Telecom fraud detection model training method and device based on blockchain and federated learning

Abstract

The application discloses a telecommunication fraud detection model training method and device based on a blockchain and federated learning, and comprises the following steps: obtaining a model training task published by a model demander, determining a task participant and a federal committee according to the model training task, and distributing corresponding first public keys and private key secret shares to the task participant and the federal committee; downloading a current global model from a consortium chain by the task participant, training the current global model according to a local data set, obtaining a local partial model, encrypting the local partial model according to the first public key to obtain an encrypted partial model, and uploading the encrypted partial model to the consortium chain; downloading the encrypted partial model from the consortium chain by the federal committee, aggregating the encrypted partial model to obtain an encrypted global model, decrypting the encrypted global model according to the private key secret share to obtain an optimized current global model, and uploading the optimized current global model to the consortium chain. The application improves the security and reliability of the telecommunication fraud detection model training and can be applied to the technical field of communication.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method and apparatus for training a telecommunications fraud detection model based on blockchain and federated learning. Background Technology

[0002] With the increasing intelligence of modern networks, traditional criminal activities are increasingly moving into cyberspace. Non-contact crimes utilizing communication and network technologies, such as telecommunications fraud, have become a hot and challenging issue of social concern. Researching and combating telecommunications fraud methods has become a necessary and urgent need. The telecommunications industry generates massive amounts of data from users and devices during its business development. These data resources are rich and diverse, including high-value data such as blacklists, malicious websites, user tag libraries, call data, internet access data, and network experience data. Sharing and collaborating on anti-telecom fraud detection models using this data would greatly help combat fraud. However, in real-world applications, on the one hand, due to industry competition and conflicts of interest, telecommunications operator data largely remains in information silos; on the other hand, because telecommunications operator data involves user privacy and security issues, sharing and collaborating on anti-fraud governance requires careful consideration of data privacy protection and security.

[0003] Federated learning applies distributed computing principles to machine learning, where multiple participants collaborate to train a model with the assistance of a central server. Its main advantage is that all participants retain their training data locally, and they do not directly exchange local datasets during training; they only exchange gradient information from the model during training. This results in a model that closely resembles the performance of centralized machine learning. However, because federated learning uses a central server for model aggregation, it is susceptible to malicious manipulation of gradient information during the aggregation process, leading to risks such as easy information tampering and data privacy breaches. Summary of the Invention

[0004] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0005] Therefore, one objective of this invention is to provide a method for training a telecommunications fraud detection model based on blockchain and federated learning, which improves the security and reliability of training the telecommunications fraud detection model.

[0006] Another objective of this invention is to provide a training device for a telecommunications fraud detection model based on blockchain and federated learning.

[0007] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of the present invention include:

[0008] On one hand, embodiments of the present invention provide a method for training a telecommunications fraud detection model based on blockchain and federated learning, including the following steps:

[0009] Obtain the model training task published by the model requester, determine the task participants and the Federal Committee based on the model training task, and allocate corresponding first public key and private key secret shares to the task participants and the Federal Committee.

[0010] The task participants download the current global model from the consortium blockchain, train the current global model based on the local dataset to obtain a local model, encrypt the local model based on the first public key to obtain an encrypted local model, and upload the encrypted local model to the consortium blockchain.

[0011] The Federal Council downloads the encrypted local model from the consortium blockchain, aggregates the encrypted local model to obtain the encrypted global model, and then decrypts the encrypted global model according to the private key secret share to obtain the optimized current global model, and uploads the optimized current global model to the consortium blockchain.

[0012] Furthermore, in one embodiment of the present invention, the telecommunications fraud detection model training method further includes a step of registering and joining the consortium blockchain, the specific steps of which are as follows:

[0013] Receive a user's consortium blockchain registration request, the consortium blockchain registration request including the user's identity identifier and current timestamp;

[0014] The identity identifier and the current timestamp are verified through the consortium blockchain;

[0015] If the verification is successful, the user is confirmed as a member of the consortium blockchain, and a second public key and a second private key are assigned to the user.

[0016] Furthermore, in one embodiment of the present invention, the step of obtaining the model training task published by the model requester, determining the task participants and the Federal Committee based on the model training task, and allocating corresponding first public key and private key secret shares to the task participants and the Federal Committee specifically includes:

[0017] Obtain the model training task published by the model requester, the model training task including an initial global model and task incentives;

[0018] According to the model training task, multiple task participants are determined from the members of the consortium blockchain, and multiple task participants are selected to form the federation committee based on reputation value, thereby generating a homomorphic key, the homomorphic key including the first public key and the corresponding first private key;

[0019] The first private key is divided into multiple private key secret shares according to the number of task participants, and the first public key and the private key secret shares are distributed to the corresponding task participants.

[0020] Furthermore, in one embodiment of the present invention, the step of training the current global model based on the local dataset to obtain a local model specifically includes:

[0021] The local dataset is divided into a training set and a test set;

[0022] Based on the training set, the current global model is trained using the gradient boosting tree algorithm to obtain the current local model;

[0023] The current local model is tested based on the test set to obtain the local model.

[0024] Furthermore, in one embodiment of the present invention, the step of encrypting the local local model using the first public key to obtain an encrypted local model, and then uploading the encrypted local model to the consortium blockchain, specifically includes:

[0025] The local model is encrypted using the first public key to obtain the first local model;

[0026] The first local model is signed based on the identity identifier, the local training duration, and the second private key to obtain the encrypted local model;

[0027] The encrypted local model is uploaded to the consortium blockchain, so that the consortium blockchain verifies the encrypted local model based on the second public key, and receives the encrypted local model when the verification is successful.

[0028] Furthermore, in one embodiment of the present invention, the step of aggregating the encrypted local model to obtain an encrypted global model, and then decrypting the encrypted global model according to the private key secret share to obtain the optimized current global model, specifically includes:

[0029] A member is randomly selected from the Federal Committee to serve as the leader of the Federal Committee. The leader of the Federal Committee then performs aggregate calculations on multiple local cryptographic models to obtain the global cryptographic model.

[0030] The encrypted global model is aggregated and verified by all members of the Federal Committee.

[0031] When the aggregation verification is successful, the corresponding decryption share is calculated based on the secret share of the private key. Then, all members of the Federal Committee collaboratively decrypt the encrypted global model based on the corresponding decryption share to obtain the optimized current global model.

[0032] Furthermore, in one embodiment of the present invention, the telecommunications fraud detection model training method further includes the following steps:

[0033] The reward value and reputation value incentives for each task participant are calculated according to the preset incentive mechanism;

[0034] Each task participant is rewarded according to the reward value, and the reputation value of each task participant is updated based on the reputation value incentive.

[0035] On the other hand, embodiments of the present invention provide a training device for a telecommunications fraud detection model based on blockchain and federated learning, comprising:

[0036] The task publishing module is used to obtain model training tasks published by model requesters, determine task participants and the Federal Committee based on the model training tasks, and allocate corresponding first public key and private key secret shares to the task participants and the Federal Committee.

[0037] The local model training module is used to download the current global model from the consortium blockchain through the task participants, train the current global model according to the local dataset to obtain a local local model, encrypt the local local model according to the first public key to obtain an encrypted local model, and upload the encrypted local model to the consortium blockchain.

[0038] The global model aggregation module is used to download the encrypted local model from the consortium blockchain through the Federal Council, aggregate the encrypted local model to obtain the encrypted global model, decrypt the encrypted global model according to the private key secret share to obtain the optimized current global model, and upload the optimized current global model to the consortium blockchain.

[0039] On the other hand, embodiments of the present invention provide an electronic device, which includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for enabling communication between the processor and the memory. When the program is executed by the processor, it implements the training method for a telecommunications fraud detection model based on blockchain and federated learning as described above.

[0040] On the other hand, embodiments of the present invention also provide a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the training method for the telecommunications fraud detection model based on blockchain and federated learning as described above.

[0041] The advantages and beneficial effects of the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention:

[0042] This invention obtains model training tasks published by model requesters, determines task participants and a federation committee based on the training tasks, and allocates corresponding first public key and private key secret shares to the task participants and the federation committee. The task participants download the current global model from the consortium blockchain, train the current global model based on the local dataset to obtain a local model, encrypt the local model using the first public key to obtain an encrypted local model, and upload the encrypted local model to the consortium blockchain. The federation committee downloads the encrypted local model from the consortium blockchain, aggregates the encrypted local models to obtain an encrypted global model, and decrypts the encrypted global model using the private key secret share to obtain the optimized current global model, which is then uploaded to the consortium blockchain. This invention combines blockchain technology with federated learning and cryptographic mechanisms. Because blockchain possesses characteristics such as decentralization, tamper-proofing, and traceability, it can effectively complement federated learning, providing privacy protection for collaborative training. This effectively addresses security issues such as the single data source for training telecom fraud detection models and the ease with which data privacy can be leaked. It facilitates multi-party collaboration among telecom operators, public security departments, financial institutions, internet companies, and other organizations in a decentralized environment, enabling secure and reliable joint training of telecom fraud detection models. This allows for the value circulation of telecom operator data, improves the security and reliability of telecom fraud detection model training, effectively protects the data privacy of all participating parties, and achieves efficient detection of telecom fraud. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments of the present invention are described below. It should be understood that the drawings described below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart illustrating the steps of a method for training a telecommunications fraud detection model based on blockchain and federated learning, as provided in an embodiment of the present invention.

[0045] Figure 2 A schematic diagram illustrating the multi-party interaction of a training method for a telecommunications fraud detection model based on blockchain and federated learning, provided in an embodiment of the present invention.

[0046] Figure 3 A timing diagram illustrating the training method for a telecommunications fraud detection model based on blockchain and federated learning provided in an embodiment of the present invention;

[0047] Figure 4 A flowchart illustrating the steps of registering and joining a consortium blockchain, as provided in an embodiment of the present invention;

[0048] Figure 5 A flowchart of step S101 provided in an embodiment of the present invention;

[0049] Figure 6 A flowchart of step S102 provided in an embodiment of the present invention;

[0050] Figure 7 Another flowchart of step S102 provided in an embodiment of the present invention;

[0051] Figure 8 A flowchart of step S103 provided in an embodiment of the present invention;

[0052] Figure 9 Another flowchart illustrating the steps of the training method for a telecommunications fraud detection model based on blockchain and federated learning provided in this embodiment of the invention;

[0053] Figure 10 A schematic diagram of the structure of a training device for a telecommunications fraud detection model based on blockchain and federated learning provided in an embodiment of the present invention;

[0054] Figure 11 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention;

[0055] Figure 12 This is a schematic diagram of the structure of the storage medium provided in an embodiment of the present invention. Detailed Implementation

[0056] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. It should be noted that although functional modules are divided in the system schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system schematic diagram or the order in the flowchart. The step numbers in the following embodiments are only set for ease of explanation and do not limit the order between steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0057] In the description of this invention, "multiple" means two or more. The use of "first" and "second" is for distinguishing technical features only and should not be construed as indicating or implying relative importance, the number of indicated technical features, or the order of the indicated technical features. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0058] The blockchain- and federated learning-based telecom fraud detection model training method provided in this application can be applied to terminals, servers, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, set-top box, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the blockchain- and federated learning-based telecom fraud detection model training method, but is not limited to the above forms.

[0059] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0060] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0061] The implementation process of this invention includes four main entities: the Telecom Fraud Detection Model Requester (MD), the Task Participant (TP), the Consortium Blockchain (CBC), and the Federal Council (FC).

[0062] Model Demander (MD): Responsible for publishing training requirements for telecom fraud detection models. The Model Demander first selects a machine learning model and initializes the selected model; then organizes participants to join the telecom fraud detection model training team. The Model Demander will eventually obtain an efficient telecom fraud detection model.

[0063] Task Participant (TP): Participants possess some of the data needed to train the telecom fraud detection model. These can be organizations such as telecom operators, public security departments, and financial institutions. Public security departments obtain customer data related to telecom fraud during case investigations, telecom operators possess customer communication data, and financial institutions accumulate customer behavioral data in the financial sector. However, due to limited data resources and security issues such as data privacy protection, they cannot independently complete the entire telecom fraud training task effectively, or models trained using only a single data resource may perform poorly.

[0064] Consortium Blockchain (CBC): A consortium blockchain is formed by selecting organizations such as telecommunications operators, public security departments, and financial institutions as consensus nodes. Consortium blockchains have an admission mechanism; the system must verify the qualifications of nodes joining the consortium blockchain, and only users who pass the verification can join, thus preventing attacks and damage from malicious nodes.

[0065] The Federal Council (FC) is a group of participants with high reputation scores who are responsible for aggregating intermediate parameters of the telecom fraud detection model, collaboratively decrypting them, and allocating incentives.

[0066] like Figure 1 The diagram shows a flowchart of a training method for a telecommunications fraud detection model based on blockchain and federated learning, provided in an embodiment of the present invention. (Refer to...) Figure 1 This invention provides a method for training a telecommunications fraud detection model based on blockchain and federated learning, specifically including the following steps:

[0067] S101. Obtain the model training task published by the model requester, determine the task participants and the Federal Committee based on the model training task, and allocate the corresponding first public key and private key secret share to the task participants and the Federal Committee.

[0068] S102. The task participants download the current global model from the consortium blockchain, train the current global model based on the local dataset to obtain the local model, encrypt the local model based on the first public key to obtain the encrypted local model, and upload the encrypted local model to the consortium blockchain.

[0069] S103. The Federal Council downloads the encrypted local model from the consortium blockchain, aggregates the encrypted local models to obtain the encrypted global model, and then decrypts the encrypted global model according to the secret share of the private key to obtain the optimized current global model, and uploads the optimized current global model to the consortium blockchain.

[0070] Repeat steps S102 to S103 as described above until the model converges or reaches the preset accuracy. Finally, the user downloads the final telecom fraud detection model from the consortium blockchain.

[0071] Specifically, such as Figure 2 The diagram shown illustrates the multi-party interaction of a training method for a telecommunications fraud detection model based on blockchain and federated learning, as provided in an embodiment of the present invention.

[0072] The MD (Model Requester) for a telecom fraud detection model publishes its training requirements, selects a machine learning model, uploads its initial parameters to the consortium blockchain, and contributes a sum of tokens as a reward pool. Interested participants with training datasets can apply to join the federated training task. As more participants join the model training team, the MD will eventually obtain a relatively efficient telecom fraud detection model.

[0073] Task Participants (TPs): These participants possess the data required for the federated learning task of combating telecom fraud. They can be organizations such as telecom operators, public security departments, and financial institutions. Public security departments obtain customer data related to telecom fraud during investigations; telecom operators possess customer communication data; and financial institutions accumulate customer behavioral data in the financial sector. However, due to limited data resources and security issues such as data privacy protection, they cannot independently complete the entire training task, or the model trained using a single data resource performs poorly. Participants can apply to join the federated training team. After approval by the requester, they become federated learning members. When uploading local model parameters to the consortium blockchain, participants must also submit information such as the size of their local training dataset and training time. The system uses this information as a basis for evaluating the participant's contribution.

[0074] Consortium Blockchain (CBC) comprises organizations such as Telecom Operator A, Telecom Operator B, Telecom Operator C, Public Security Bureau, Bank A, Bank B, and Bank C as consensus nodes. Because of its access control mechanism, the system verifies the qualifications of nodes joining the consortium blockchain; only users who pass the verification can join, thus preventing attacks and damage from malicious nodes. The blockchain ledger is jointly maintained by every user in the system, meaning that the failure of a single node will not affect the operation of the entire system, avoiding the impact of single points of failure. Furthermore, due to the open, traceable, and non-repudiable nature of blockchain information, the system stores the historical incentive and evaluation information of all participants on the consortium blockchain, making model training more transparent, reliable, and fair.

[0075] The Federation Council (FC) is a group of participants randomly selected from the federated learning team to form a FC. A member is randomly chosen as the leader for the training task. The FC downloads the local model parameters from each participant on the consortium blockchain, aggregates them, and then collaboratively decrypts the aggregated model before uploading it to the blockchain. When the FC leader goes offline or the system-defined term expires, the system re-ranks the participants based on their reputation scores and elects a new FC and leader. The FC's main tasks are model parameter aggregation, collaborative decryption, and incentive allocation.

[0076] It can be recognized that the embodiments of the present invention combine blockchain technology with federated learning and cryptographic mechanisms. Because blockchain has the characteristics of decentralization, tamper-proof, and traceability, it can complement federated learning well, provide privacy protection for collaborative training, and effectively solve the security problems such as the single data source and easy leakage of data privacy in the training of telecom fraud detection models. It is conducive to multi-party collaboration between telecom operators, public security departments, financial institutions, Internet companies and other organizations in a decentralized environment, to conduct joint training of telecom fraud detection models in a secure and reliable manner, realize the value circulation of telecom operator data, improve the security and reliability of telecom fraud detection model training, effectively protect the data privacy of all parties, and achieve efficient detection of telecom fraud.

[0077] like Figure 4 The diagram shown is a flowchart of one step in registering and joining a consortium blockchain according to an embodiment of the present invention. Figure 4 As an optional implementation, the telecom fraud detection model training method also includes a registration and joining of a consortium blockchain, the specific steps of which are as follows:

[0078] S201. Receive the user's consortium blockchain registration request. The consortium blockchain registration request includes the user's identity identifier and current timestamp.

[0079] S202. Verify the identity identifier and current timestamp through the consortium blockchain;

[0080] S203. When the verification is successful, the user is confirmed as a member of the consortium blockchain, and a second public key and a second private key are assigned to the user.

[0081] Specifically, assume there are N users in the system, referred to as u1, u2, L, u... N This indicates that user u i For each i ∈ [1, N], select an identity identifier (UID). i Get the system's current timestamp. i Then {UID i ,ut iThe request is sent to the consortium blockchain for registration. When the consortium blockchain receives the user's registration request, it verifies the data. i and UID i The system verifies the validity of the public-private key pair (PK). If valid, the registration request is accepted, and a public-private key pair (PK) is assigned to the user. i ,sk i Otherwise, the user's registration request will be rejected.

[0082] like Figure 5 The diagram shown is a flowchart of step S101 provided in an embodiment of the present invention. (Refer to...) Figure 5 As an optional implementation, the model training task published by the model requester is obtained, the task participants and the Federal Committee are determined based on the model training task, and corresponding first public key and private key secret shares are allocated to the task participants and the Federal Committee, specifically including:

[0083] S1011. Obtain the model training task published by the model requester. The model training task includes the initial global model and task incentives.

[0084] S1012. Based on the model training task, determine multiple task participants from the members of the consortium blockchain, and select multiple task participants to form a federation committee based on reputation value, thereby generating a homomorphic key, which includes a first public key and a corresponding first private key.

[0085] S1013. Divide the first private key into multiple private key secret shares according to the number of task participants, and distribute the first public key and private key secret shares to the corresponding task participants.

[0086] Specifically, a user requesting a telecom fraud detection model publishes a model learning task, selecting to initialize the model parameters Δw. 0 and pay a certain amount of tokens b T The data is uploaded to the consortium blockchain as a task incentive. Participants apply to join the training task based on their own data resources and pay a certain amount of tokens (b). u As a deposit, a certain percentage of tokens will be deducted as punishment if a user makes a mistake.

[0087] Assume the model user selects k task participants u1, u2, ..., u k A federated learning team is formed, and the model requester selects M = k / 2 members from the federated learning team to form a federated committee, and generates a homomorphic key (pk). m ,sk m Then use the private key sk. m The task is divided into k secret shares and shared among the participants, with the set being S = {s1, s2, ..., s...}. k}, each participant ui Will obtain public key pk m and secret shares i .

[0088] In some optional embodiments, the homomorphic key is generated as follows:

[0089] Randomly select two large prime numbers p and q, and let n = pq, where p = 2p' + 1 and q = 2q' + 1. Let m = p'q', choose any Then the private key sk m =βm, randomly selected Let g = (1 + n) a b n modn 2 θ = amβmodn, generate the public key.

[0090] In some optional embodiments, the method for dividing the shared secret share is as follows:

[0091] Let a0 = βm, a i Randomly select from {0,1,L,nm-1}, and set sk m The shares are divided into k secret shares for sharing, and the formula for calculating the secret shares is as follows:

[0092]

[0093] Then s i Distributed to the corresponding participants u i Each participant u i Will obtain public key pk m and secret shares i Once distribution is complete, the private key sk can be... m destroy.

[0094] In some optional embodiments, the formation of the Federal Committee and the election of the Federal Learning Leader are as follows: Participants are ranked from highest to lowest based on their reputation scores, and the top half of the participants with the highest reputation scores are selected to form the Federal Committee. Initially, the Federal Committee leader is randomly selected from among the Federal Committee members. If the leader fails or withdraws, the next Federal Learning Leader will be selected through a round-robin process from the Federal Committee.

[0095] like Figure 6 The diagram shown is a flowchart of step S102 provided in an embodiment of the present invention. (Refer to...) Figure 6 As an optional implementation, the current global model is trained based on the local dataset to obtain a local model, which specifically includes:

[0096] S1021. Divide the local dataset into a training set and a test set;

[0097] S1022. Based on the training set, train the current global model using the gradient boosting tree algorithm to obtain the current local model;

[0098] S1023. Test the current local model based on the test set to obtain the local model.

[0099] Specifically, the gradient boosting tree machine learning algorithm is chosen to train the local model. The gradient boosting tree algorithm uses one or more of the following metrics to evaluate the model's performance: F1 score, accuracy, recall, and AUC. The participants' local data is divided into training and test sets in a 3:7 ratio.

[0100] In some optional embodiments, the LightGBM gradient boosting tree algorithm can be used to train the local model. The core parameters of the LightGBM algorithm are set as: (boosting_type, max_depth, num_leaves, min_child_samples, feature_fraction, bagging_fraction, min_child_weight, bagging_freq), where boosting_type is fixed at gbdt, max_depth = 8, and the other parameters are selected according to the actual situation, while satisfying the condition num_leaves ≤ 2. max_depth During the training process, parameters are continuously optimized and iterated.

[0101] In some optional embodiments, the XGBoost gradient boosting tree algorithm can also be used to train the local model. The core parameters of the XGBoost algorithm are set as follows: (objective, booster, eval_metric, learning_rate, max_depth, subsample, colsample_bytree, reg_alpha, reg_lambda), where the objective parameter is set to binary:logistic, the booster parameter is set to gbtree, the max_depth parameter is 8, and the other parameters are selected according to the actual situation. The parameters are continuously optimized and iterated during the training process.

[0102] In some optional embodiments, the CatBoost gradient boosting tree algorithm can also be used to train the local model. The core parameters of the CatBoost algorithm are set as follows: (bootstrap_type, learning_rate, depth, l2_leaf_reg, subsample, early_stopping_rounds, od_type), where the parameter bootstrap_type is set to Bernoulli, the parameter depth = 8, and the other parameters are selected according to the actual situation. The parameters are continuously optimized and iterated during the training process.

[0103] like Figure 7 The diagram shown is another flowchart of step S102 provided in an embodiment of the present invention. (Refer to...) Figure 7 As an optional implementation, the local model is further encrypted using the first public key to obtain an encrypted local model, and then uploaded to the consortium blockchain. This specifically includes:

[0104] S1024. Encrypt the local model using the first public key to obtain the first local model;

[0105] S1025. Sign the first local model according to the identity identifier, local training duration and second private key to obtain the encrypted local model;

[0106] S1026. Upload the encrypted local model to the consortium blockchain, so that the consortium blockchain can verify the encrypted local model based on the second public key, and receive the encrypted local model when the verification is successful.

[0107] Specifically, task participant u i Download the global model parameters Δw from the consortium blockchain for round T. T This is then used as the input model for the next round of local training. Task participant u i The model is trained using a local dataset, and the intermediate gradient parameters of the local model are obtained after training. Random selection Encryption yields ciphertext Then, the signature is calculated using the private key. And Uploaded to the consortium blockchain. When the consortium blockchain receives a notification from participant u... i The uploaded message first verifies the participant's identity identifier (UID). i and local training duration If all are valid, the validity of the signature information is verified by the following equation;

[0108]

[0109] If the equation is true, the upload request is accepted; otherwise, the upload is rejected.

[0110] like Figure 8 The diagram shown is a flowchart of step S103 provided in an embodiment of the present invention. (Refer to...) Figure 8 As an optional implementation, the encrypted local model is aggregated to obtain an encrypted global model, and then the encrypted global model is decrypted according to the private key secret share to obtain the optimized current global model, which specifically includes:

[0111] S1031. Randomly select a member from the Federal Council to serve as the leader of the Federal Council. The leader of the Federal Council performs aggregate calculations on multiple local encryption models to obtain the global encryption model.

[0112] S1032. The global encryption model is aggregated and verified by all members of the Federal Council;

[0113] S1033. When the aggregation verification is successful, the corresponding decryption share is calculated based on the secret share of the private key. Then, all members of the Federal Committee collaboratively decrypt the encrypted global model based on the corresponding decryption share to obtain the optimized current global model.

[0114] Specifically, the Federation Council leader downloads the encrypted local model parameters of all participants from the consortium blockchain, and then uses an incentive mechanism to calculate participant u. i Contribution rewards, if u i Those who misreport the amount of data will have a certain amount of their deposit frozen as a penalty. After calculating the data contribution rewards for all participants, based on the Paillier algorithm's addition and scalar multiplication homomorphic properties, the leader will aggregate the local model parameters to form the global model parameter ciphertext C(Δw). T+1 The system will allocate a certain amount of computing contribution reward to the leader.

[0115] The leaders of the Federal Council, based on the additive homomorphism of homomorphic encryption algorithms, ciphertexted the parameters of a local fraud detection model for k task participants. Perform aggregation calculations to generate global model parameter ciphertext. Each member of the Federal Council calculated C'(Δw) separately. T+1 When at least M members have calculated C'(Δw) T+1 )=C(Δw T+1 The aggregation result was verified.

[0116] Then, the members of the Federal Council used their respective secret shares... j Calculate decryption share When at least t committee members provide the correct decryption share, the aggregation yields the global plaintext model M(Δw).T ), and upload it to the consortium blockchain.

[0117] Assuming the number of members in the Federal Council is t, if the effective decryption share received by each member of the Federal Council is c... i If there are fewer than t shares, joint decryption is not possible; if the number of valid shares received is not less than t, let S′ be the set of t valid shares, and the plaintext can be calculated as:

[0118]

[0119] in

[0120] like Figure 9 The diagram shown illustrates another step of the training method for a telecommunications fraud detection model based on blockchain and federated learning provided in this embodiment of the invention. (Refer to...) Figure 9 As an optional implementation, the telecom fraud detection model training method further includes the following steps:

[0121] S104. Calculate the reward value and reputation value incentives for each task participant according to the preset incentive mechanism;

[0122] S105. Reward each task participant based on the reward value, and update the reputation value of each task participant based on the reputation value incentive.

[0123] Specifically, this embodiment of the invention attracts participants to actively participate in the collaborative training process through an incentive mechanism, which consists of two parts: reputation value incentive and reward value.

[0124] The incentive mechanism's first part, the calculation method for the reputation value incentive obtained by participants upon completing the training task, is as follows:

[0125]

[0126] Where i represents the i-th task participant, and j represents the j-th training task. This represents the reputation reward earned by participant i for completing the j-th training task; the weight η reflects the participant's data contribution during local training. This indicates the task duration (training time required for participants with a large local dataset). The training dataset size is relatively long, and is positively correlated with the training duration. ρ(Δw) j ) represents the accuracy of the global model obtained by all participants in the j-th training iteration, rpt i represents the total reputation reward for the i-th task participant, initially set to 0; m represents the total number of training sessions for the i-th task participant.

[0127] The second part of the incentive mechanism, the method for calculating the reward value obtained by the task participant upon completing the training task, is as follows:

[0128]

[0129] in Let b represent the reward value for the i-th model participant completing the j-th task. j δ represents the base reward for the j-th task. i RWD represents the reputation reward coefficient of the i-th task participant; n represents the number of participants in the collaborative training task; and RWD represents the total reward value of the current task set by the system.

[0130] The method flow of the embodiments of the present invention has been described above. For example... Figure 3 The figure shown is a timing diagram of the training method for a telecommunications fraud detection model based on blockchain and federated learning provided in an embodiment of the present invention. The following is a combination of... Figure 3 The implementation process of the embodiments of the present invention will be further described.

[0131] 1) User Registration. Users apply to join the consortium blockchain. The consortium blockchain will verify the user's qualifications. Users who pass the verification will receive the user's public and private keys (PK). i ,sk i The system successfully registered and joined the system, avoiding attacks and damage from malicious nodes. Let the consortium blockchain system contain a total of N users, namely u1, u2, L, u... N Users in a consortium blockchain mainly include four types: MD (Demand Provider), TP (Task Participant), CBC (Consortium Blockchain), and FC (Federal Council).

[0132] 2) Construction of Telecom Fraud Training Task. The user (MD) requesting the telecom fraud detection model publishes a model training task, selecting to initialize the model parameters Δw. 0 and pay a certain amount of tokens b T The data is uploaded to the consortium blockchain as a task incentive. Task participants (TPs) apply to join the training task based on their own data resources and pay a certain amount of tokens (b). u As a deposit, a certain percentage of tokens will be deducted as punishment if a user makes a mistake.

[0133] The telecom fraud detection model requires k task participants (u1, u2, ..., u) to be selected by the MD user. k A federated learning team is formed, and the MD selects the M from the federated learning team based on the TP's reputation score (where...). A Federal Committee (FC) is composed of 10 members and generates a homomorphic key (pk). m ,sk m Then use the private key sk. mDivide the data into k secret shares and share them with TP, i.e., S = {s1, s2, ..., s...} k}, each participant u i Will obtain public key pk m and secret shares i .

[0134] 3) Localized telecom fraud detection model on-chain. Task participant u i Download the global model parameters Δw from the T-th round of training on the consortium blockchain CBC. T And use it as the input model for the next round of local training. i The model is trained using a local dataset, and the intermediate gradient parameters of the local fraud detection model are obtained after training. Encrypting the gradient parameters of the local model yields ciphertext. Then calculate the signature. And Uploaded to the consortium blockchain. When the consortium blockchain receives a notification from participant u... i The uploaded message first verifies the participant's identity identifier (UID). i and local training duration Is it valid? If valid, verify the validity of the signature information using the following equation:

[0135]

[0136] If the equation is true, the upload request is accepted; otherwise, the upload is rejected.

[0137] 4) Global Model Aggregation. The Federation Council leader downloads the encrypted local model parameters of all participants from the consortium blockchain, and then uses the incentive mechanism to calculate participant u. i Contribution rewards, if u i Those who misreport the amount of data will have a certain amount of their deposit frozen as a penalty. After calculating the data contribution rewards for all participants, the leader will aggregate the local model parameters to form the global model parameter ciphertext C(Δw). T+1 The system will allocate a certain amount of computational contribution reward to the leader. Each member of the Federal Committee calculates C'(Δw) separately. T+1 C'(Δw) is calculated if and only if at least M members are available. T+1 )=C(Δw T+1 The aggregation result verification passed. Then, members of the Federal Council used their held secret shares... j Calculate the decryption share c j The global plaintext model M(Δw) is obtained when at least t committee members provide the correct decryption share. T ), and upload it to the consortium blockchain CBC.

[0138] 5) Repeat steps 3) to 4) above until the model converges or reaches the preset accuracy. Finally, the model requester (MD) downloads the final telecom fraud detection model from the consortium blockchain.

[0139] It can be recognized that this invention utilizes federated learning to jointly model and train multi-source fraud data, and employs blockchain to replace the central server, endowing federated learning with tamper-proof and single-point-of-failure resistance characteristics, thereby enabling management of the training process. Collaborative training participants use local datasets to train the telecom fraud detection model. Privacy data from telecom operators, public security departments, financial institutions, etc., does not need to leave the local database. Homomorphic encryption and secret-sharing algorithms protect the privacy of intermediate model parameters for training participants. Model aggregation and collaborative decryption are performed through a federated committee, providing privacy protection for collaborative training. An incentive mechanism is established to ensure the fairness of collaborative training, encouraging task participants to join the federated learning collaborative training to improve the accuracy of the telecom fraud detection model training. Combining blockchain technology with federated learning and cryptographic mechanisms can form a complementary relationship, facilitating multi-party collaboration among telecom operators, public security departments, financial institutions, internet companies, and other organizations in a decentralized environment. This enables secure and reliable joint anti-fraud model training, realizing the value circulation of telecom operator data, effectively protecting the data privacy of all participants while achieving effective detection of telecom fraud.

[0140] like Figure 10 The diagram shown is a structural schematic of a training device for a telecommunications fraud detection model based on blockchain and federated learning provided in an embodiment of the present invention. (Refer to...) Figure 10 This invention provides a training device for a telecommunications fraud detection model based on blockchain and federated learning, comprising:

[0141] The task publishing module is used to obtain model training tasks published by model requesters, determine task participants and the Federal Committee based on the model training tasks, and allocate corresponding first public key and private key secret shares to task participants and the Federal Committee.

[0142] The local model training module is used to download the current global model from the consortium blockchain through task participants, train the current global model based on the local dataset to obtain a local model, encrypt the local model based on the first public key to obtain an encrypted local model, and then upload the encrypted local model to the consortium blockchain.

[0143] The global model aggregation module is used to download encrypted local models from the consortium blockchain through the Federal Council, aggregate the encrypted local models to obtain an encrypted global model, and then decrypt the encrypted global model according to the secret share of the private key to obtain the optimized current global model, and then upload the optimized current global model to the consortium blockchain.

[0144] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0145] This invention also provides an electronic device, comprising: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the aforementioned training method for a telecommunications fraud detection model based on blockchain and federated learning. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0146] like Figure 11 The diagram shown is a hardware structure schematic of an electronic device provided in an embodiment of the present invention. (Refer to...) Figure 11 This invention provides an electronic device, comprising:

[0147] The processor 1101 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0148] The memory 1102 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1102 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1102 and is called and executed by the processor 1101 to execute the training method for the telecommunications fraud detection model based on blockchain and federated learning in the embodiments of this invention.

[0149] Input / output interface 1103 is used to implement information input and output;

[0150] The communication interface 1104 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0151] Bus 1105 transmits information between various components of the device (e.g., processor 1101, memory 1102, input / output interface 1103, and communication interface 1104);

[0152] The processor 1101, memory 1102, input / output interface 1103 and communication interface 1104 are connected to each other within the device via bus 1105.

[0153] like Figure 12 The diagram shown is a structural schematic of the storage medium provided in an embodiment of the present invention. (Refer to...) Figure 12 The present invention also provides a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs 1201, which can be executed by one or more processors to implement the above-mentioned training method for the telecommunications fraud detection model based on blockchain and federated learning.

[0154] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0155] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0156] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the aforementioned blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0157] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0158] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0160] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the aforementioned program can be printed, because the aforementioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or, if necessary, processing in other suitable ways, and then stored in computer memory.

[0161] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0162] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0163] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0164] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for training a telecom fraud detection model based on blockchain and federated learning, characterized in that, The method comprises the following steps: obtaining a model training task published by a model demander, determining a task participant and a federal committee according to the model training task, and assigning a corresponding first public key and private key secret share to the task participant and the federal committee; downloading a current global model from a consortium chain by the task participant, training the current global model according to a local data set to obtain a local partial model, then encrypting the local partial model according to the first public key to obtain an encrypted partial model, and uploading the encrypted partial model to the consortium chain; downloading the encrypted partial model from the consortium chain by the federal committee, aggregating the encrypted partial model to obtain an encrypted global model, then decrypting the encrypted global model according to the private key secret share to obtain an optimized current global model, and uploading the optimized current global model to the consortium chain; The method for obtaining a model training task published by a model demander, determining a task participant and a federal committee according to the model training task, and assigning a corresponding first public key and private key secret share to the task participant and the federal committee, specifically comprises: obtaining the model training task published by the model demander, wherein the model training task comprises an initial global model and a task incentive; determining a plurality of task participants from members of the consortium chain according to the model training task, selecting a plurality of task participants to form the federal committee according to a credit value, and then generating a homomorphic key, wherein the homomorphic key comprises the first public key and a corresponding first private key; dividing the first private key into a plurality of private key secret shares according to the number of task participants, and distributing the first public key and the private key secret shares to the corresponding task participants; The method for aggregating the encrypted partial model to obtain an encrypted global model, then decrypting the encrypted global model according to the private key secret share to obtain an optimized current global model, specifically comprises: randomly selecting a member from the federal committee as a federal committee leader, aggregating and calculating a plurality of encrypted partial models by the federal committee leader to obtain the encrypted global model; aggregating and verifying the encrypted global model by all members of the federal committee; when the aggregation and verification is successful, calculating a corresponding decryption share according to the private key secret share, then collaboratively decrypting the encrypted global model according to the corresponding decryption share by all members of the federal committee to obtain the optimized current global model.

2. The blockchain and federated learning-based telecommunications fraud detection model training method of claim 1, wherein The method for training a telecom fraud detection model further comprises the step of registering to join the consortium chain, which specifically comprises: receiving a consortium chain registration request of a user, wherein the consortium chain registration request comprises an identity of the user and a current timestamp; verifying the identity and the current timestamp by the consortium chain; when the verification is successful, determining that the user is a member of the consortium chain, and assigning a second public key and a second private key to the user.

3. The blockchain and federated learning-based telecommunications fraud detection model training method of claim 1, wherein The training of the current global model according to the local data set to obtain a local partial model specifically includes: dividing the local data set into a training set and a test set; training the current global model according to the training set by a gradient boosting tree algorithm to obtain a current partial model; testing the current partial model according to the test set to obtain the local partial model.

4. The blockchain and federated learning-based telecommunications fraud detection model training method of claim 2, wherein, The encryption of the local partial model according to the first public key to obtain an encrypted partial model, and uploading the encrypted partial model to the consortium chain specifically includes: encrypting the local partial model according to the first public key to obtain a first partial model; signing the first partial model according to the identity, the local training duration and the second private key to obtain the encrypted partial model; uploading the encrypted partial model to the consortium chain, so that the consortium chain verifies the encrypted partial model according to the second public key, and receives the encrypted partial model when the verification is passed.

5. The blockchain and federated learning-based telecommunications fraud detection model training method of claim 1, wherein, The telecom fraud detection model training method further includes the following steps: According to the preset incentive mechanism, the reward value and the reputation value of each task participant are calculated; According to the reward value, each task participant is rewarded, and the reputation value of each task participant is updated according to the reputation value.

6. A blockchain and federated learning based telecommunication fraud detection model training apparatus, characterized in that, The method for training a telecom fraud detection model based on blockchain and federated learning according to any one of claims 1 to 5 comprises: a task publishing module configured to obtain a model training task published by a model demander, determine a task participant and a federation committee according to the model training task, and assign corresponding first public keys and private key secret shares to the task participant and the federation committee; a partial model training module configured to download a current global model from a consortium chain by the task participant, train the current global model according to a local data set to obtain a local partial model, and then encrypt the local partial model according to the first public key to obtain an encrypted partial model, and upload the encrypted partial model to the consortium chain; a global model aggregation module configured to download the encrypted partial model from the consortium chain by the federation committee, aggregate the encrypted partial model to obtain an encrypted global model, then decrypt the encrypted global model according to the private key secret share to obtain an optimized current global model, and upload the optimized current global model to the consortium chain.

7. An electronic device, comprising: The electronic device includes a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for connecting and communicating between the processor and the memory, and the program is executed by the processor to implement the steps of the method for training a telecom fraud detection model based on blockchain and federated learning according to any one of claims 1 to 5.

8. A storage medium, the storage medium being a computer-readable storage medium for computer-readable storage, characterized in that, The storage medium stores one or more programs, and the one or more programs are executable by one or more processors to implement the steps of the method for training a telecom fraud detection model based on blockchain and federated learning according to any one of claims 1 to 5.

Images