User operation strategy determination method and device, electronic equipment and storage medium

Abstract

The invention discloses a user operation strategy determination method and device, electronic equipment and a storage medium, and relates to the field of artificial intelligence and the field of financial science and technology. The method comprises the following steps: acquiring target object data; the target object data are input into a pre-trained semi-supervised learning model, feature vectors are extracted from the target object data through the semi-supervised learning model, feature tags of the target objects are predicted according to the feature vectors, and the feature tags are used for marking preference features of the target objects when the target objects communicate with a financial institution; and generating a target operation strategy corresponding to the target object in combination with a preset operation strategy template according to the feature tag of the target object. The technical problem that in the prior art, due to the fact that the user operation strategy is too single, different operation strategies are difficult to generate for different users, and the personalized generation accuracy of the operation strategies is poor is solved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and financial technology, and specifically to a method, device, electronic device and storage medium for determining a user operation strategy. Background Art

[0002] As digital transformation deepens, businesses face an increasingly diverse and complex customer base. To stand out in the fiercely competitive market, accurately understanding and meeting the needs of each customer has become crucial. Traditional customer operations strategies often adopt a one-size-fits-all approach, formulating the same operations plan for all customers or a specific customer category. While this approach is simple to implement, it is insufficient in today's highly personalized and segmented market environment.

[0003] Existing technologies typically rely on expert experience and rule-based approaches to generate customer operations strategies. While effective in certain scenarios, this rule-based approach is primarily limited by its lack of flexibility and adaptability. When customer behavior patterns and preferences change, or when new competitors emerge in the market, fixed rules may not be quickly adjusted, impacting the effectiveness and responsiveness of the strategy. Furthermore, rule-based development is often based on limited data and assumptions, making it difficult to fully capture all customer characteristics and market dynamics. This results in poorly accurate personalized operations strategies.

[0004] In summary, existing technologies often rely too simplistically on data, particularly high-quality labeled data, to generate personalized customer operations strategies, resulting in poor accuracy. This technical issue directly impacts the effectiveness of communication between businesses and their customers, reduces marketing efficiency and customer satisfaction, and becomes a significant bottleneck restricting businesses' customer relationship management and market expansion. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, electronic device and storage medium for determining a user operation strategy, so as to at least solve the technical problem in the prior art that the user operation strategy is too single and it is difficult to generate different operation strategies for different users, resulting in poor accuracy in personalized generation of operation strategies.

[0006] According to one aspect of an embodiment of the present application, a method for determining a user operation strategy is provided, comprising: obtaining target object data with authorization from the target object, wherein the target object data includes: channel preference data of the target object, historical transaction records between the target object and financial institutions, and interactive behavior data; inputting the target object data into a pre-trained semi-supervised learning model, extracting a feature vector from the target object data through the semi-supervised learning model, and predicting a feature label of the target object based on the feature vector, wherein the feature label is used to mark the preference characteristics of the target object when communicating with the financial institution; generating a target operation strategy corresponding to the target object based on the feature label of the target object and a preset operation strategy template.

[0007] Optionally, after generating a target operation strategy corresponding to the target object, the target operation content and the promotion timing and promotion method of the target operation content are determined according to the target operation strategy; the promotion method is used at the promotion timing to send the target operation content to the terminal device of the target object.

[0008] Optionally, after the target operation content is sent to the terminal device of the target object, feedback status information of the target object regarding the target operation content is collected; and the operation strategy template and / or the semi-supervised learning model is updated according to the feedback status information.

[0009] Optionally, with authorization from the target object, the target object data is obtained, including: with authorization from the target object, configuring a data acquisition interface related to the target object; synchronously acquiring initial data of the target object through the data acquisition interface; converting the initial data into intermediate data in a target format; performing a repair operation on the intermediate data to obtain the target object data, wherein the repair operation includes: filling in empty values ​​of the intermediate data, correcting erroneous values ​​of the intermediate data, and correcting the acquisition time corresponding to the intermediate data.

[0010] Optionally, a feature vector is extracted from the target object data through a semi-supervised learning model, and a feature label of the target object is predicted based on the feature vector, including: converting the target object data into the format required by the semi-supervised learning model to obtain model input data; compressing the model input data to obtain compressed data; after receiving the compressed data through each neural network layer of the semi-supervised learning model, the compressed data is decompressed, and a feature vector is extracted from the decompressed data, wherein the feature vector includes at least: the channel preference characteristics, transaction behavior characteristics, and interaction frequency characteristics of the target object; and predicting the feature label of the target object based on the feature vector.

[0011] Optionally, the training steps of the semi-supervised learning model include: obtaining operation record data of the historical reference object; dividing the operation record data of the historical reference object into a labeled data set and an unlabeled data set, wherein each data sample in the labeled data set is accompanied by one or more labels, and the labels are used to indicate the category and attribute information of the data sample; the data samples in the unlabeled data set are not accompanied by any labels; initializing the semi-supervised learning model according to the labeled data set; performing iterative training on the unlabeled data set to generate pseudo labels, and feeding the pseudo labels back to the model training process of the semi-supervised learning model, and using reinforcement learning to optimize the pseudo label generation process; evaluating the performance of the semi-supervised learning model, and optimizing the model based on the evaluation results until the model performance index reaches a preset threshold, and determining that the semi-supervised learning model has ended training.

[0012] Optionally, based on the characteristic label of the target object and in combination with a preset operation strategy template, a target operation strategy corresponding to the target object is generated, including: selecting an operation strategy template that matches the characteristic label from a preset operation strategy template library based on the characteristic label of the target object; performing semantic analysis on the operation strategy template; adjusting the selected operation strategy template based on the analysis results of the semantic analysis and the characteristic label of the target object to obtain an initial operation strategy; and using a generative adversarial network to update the initial operation strategy to obtain a target operation strategy.

[0013] According to another aspect of an embodiment of the present application, a device for determining a user operation strategy is also provided, which includes: an acquisition unit for acquiring target object data with the authorization of the target object, wherein the target object data includes: channel preference data of the target object, historical transaction records between the target object and financial institutions, and interactive behavior data; a first processing unit for inputting the target object data into a pre-trained semi-supervised learning model, extracting a feature vector from the target object data through the semi-supervised learning model, and predicting a feature label of the target object based on the feature vector, wherein the feature label is used to mark the preference characteristics of the target object when communicating with the financial institution; a second processing unit for generating a target operation strategy corresponding to the target object based on the feature label of the target object in combination with a preset operation strategy template.

[0014] According to another aspect of an embodiment of the present application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer-readable storage medium, wherein when the computer program runs, the device where the computer-readable storage medium is located executes the above-mentioned method for determining the user operation strategy.

[0015] According to another aspect of an embodiment of the present application, an electronic device is also provided, wherein the electronic device includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors execute the above-mentioned method for determining the user operation strategy.

[0016] According to another aspect of an embodiment of the present application, a computer program product is also provided, wherein the computer program product includes a computer program or instructions, and the computer program or instructions implement the above-mentioned method for determining the user operation strategy when executed by a processor.

[0017] In this application, with the authorization of the target subject, target subject data is obtained, where the target subject data includes: the target subject's channel preference data, the target subject's historical transaction records with financial institutions, and interactive behavior data. The target subject data is then input into a pre-trained semi-supervised learning model, which extracts a feature vector from the target subject data. Based on the feature vector, the target subject's feature label is predicted, where the feature label is used to identify the target subject's preference characteristics when communicating with financial institutions. Finally, based on the target subject's feature label and combined with a preset operational strategy template, a target operational strategy corresponding to the target subject is generated.

[0018] It can be seen from the above content that according to the technical solution of this application, the collected target object data is input into a pre-trained semi-supervised learning model. By extracting feature vectors from the data through the model, the unique behavior patterns and preference characteristics of the target object can be identified. The semi-supervised learning model predicts the feature labels of the target object based on the extracted feature vectors. These feature labels can carefully reflect the preference characteristics of the target object, especially the specific needs and tendencies when communicating with financial institutions. Based on the predicted feature labels, this application combines the preset operation strategy template to generate a highly personalized operation strategy. This method can not only generate differentiated strategies for different users, but also ensure that the strategy is closely matched with the user's preferences, greatly improving the pertinence and effectiveness of the operation strategy.

[0019] By implementing the above technical solutions, this application can more precisely understand and predict user behavior and preferences, so that the generated operation strategies are more in line with the actual needs of users, thereby improving the success rate of strategy execution and customer satisfaction.

[0020] In summary, the customer operation strategy determination method based on semi-supervised learning proposed in this application effectively solves the problem of poor accuracy of personalized operation strategy generation in the existing technology through steps such as authorized data collection, feature vector extraction, feature label prediction and personalized strategy generation, realizes the refinement and intelligence of the strategy, and significantly improves the efficiency and quality of customer operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 is a flowchart of an optional method for determining a user operation strategy according to an embodiment of the present application;

[0023] Figure 2 This is a flow chart of generating an optional feature tag according to an embodiment of the present application;

[0024] Figure 3 is a training flow chart of an optional semi-supervised learning model according to an embodiment of the present application;

[0025] Figure 4 This is a schematic diagram of an optional device for determining a user operation strategy according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] It should also be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) collected by this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or institution through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or institution.

[0029] According to an embodiment of the present application, an embodiment of a method for determining a user operation strategy is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Optionally, according to an embodiment of the present application, a user operation strategy determination system (hereinafter referred to as the system) is provided as the execution subject of the user operation strategy determination method of the embodiment of the present application, wherein the system can be a software system or an embedded system combining software and hardware. Of course, the method execution subject in the embodiment of the present application can also be other forms of execution subjects, such as devices, equipment, etc. Those skilled in the art should know that this application does not specifically limit the specific form of expression of the method execution subject.

[0031] Figure 1 This is a flow chart of an optional method for determining a user operation strategy according to an embodiment of the present application. Figure 1 As shown, the method includes the following steps:

[0032] Step S101: Acquire target object data when authorization is obtained from the target object.

[0033] In step S101 , the target object data includes: the target object's channel preference data, the target object's historical transaction records with financial institutions, and interaction behavior data.

[0034] It is important to emphasize that obtaining explicit authorization from the target subject includes informing the target subject of the purpose of data collection, how it will be used, and how their privacy will be protected. The target subject is also clearly informed of the privacy policy, including policies for data collection, use, storage, and deletion, as well as how the target subject can exercise their rights, such as accessing, correcting, or deleting their data. The target subject has the right to choose whether to consent to the collection and use of data.

[0035] Furthermore, when collecting subject data, the system uses encryption technology to ensure data security during transmission and storage. Use of subject data will strictly comply with relevant laws and regulations and the scope of subject authorization. Subject data is stored in a secure database and undergoes regular security audits to ensure data security and the effectiveness of privacy protection measures. When subject data is no longer needed, the system securely deletes it according to the subject's requirements and legal requirements, ensuring it cannot be recovered.

[0036] Optionally, channel preference data refers to information about target customers' preferred communication and interaction methods, including but not limited to email, text messages, social media, phone calls, and physical stores. This data can help financial institutions understand which channels customers prefer for communication, allowing them to adopt the most effective channels to reach customers in future operational strategies, improving communication efficiency and customer satisfaction. Historical transaction records cover all financial transactions between the target customer and the financial institution, such as details on deposits, withdrawals, transfers, investments, and loans. Historical transaction records not only reveal customers' financial behavior habits but also reflect their credit status and risk appetite, serving as an important basis for developing personalized financial services and product recommendation strategies.

[0037] Optionally, interactive behavior data includes records of various interactions between customers and financial institutions, such as the frequency of customer service inquiries, interaction channels chosen, content preferences, and responses to financial institution promotions. This type of data is crucial for understanding customers' service needs, emotional attitudes, and reactions to brand promotions, helping financial institutions optimize the customer experience and develop more attractive promotional strategies.

[0038] By legally collecting these three types of data, financial institutions can build a comprehensive and detailed customer profile, providing a rich foundation for subsequent feature vector extraction and feature label prediction using semi-supervised learning models. Only with complete and accurate customer data can the depth and breadth of subsequent analysis be ensured, leading to the generation of highly personalized and effective operational strategies, ultimately improving customer satisfaction and business conversion rates.

[0039] In step S102 , the target object data is input into a pre-trained semi-supervised learning model, a feature vector is extracted from the target object data by the semi-supervised learning model, and a feature label of the target object is predicted based on the feature vector.

[0040] In step S102 , the feature tags are used to mark the target object's preference features when communicating with the financial institution.

[0041] Optionally, the system can first organize and preprocess the collected data on the target audience's channel preferences, historical transaction records, and interactive behavior. Preprocessing steps may include data cleaning (e.g., removing erroneous or irrelevant information), format conversion, and normalization to ensure data quality and consistency, making it suitable for the input requirements of the machine learning model.

[0042] Alternatively, a semi-supervised learning model is a machine learning model that can learn and predict using a large amount of unlabeled data, guided by limited labeled data. During the model training phase, a portion of labeled target object data (i.e., known preferred characteristics) is used as guidance, allowing the model to learn to identify the association between preferred characteristics and data using this labeled data. Simultaneously, the model can leverage the majority of unlabeled target object data to enhance learning and discover more data structures and patterns.

[0043] Alternatively, a feature vector is a data representation in machine learning that converts raw data into a collection of numerical features that capture the data's key attributes and patterns. In customer data analysis, a feature vector may contain quantitative information on multiple dimensions, such as a user's transaction frequency, transaction amount, communication channel preferences, and interactive activity. Semi-supervised learning models automatically extract feature vectors from the input target object data. This process is essentially an internal computational mechanism within the model, which uses neural networks or other machine learning algorithms to perform multi-level abstraction and transformation of data. The purpose of feature extraction is to simplify complex and diverse customer data into a set of numerical features that are easy to understand and analyze, facilitating subsequent preference prediction.

[0044] Optionally, the feature label is an identifier used to mark the target object's preference characteristics when communicating with the financial institution. It can be a category label (such as "prefer social media communication"), a level label (such as "highly active customers"), or a continuous value label (such as "average number of transactions per month"). These labels can intuitively reflect the customer's behavior and preferences, and provide direct guidance for designing personalized operation strategies. Based on the feature vector, the semi-supervised learning model can predict the feature label of the target object. The model will analyze the feature vector of each target object according to the data pattern learned during training, and output the label that best represents the preference characteristics of the object. This process is a reflection of the application of the model. By applying the results of model training to new data, the preference characteristics of unknown target objects can be predicted.

[0045] In summary, the above steps fully utilize the advantages of semi-supervised learning. Even when labeled data is limited, it can effectively mine potential preference characteristics from the target object data, thereby generating highly personalized operation strategies, effectively solving the problem of traditional methods where strategy generation is too simplistic and difficult to accurately match customer preferences.

[0046] Step S103 : generating a target operation strategy corresponding to the target object based on the characteristic tag of the target object and in combination with a preset operation strategy template.

[0047] Optionally, feature labels are derived from analyzing target customer data using a semi-supervised learning model. Feature labels represent specific preferences and behavioral characteristics of target customers when interacting with financial institutions. For example, one feature label might be "preferring frequent email communication," while another might be "inclined to pay via mobile." Feature labels may encompass multiple aspects, including but not limited to customer contact channel preferences, transaction behavior characteristics, interaction frequency preferences, product and service needs, and customer satisfaction levels. These multi-dimensional labels collectively paint a comprehensive picture of a customer, providing a comprehensive perspective for subsequent strategy customization.

[0048] Optionally, an operational strategy template refers to a set of standardized policy frameworks designed in advance by financial institutions to guide how to communicate and interact with customer groups with specific characteristics. These templates typically include a variety of communication methods, message types, customer contact frequency, product and service recommendations, and are intended to cover a wide range of situations and needs. Once the characteristic labels of the target audience are obtained, financial institutions can personalize the preset operational strategy templates based on these labels to ensure that the strategy accurately matches customer preferences. For example, if a customer prefers to communicate via social media, the original email promotion plan in the strategy template may need to be adjusted to a social media interaction plan.

[0049] In this embodiment of the present application, the system can use feature tags to enable financial institutions to select the most appropriate portion of a pre-set operational strategy template from a customer's profile, or fine-tune and combine existing templates to form a set of operational strategies tailored to specific target customers. This process can also include automated strategy selection and adjustment, as well as manual review and supplementation steps to ensure the rationality and effectiveness of the strategy.

[0050] In an optional embodiment, after generating a target operation strategy corresponding to the target object, the system can determine the target operation content and the promotion timing and promotion method of the target operation content based on the target operation strategy, and use the promotion method at the promotion timing to send the target operation content to the terminal device of the target object.

[0051] Optionally, based on the target audience's characteristic tags and customized operational strategies, operational content can be designed to meet the specific needs and preferences of that customer. This can include product introductions, promotional information, market trends, educational materials, or special customer care information. For example, if analysis results indicate that the target audience is young professionals who frequently use mobile banking apps, operational content might focus on introducing the latest fintech products and providing financial advice related to career development to attract and maintain the interest of this customer group.

[0052] Secondly, determining promotional timing is another key factor in personalized marketing strategies. Data analysis can identify target audiences' active periods, peak transaction times, and special event windows (such as birthdays and anniversaries), allowing for optimal timing in marketing content delivery. This approach aims to maximize the relevance and willingness of customers to receive messages, avoiding sending messages during periods when customers are less likely to receive them, which could reduce communication effectiveness.

[0053] Furthermore, the selection of promotional methods should be based on the target audience's channel preference data. For example, if a customer prefers social media interaction, then content should be pushed through social media platforms; if a customer prefers email communication, then email should be used as a promotional channel. Furthermore, you can consider combining the advantages of multiple channels and adopting a cross-channel promotion strategy to cover multiple possible customer touchpoints.

[0054] As can be seen from the above, this application implements a complete personalized customer operation process by determining personalized operation content, optimizing promotion timing and methods, and accurately delivering them to target customer terminals. This process not only improves the efficiency and effectiveness of customer communication, but also provides financial institutions with the opportunity to continuously optimize and iterate their operation strategies, enabling them to better serve customers and improve overall business performance.

[0055] In an optional embodiment, after the target operation content is sent to the terminal device of the target object, feedback status information of the target object regarding the target operation content is collected, and the operation strategy template and / or semi-supervised learning model is updated based on the feedback status information.

[0056] Optionally, feedback status information refers to the target customer's response to the campaign content. This information can include behavioral data (such as click-through rate, reply rate, and purchase conversion rate), emotional feedback (such as satisfaction ratings, comments, and complaints), and changes in customer engagement (such as the number of social media interactions). This information directly reflects the actual effectiveness of the campaign strategy and is an important basis for evaluating its success or failure.

[0057] Based on the collected feedback status information, financial institutions can evaluate the actual effectiveness of each operational strategy template and modify or replace templates that perform poorly or are no longer applicable. For example, if the data shows that the open rate of email promotion has dropped significantly, it may be necessary to reduce the frequency of emails or change the design of the email content. Similarly, feedback status information is also used to update and optimize semi-supervised learning models. Newly collected data (including feedback status information) is used as training samples to calibrate the model's predictive ability and ensure that the model can continue to adapt to market changes and customer preference updates. Through the retraining process, the accuracy and generalization ability of the model are enhanced, and the feature labels for future predictions will be more accurate, and the generated operational strategies will be more effective.

[0058] As can be seen from the above, collecting feedback from target entities and updating operational strategy templates and semi-supervised learning models accordingly is a key step in achieving continuous optimization and intelligent upgrades of customer operational strategies. This closed-loop mechanism not only improves the relevance and effectiveness of operational strategies but also promotes the continuous evolution of machine learning models.

[0059] In an optional embodiment, upon obtaining authorization from the target object, obtaining target object data includes: upon obtaining authorization from the target object, the system may configure a data acquisition interface related to the target object. Through the data acquisition interface, the system may synchronously acquire initial data from the target object, then convert the initial data into intermediate data in a target format. Finally, the system may perform a repair operation on the intermediate data to obtain the target object data, wherein the repair operation includes filling null values ​​in the intermediate data, correcting erroneous values ​​in the intermediate data, and correcting the collection time corresponding to the intermediate data.

[0060] Optionally, the system can be configured with a dedicated data collection interface. This serves as a channel for the system to interact with data sources, responsible for data extraction, transmission, and preliminary processing. This interface may connect to various data sources, such as bank transaction systems, social media platforms, and third-party customer relationship management systems. Through this data collection interface, the system can synchronously collect initial data on target entities in real time or on a regular basis. This ensures data timeliness and continuity, which is crucial for capturing changes in customer behavior.

[0061] Furthermore, initial data may come from various sources and formats. To ensure unified processing, the system converts this initial data into intermediate data in the target format. This typically involves operations such as data field mapping, encoding conversion, and data type adjustment. Intermediate data is data that has undergone preliminary conversion but has not yet undergone quality inspection and complete processing. It serves as a bridge between the original data and the final analyzed data, ensuring smooth data flow within the system.

[0062] Optionally, during the data collection process, some fields may be missing information. The system can use statistical methods (such as mean filling), predictive models, or other logical rules to fill these empty values ​​to ensure data integrity and avoid affecting the accuracy of subsequent analysis due to missing data. Secondly, erroneous values ​​in intermediate data may be caused by data entry errors, damage during transmission, or anomalies in the data itself. The system needs to identify and correct these erroneous values, such as correcting impossible negative balances to zero, or correcting inconsistent date information through contextual logic judgment.

[0063] Optionally, the accuracy of data collection time is particularly important for time series analysis. The system needs to calibrate the data timestamp to resolve issues such as time zone differences and system time deviations to ensure the consistency and accuracy of all data in the time coordinate.

[0064] After these conversion and repair operations, the system obtains the target object data—a cleansed, formatted, and validated dataset that can be directly used for data analysis and machine learning model training without worrying about data quality issues. In short, this data processing process ensures data quality and compliance throughout the acquisition, conversion, and repair processes, providing a clean, uniformly formatted dataset for subsequent analysis. This is key to implementing refined, data-driven operational strategies.

[0065] In an optional embodiment, Figure 2 This is a flow chart of generating an optional feature tag according to an embodiment of the present application, such as Figure 2 As shown, the following steps are included:

[0066] Step S201: convert the target object data into the format required by the semi-supervised learning model to obtain model input data.

[0067] Step S202: compress the model input data to obtain compressed data.

[0068] In step S203 , after receiving the compressed data, the compressed data is decompressed by each neural network layer of the semi-supervised learning model, and a feature vector is extracted from the decompressed data.

[0069] The feature vector includes at least: channel preference characteristics, transaction behavior characteristics, and interaction frequency characteristics of the target object.

[0070] Step S204: predicting a feature label of the target object based on the feature vector.

[0071] Alternatively, semi-supervised learning models often require data input in a specific format. For example, numerical, categorical, or textual data may need to be converted to vector representations. Data conversion ensures that the original target object data can be correctly interpreted and processed by the model. Specifically, the data conversion process can include the following steps:

[0072] Encode categorical data into numerical values, such as using one-hot encoding to represent customer channel preferences;

[0073] Convert text data into word embedding vectors to capture semantic information;

[0074] Format date and time data so that it can be used for time series analysis.

[0075] Optionally, when the amount of data is huge, the model input data can be compressed to reduce storage space requirements and speed up data processing. Data compression techniques such as PCA (Principal Component Analysis) and t-SNE (t-Distributed Stochastic Neighbor Embedding) can reduce the dimensionality of the data, retaining the main information while reducing the data size. When compressed data enters the semi-supervised learning model, there may be a specific layer within the model (such as a decoder) responsible for restoring it to its original size or a state close to the original size to enable more detailed data feature extraction and analysis. This process is called data decompression, which ensures that the model can fully understand the details of the data.

[0076] Optionally, the neural network layers of the semi-supervised learning model process the decompressed version of the compressed data, extracting deeper features from the data through multiple layers of nonlinear transformations. This feature information is presented as a feature vector, an array of numbers that captures the inherent structure of the data. The extracted feature vector contains at least the following three key aspects:

[0077] Channel preference characteristics: reflects the target audience's preference for different communication channels, such as email, text messages, social media, etc.

[0078] Transaction behavior characteristics: summarize the customer's trading habits, such as transaction frequency, transaction amount, transaction time, etc.

[0079] Interaction frequency characteristics: Describes the frequency of communication between the target object and the financial institution, including statistical analysis of various interaction records.

[0080] Optionally, a semi-supervised learning model uses the extracted feature vectors to predict the target object's feature labels. Because the model has already learned from a large amount of labeled and unlabeled data during the training phase, it can accurately identify the association between feature vectors and specific labels and make predictions. Furthermore, feature labels are the result of model predictions and are used to identify the preferences and behavioral characteristics of the target object. For example, the model might predict that a customer has feature labels such as "high-interaction customer," "email preference," or "high-transaction customer." These labels provide specific direction for customized operational strategies.

[0081] In summary, the above technical process achieves the transformation from raw customer data to refined operational strategy generation through data format conversion, compression and decompression, feature vector extraction, and feature label prediction.

[0082] In an optional embodiment, Figure 3 is a training flow chart of an optional semi-supervised learning model according to an embodiment of the present application, such as Figure 3 As shown, the following steps are included:

[0083] Step S301: Acquire operation record data of a historical reference object.

[0084] Step S302 : dividing the operation record data of the historical reference object into a labeled data set and an unlabeled data set.

[0085] In step S302, each data sample in the labeled data set is accompanied by one or more labels, and the labels are used to indicate the category and attribute information of the data sample; the data samples in the unlabeled data set are not accompanied by any labels.

[0086] Step S303: Initialize the semi-supervised learning model according to the labeled data set.

[0087] Step S304: Iterative training is performed on the unlabeled dataset to generate pseudo labels, and the pseudo labels are fed back into the model training process of the semi-supervised learning model to optimize the pseudo label generation process using reinforcement learning.

[0088] Step S305 , evaluating the performance of the semi-supervised learning model, and optimizing the model according to the evaluation results, until the model performance index reaches a preset threshold, and determining that the semi-supervised learning model training ends.

[0089] Optionally, labeled datasets contain sample data with known labels. These labels can be manually added through customer surveys, historical data analysis, or expert knowledge. They provide the learning objectives required for model training, enabling the model to understand the association between specific data features and labels. Unlabeled datasets consist of a large number of unlabeled data samples. Although these data have not been manually annotated, they contain rich customer behavior information and are an important data resource for semi-supervised learning models to learn and generate pseudo-labels.

[0090] Optionally, all data used in the model training process are user-authorized data, and during the training process, the training data is desensitized and encrypted as necessary, and is subject to audit by a third-party organization at any time.

[0091] Alternatively, the semi-supervised learning model can be a model based on a deep learning architecture, such as a convolutional neural network (CNN), a long short-term memory network (LSTM), or an autoencoder (AE), or it may use a traditional machine learning algorithm, such as a support vector machine (SVM) or a decision tree. Model initialization usually involves steps such as setting hyperparameters and randomly initializing weights. Among them, the initial training of the semi-supervised learning model is performed on a labeled dataset. Through this process, the model begins to learn how to distinguish between data samples of different categories, that is, it begins to have the ability to classify data samples. After the model has initially learned, an unlabeled dataset is fed into the model, and the model will attempt to assign labels to it. These labels generated by the model are called "pseudo-labels." Although they may not be completely accurate initially, with subsequent training, the model will gradually correct and improve these pseudo-labels.

[0092] In addition, reinforcement learning is applied to optimize the pseudo-label generation process. By setting a reward mechanism, the model is encouraged to generate more accurate pseudo-labels. Therefore, during iterative training on unlabeled datasets, the model can continuously adjust and optimize its internal parameters to improve prediction performance.

[0093] Optionally, model performance indicators include but are not limited to accuracy, recall, F1 score or area under the AUC-ROC curve, which are used to quantify the accuracy and robustness of model predictions. Optionally, a performance threshold can also be set. When the evaluation results of the model reach or exceed this threshold, the model training is considered complete and has reached the expected maturity and usability. Based on the performance evaluation results, it may be necessary to adjust the model architecture, hyperparameters or optimization algorithms. This process may include increasing or decreasing the number of neural network layers, changing the learning rate, using different loss functions or optimizers, etc., until the model performance meets the standards.

[0094] Training can be terminated when the model performance reaches a preset threshold or when the model performance stops improving after multiple iterations. Once the termination criteria are met, model training stops and the model is considered sufficiently trained to be used to predict feature labels for new target object data.

[0095] Through the above process, the semi-supervised learning model fully utilizes limited labeled data and abundant unlabeled data, not only reducing the cost of manual annotation but also enhancing the model's generalization ability, enabling it to more accurately identify and predict customer behavior. This model training method is particularly suitable for customer operations scenarios because it can handle the diversity and uncertainty of customer behavior, providing financial institutions with more refined and intelligent customer operations decision support.

[0096] In an optional embodiment, a target operation strategy corresponding to a target object is generated based on the target object's feature tags and a preset operation strategy template. The system first selects an operation strategy template that matches the target object's feature tags from a preset operation strategy template library. The system then performs semantic analysis on the operation strategy template and, based on the semantic analysis results and the target object's feature tags, adjusts the selected operation strategy template to obtain an initial operation strategy. Finally, the system uses a generative adversarial network to update the initial operation strategy to obtain the target operation strategy.

[0097] Optionally, the operation strategy template library is a pre-prepared database of various customer operation strategies. Each strategy template is designed for different customer characteristics. For example, one template might focus on email outreach, while another might emphasize social media engagement. The system automatically selects the most personalized strategy from the template library based on the target customer's characteristic tags. For example, for customers who prefer high-frequency interactions, the system will select strategy templates designed for a higher communication frequency.

[0098] Based on the selected template, the system conducts an in-depth semantic analysis of its content to understand the meaning and potential effects of each activity in the strategy, ensuring accurate communication of the strategy's message. The output of this semantic analysis is combined with the target audience's characteristic tags, allowing the system to identify which elements of the strategy may require adjustment for a specific customer. For example, if the characteristic tags indicate a customer preference for concise and clear information, the copy in the strategy may need to be streamlined. Combining the results of the semantic analysis with the details of the characteristic tags, the system then personalizes the selected operational strategy template, removing inappropriate activities and enhancing elements that align with customer preferences, ultimately forming an initial operational strategy for the specific target audience.

[0099] Alternatively, a generative adversarial network (GAN) is a specialized deep learning model consisting of a generator and a discriminator. The generator attempts to generate samples that mimic real data, while the discriminator attempts to distinguish real data from fake data generated by the generator. In embodiments of the present application, GANs are used to optimize operational strategies and generate more efficient and personalized strategic solutions.

[0100] First, the initial operation strategy is input into the generator of the generative adversarial network, which attempts to propose multiple strategy variations. Simultaneously, the discriminator evaluates the effectiveness and personalization of these variations and provides feedback to the generator, prompting it to continuously adjust the strategy and produce higher-quality strategy recommendations. After multiple rounds of iteration, the generative adversarial network is able to generate an operation strategy that not only conforms to the target object's feature labels, but is also semantically clear, highly personalized, and targeted. This final strategy, the target operation strategy, is customized to the customer's specific characteristics and needs to maximize its operational effectiveness.

[0101] As can be seen from the above, through precise matching, semantic analysis, and dynamic optimization using generative adversarial networks, the system not only generates highly personalized strategies but also innovates in both content and form, surpassing traditional strategy formulation methods and improving customer satisfaction and operational effectiveness. The introduction of generative adversarial networks adds a layer of automated innovation and optimization to the customization of operational strategies, enabling them to not only match current customer needs but also anticipate potential customer reactions, further enhancing the foresight and effectiveness of operational strategies.

[0102] According to another aspect of the embodiment of the present application, a device for determining a user operation strategy is also provided, wherein: Figure 4 is a schematic diagram of an optional device for determining a user operation strategy according to an embodiment of the present application, such as Figure 4 As shown, the device includes: an acquisition unit 401, a first processing unit 402, and a second processing unit 403.

[0103] Optionally, the acquisition unit 401 is used to obtain the target object data with the authorization of the target object, wherein the target object data includes: the target object's channel preference data, the target object's historical transaction records with financial institutions, and interactive behavior data; the first processing unit 402 is used to input the target object data into a pre-trained semi-supervised learning model, extract a feature vector from the target object data through the semi-supervised learning model, and predict the feature label of the target object based on the feature vector, wherein the feature label is used to mark the preference characteristics of the target object when communicating with the financial institution; the second processing unit 403 is used to generate a target operation strategy corresponding to the target object based on the feature label of the target object and a preset operation strategy template.

[0104] Optionally, the user operation strategy determination device also includes: a first determination unit, used to determine the target operation content and the promotion timing and promotion method of the target operation content based on the target operation strategy; a sending unit, used to use the promotion method at the promotion timing to send the target operation content to the terminal device of the target object.

[0105] Optionally, the user operation strategy determination device further includes: a collection unit for collecting feedback status information of the target object for the target operation content; and an updating unit for updating the operation strategy template and / or the semi-supervised learning model based on the feedback status information.

[0106] Optionally, the acquisition unit 401 includes: a configuration subunit, which is used to configure a data acquisition interface related to the target object with the authorization of the target object; a data acquisition subunit, which is used to synchronously acquire the initial data of the target object through the data acquisition interface; a conversion subunit, which is used to convert the initial data into intermediate data in a target format; and a repair subunit, which is used to perform a repair operation on the intermediate data to obtain the target object data, wherein the repair operation includes: filling in the empty values ​​of the intermediate data, correcting the erroneous values ​​of the intermediate data, and correcting the acquisition time corresponding to the intermediate data.

[0107] Optionally, the first processing unit 402 includes: a first conversion subunit, used to convert the target object data into the format required by the semi-supervised learning model to obtain model input data; a data compression subunit, used to compress the model input data to obtain compressed data; a first processing subunit, used to decompress the compressed data after receiving the compressed data through each neural network layer of the semi-supervised learning model, and extract a feature vector from the decompressed data, wherein the feature vector includes at least: the channel preference characteristics, transaction behavior characteristics, and interaction frequency characteristics of the target object; a second processing subunit, used to predict the feature label of the target object based on the feature vector.

[0108] Optionally, the device for determining the user operation strategy also includes: a first acquisition unit, used to acquire the operation record data of the historical reference object; a data set division unit, used to divide the operation record data of the historical reference object into a labeled data set and an unlabeled data set, wherein each data sample in the labeled data set is accompanied by one or more labels, and the labels are used to indicate the category and attribute information of the data sample; the data samples in the unlabeled data set are not accompanied by any labels; an initialization unit, used to initialize the semi-supervised learning model according to the labeled data set; a training unit, used to perform iterative training on the unlabeled data set, generate pseudo labels, and feed the pseudo labels back to the model training process of the semi-supervised learning model, and use reinforcement learning to optimize the pseudo label generation process; an optimization unit, used to evaluate the performance of the semi-supervised learning model, and optimize the model according to the evaluation results until the model performance index reaches a preset threshold, and determines that the semi-supervised learning model has ended training.

[0109] Optionally, the second processing unit 403 includes: a selection subunit, used to select an operation policy template that matches the feature label from a preset operation policy template library based on the feature label of the target object; a semantic analysis subunit, used to perform semantic analysis on the operation policy template; an adjustment subunit, used to adjust the selected operation policy template based on the analysis results of the semantic analysis and the feature label of the target object to obtain an initial operation policy; and a strategy update subunit, used to update the initial operation policy using a generative adversarial network to obtain a target operation policy.

[0110] According to another aspect of an embodiment of the present application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer-readable storage medium, wherein when the computer program is run, the device where the computer-readable storage medium is located executes the above-mentioned method for determining the user operation strategy.

[0111] According to another aspect of an embodiment of the present application, an electronic device is also provided, wherein the electronic device includes one or more processors and a memory, and the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned method for determining the user operation strategy.

[0112] According to another aspect of an embodiment of the present application, a computer program product is further provided, wherein the computer program product includes a computer program or instructions, and the computer program or instructions implement the above-mentioned method for determining the user operation strategy when executed by a processor.

[0113] The above-mentioned embodiments or examples disclosed in this application are not exhaustive, but are only illustrations of some embodiments or examples, and are not intended to be specific limitations on the scope of protection disclosed in this application. In the absence of contradiction, each step in a certain embodiment or example in this application can be implemented as an independent example, and the steps can be arbitrarily combined. For example, the solution after removing some steps in a certain embodiment or example can also be implemented as an independent example, and the order of the steps in a certain embodiment or example can be arbitrarily exchanged. In addition, the optional methods or optional examples in a certain embodiment or example can be arbitrarily combined; in addition, the various embodiments or examples can be arbitrarily combined. For example, some or all of the steps in different embodiments or examples can be arbitrarily combined, and a certain embodiment or example can be arbitrarily combined with the optional methods or optional examples of other embodiments or examples.

[0114] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0115] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0116] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0117] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0118] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0119] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0120] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining a user operation strategy, characterized in that: include: Obtaining target object data with authorization from the target object, wherein the target object data includes: channel preference data of the target object, historical transaction records of the target object with financial institutions, and interactive behavior data; Inputting the target object data into a pre-trained semi-supervised learning model, extracting a feature vector from the target object data using the semi-supervised learning model, and predicting a feature label of the target object based on the feature vector, wherein the feature label is used to mark the target object's preference characteristics when communicating with the financial institution; Based on the characteristic tags of the target object and in combination with a preset operation strategy template, a target operation strategy corresponding to the target object is generated.

2. The method according to claim 1, characterized in that After generating the target operation strategy corresponding to the target object, the method further includes: Determine the target operation content and the promotion timing and promotion method of the target operation content based on the target operation strategy; The target operation content is sent to the terminal device of the target object by using the promotion method at the promotion opportunity.

3. The method according to claim 2, characterized in that After sending the target operation content to the terminal device of the target object, the method further includes: Collecting feedback status information of the target object regarding the target operation content; The operation strategy template and / or the semi-supervised learning model is updated according to the feedback status information.

4. The method according to claim 1, wherein With authorization from the target object, obtain the target object's data, including: When authorization is obtained from the target object, a data collection interface related to the target object is configured; Synchronously collecting initial data of the target object through the data collection interface; Converting the initial data into intermediate data in a target format; Performing a repair operation on the intermediate data to obtain the target object data, wherein the repair operation includes: filling null values ​​of the intermediate data, correcting erroneous values ​​of the intermediate data, and correcting the collection time corresponding to the intermediate data.

5. The method according to claim 1, characterized in that Extracting a feature vector from the target object data using the semi-supervised learning model, and predicting a feature label of the target object based on the feature vector, including: Converting the target object data into a format required by the semi-supervised learning model to obtain model input data; Compressing the model input data to obtain compressed data; After receiving the compressed data, the compressed data is decompressed by each neural network layer of the semi-supervised learning model, and the feature vector is extracted from the decompressed data, wherein the feature vector includes at least: the channel preference feature, transaction behavior feature, and interaction frequency feature of the target object; Predicting a feature label of the target object based on the feature vector.

6. The method according to claim 1, characterized in that The training steps of the semi-supervised learning model include: Obtain operational record data of historical reference objects; Dividing the operation record data of the historical reference object into a labeled data set and an unlabeled data set, wherein each data sample in the labeled data set is accompanied by one or more labels, and the labels are used to indicate the category and attribute information of the data sample; the data samples in the unlabeled data set are not accompanied by any labels; Initializing the semi-supervised learning model according to the labeled dataset; Iterative training is performed on the unlabeled dataset to generate pseudo labels, and the pseudo labels are fed back into the model training process of the semi-supervised learning model, and the generation process of the pseudo labels is optimized using reinforcement learning; The performance of the semi-supervised learning model is evaluated, and the model is optimized according to the evaluation results until the model performance index reaches a preset threshold, and the semi-supervised learning model training is determined to be completed.

7. The method according to claim 1, characterized in that Based on the characteristic tags of the target object and in combination with a preset operation strategy template, a target operation strategy corresponding to the target object is generated, including: According to the characteristic tag of the target object, an operation strategy template that matches the characteristic tag is selected from a preset operation strategy template library; Performing semantic analysis on the operation strategy template; Adjusting the selected operation strategy template according to the analysis result of the semantic analysis and the characteristic label of the target object to obtain an initial operation strategy; The initial operation strategy is updated using a generative adversarial network to obtain the target operation strategy.

8. A device for determining a user operation strategy, characterized in that: include: an acquisition unit, configured to acquire target object data upon authorization by the target object, wherein the target object data includes: channel preference data of the target object, historical transaction records of the target object with financial institutions, and interactive behavior data; a first processing unit, configured to input the target object data into a pre-trained semi-supervised learning model, extract a feature vector from the target object data using the semi-supervised learning model, and predict a feature label of the target object based on the feature vector, wherein the feature label is used to mark the target object's preference characteristics when communicating with the financial institution; The second processing unit is configured to generate a target operation strategy corresponding to the target object based on the feature tag of the target object and in combination with a preset operation strategy template.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the method for determining the user operation strategy according to any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method for determining the user operation strategy described in any one of claims 1 to 7.

11. A computer program product, characterized in that The method comprises a computer program or an instruction, which, when executed by a processor, implements the method for determining the user operation strategy of any one of claims 1 to 7.

Images