjusting location precision. Joint modeling is performed through vertical federated learning. While ensuring the security of the original data from both parties, data fusion is achieved to obtain a customer profile. Compared with existing technologies, this method effectively ensures the security of data interaction actions and the security of the original data, is more compliant with compliance requirements, has higher security, more flexible calculation methods, and higher profile accuracy.

[0037] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0038] A customer profiling method based on federated learning for operators includes the following steps:

[0039] S1. Obtain DPI log data, analyze DPI log data, obtain ur1, mdn, userID, and establish a mapping relationship between userID and mdn;

[0040] S2. Encode the POI location information and adjust the number of bits in the encoding;

[0041] S3. Encrypt and align the sample sets, and perform joint modeling through federated learning.

[0042] In this embodiment, as Figure 1 As shown, the partner issues a unique user ID: userID. The operator obtains DPI log data from the partner's designated domain, extracts the ur1 and mobile phone number mdn from the DPI log data, and retrieves the userID issued by the partner from the ur1 according to the negotiated rules. A mapping relationship between userID and mdn is established. The POI location information specified by the partner is encoded, and the encoding bit length is adjusted according to the precision requirements. The intersection of the sample sets is performed by aligning encrypted samples. The partner provides user features and tags, and the operator provides tags. Joint modeling is performed through federated learning to complete the user profile. The userID issued by the partner ensures compliance with cooperation requirements; geohash encoding increases the flexibility of adjusting location precision and reduces the complexity of location distance calculation; and joint modeling through vertical federated learning achieves data fusion and obtains customer profiles while ensuring the security of the original data of both parties. Compared with existing technologies, this approach is more compliant with compliance requirements, more secure, has a more flexible calculation method, and provides more accurate profiles.

[0043] S1 specifically includes the following steps:

[0044] S11. Obtain DPI log data for the domain specified by the partner;

[0045] S12. Analyze the DPI log data of the specified domain name to obtain ur1 and mdn;

[0046] S13. According to the negotiation rules, analyze ur1 to obtain userID;

[0047] S14. Establish the mapping relationship between userID and mdn.

[0048] The domain name specified in S11 is abc.com, and the negotiation rule in S13 is mu&userid=.

[0049] In this embodiment, the compliance of the cooperative content is ensured by issuing userIDs through the partner: the partner has the registration information of its own app or website users, including user IDs (userIDs) issued by the partner's network with internal identifiers for its own users, ensuring that the users involved in subsequent federated learning are authorized users. The format is http: / / abc.com / api / mu&userid=11111111. When a user uses the app or website, the partner appends the userID to the request link. The operator obtains DPI log data from the domain specified by the partner. The DPI log data for the specified domain "abc.com" includes ur1 and mdn. The operator analyzes the URL data in the DPI log data and obtains the userID according to the negotiation rule "mu&userid=". The operator establishes a mapping relationship between userID and mdn internally.

[0050] S2 specifically includes the following steps:

[0051] S21. Perform geohash encoding on the specified POI location information, and adjust the encoding bit length according to the precision requirements;

[0052] S22. Obtain the MDN that has accessed the specified physical range through signaling data;

[0053] S23. Obtain the userID corresponding to S22mdn according to the mapping relationship of S1.

[0054] If the encoding in S22 is 6 bits, then the specified physical range is: grid width 1.22km, grid height 0.61km.

[0055] In this embodiment, as Figure 2As shown, geohash encoding increases the flexibility of adjusting location precision and reduces the complexity of location distance calculation: the partner provides specified POI location information, the operator performs six-bit geohash encoding on the POI location information, converting latitude and longitude into a one-dimensional string. The string length is adjusted according to precision requirements. Through the mapping relationship between signaling data collected by the operator, userIDs, and MDN, a set of userIDs that have visited a grid range of 1.22km in width and 0.61km in height is obtained, providing a sample set of userIDs who have visited the specified range and are partner users for subsequent calculations. Distance does not need to be calculated using two points of latitude and longitude, reducing computational load and improving efficiency. Compared to LBS, this improves the efficiency of latitude and longitude retrieval of spatial POI data and increases cache hit rate.

[0056] S3 specifically includes the following steps:

[0057] S31. Align the encrypted samples of userID in the free sample set and userIDs that have accessed the specified area.

[0058] S32. Perform longitudinal federated learning joint modeling on the sample set.

[0059] The sample set in S32 includes: features and tags provided by partners, and features provided by operators.

[0060] The features provided by the partner are: the duration of user use of the app, the number of times the app is used, and the terminal brand. The tags provided by the partner are: whether the user is a high-net-worth, high-spending user.

[0061] Operators provide features including: user gender, age, network duration, package amount, stability of residence, and internet usage behavior.

[0062] In this embodiment, as Figure 2As shown, joint modeling is performed through vertical federated learning. While ensuring the security of the original data from both parties, data fusion is achieved to obtain customer profiles. Compared with existing technologies, this approach is more compliant with regulations, more secure, has more flexible calculation methods, and provides more accurate profiles. The security of user sample sets from both parties is protected through encrypted sample alignment. For example, the intersection of sample sets is performed: the partner provides a free sample set of userIDs, and the operator provides userIDs that have accessed specified regions. Encrypted sample alignment is then performed. The partner provides features and labels for the sample set, and the operator provides features. Joint modeling is then performed in specified scenarios through vertical federated learning to complete the user profile. The partner provides labels indicating whether the user is a high-net-worth, high-spending user, and provides features such as app usage duration, app usage frequency, and terminal brand. The operator provides features such as user gender, age, network duration, package amount, residence stability, and internet behavior. Training and prediction sets are divided, machine learning algorithms are selected, and vertical federated learning is performed to obtain a joint model of the customer profile.

[0063] A customer profiling device based on federated learning includes a mapping module, a location encoding module, and a federated learning module, wherein:

[0064] The mapping module is used to obtain DPI log data, analyze DPI log data, obtain ur1, mdn, and userID, and establish a mapping relationship between userID and mdn;

[0065] The location encoding module is used to encode POI location information and adjust the encoding bit length.

[0066] The federated learning module is used to perform encrypted sample alignment on the sample set and to perform joint modeling through federated learning.

[0067] In this embodiment, the partner issues an internal userID, the operator collects DPI data, obtains DPI log data including URL and MDN of a specified domain, obtains the userID according to the negotiation rules, and then obtains the mapping relationship between userID and MDN. The partner provides POI data, which is geohash encoded by the operator and associated with the signaling data collected by the operator to obtain a set of userIDs that have visited a specified location range. The encrypted samples are aligned by aligning the userIDs issued by the partner with the userIDs provided by the operator that have visited the specified range. Through federated learning, the features and tags provided by the partner and the features provided by the operator are jointly modeled to obtain a joint model of customer profile.

[0068] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0069] By issuing userIDs to partners, the compliance of collaborative content is ensured; geohash encoding increases the flexibility of adjusting location precision and reduces the complexity of location distance calculation; and joint modeling is carried out through vertical federated learning, achieving data fusion and obtaining customer profiles while ensuring the security of the original data of both parties. Compared with existing technologies, this approach is more compliant with business requirements, more secure, has more flexible calculation methods, and more accurate profiles.

[0070] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0075] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A customer profiling method based on federated learning, characterized in that, For use by operators, the following steps are included: S1. Obtain DPI log data, analyze DPI log data, obtain ur1, mdn, userID, and establish a mapping relationship between userID and mdn; S2. Encode the POI location information and adjust the number of bits in the encoding; S3. Encrypt and align the sample sets, and perform joint modeling through federated learning. S1 specifically includes the following steps: S11. Obtain DPI log data for the domain specified by the partner; S12. Analyze the DPI log data of the specified domain name to obtain ur1 and mdn; S13. According to the negotiation rules, analyze ur1 to obtain userID; S14. Establish the mapping relationship between userID and mdn.

2. The customer profiling method based on federated learning as described in claim 1, characterized in that: The domain name specified in S11 is abc.com, and the negotiation rule in S13 is mu&userid=.

3. The customer profiling method based on federated learning as described in claim 1, characterized in that, S2 specifically includes the following steps: S21. Perform geohash encoding on the specified POI location information, and adjust the encoding bit length according to the precision requirements; S22. Obtain the MDN that has accessed the specified physical range through signaling data; S23. Obtain the userID corresponding to S22mdn according to the mapping relationship of S1.

4. The customer profiling method based on federated learning as described in claim 3, characterized in that, If the encoding in S22 is 6 bits, then the specified physical range is: grid width 1.22km, grid height 0.61km.

5. The customer profiling method based on federated learning as described in claim 1, characterized in that, S3 specifically includes the following steps: S31. Align the encrypted samples of userID in the free sample set and userIDs that have accessed the specified area. S32. Perform longitudinal federated learning joint modeling on the sample set.

6. The customer profiling method based on federated learning as described in claim 5, characterized in that, The sample set in S32 includes: features and tags provided by the partner, and features provided by the operator.

7. The customer profiling method based on federated learning as described in claim 6, characterized in that, The features provided by the partner are: the duration of user use of the app, the number of times the app is used, and the terminal brand. The tags provided by the partner are: whether the user is a high-net-worth, high-spending user.

8. The customer profiling method based on federated learning as described in claim 6, characterized in that, The features provided by the operator include: user gender, age, network duration, package amount, stability of permanent residence, and internet usage behavior.

9. A customer profiling device based on federated learning, characterized in that, It includes a mapping module, a position encoding module, and a federated learning module, among which: The mapping module is used to acquire and analyze DPI log data to obtain ur1, mdn, and userID, and establish a mapping relationship between userID and mdn. Specifically, it includes: Obtain DPI log data for the specified domain of the partner; Analyze the DPI log data of the specified domain name to obtain ur1 and mdn; According to the negotiation rules, ur1 is analyzed to obtain the userID; Establish a mapping relationship between userID and MDN; The location encoding module is used to encode POI location information and adjust the encoding bit length; the federated learning module is used to encrypt and align the sample set and perform joint modeling through federated learning.

Images