A multi-queue optimized online transaction task scheduling method and device

Abstract

The present application discloses a multi-queue optimized online transaction task scheduling method and device, which belongs to the field of data analysis technology. The present application can not only deeply mine the text semantics of the initial transaction behavior from the online transaction monitoring log, but also obtain the text semantics of the derivative transaction behavior that is related to the initial semantics through repetition and re-mining technology, thereby greatly enriching the semantic information of the transaction behavior. Furthermore, by projecting these semantic information into the target semantic knowledge relationship network, the deep integration and structured representation of knowledge are achieved, providing a solid foundation for the generation of transaction behavior text linkage decision semantics. Finally, the use of the linkage decision semantics for online transaction task scheduling significantly improves the intelligence level and efficiency of scheduling, and provides new technical ideas and solutions for the optimization and upgrading of the online transaction environment.

Description

Technical Field

[0001] The present application belongs to the field of data analysis technology, and specifically relates to a multi-queue optimized online transaction task scheduling method and device. Background Art

[0002] In the current online trading environment, as the scale and complexity of transactions continue to increase, how to efficiently and accurately schedule trading tasks has become a technical problem that needs to be solved. Traditional trading task scheduling methods often rely on fixed rules and patterns, which are difficult to adapt to rapidly changing trading behaviors and market demands. At the same time, since trading behaviors themselves have rich semantic information, existing technologies can often only capture superficial and explicit semantics, while ignoring deep and implicit semantic connections, resulting in limited accuracy and efficiency of task scheduling. Summary of the invention

[0003] The present application provides a multi-queue optimized online transaction task scheduling method and device, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] The embodiment of the present application provides a multi-queue optimized online transaction task scheduling method, which is applied to an online transaction task scheduling device, and the method includes: obtaining initial transaction behavior text semantics to be optimized and scheduled for queue tasks; the initial transaction behavior text semantics is obtained by an online transaction task scheduling network performing semantic mining on an online transaction monitoring log under an online transaction project that matches the online transaction task scheduling network; obtaining at least one derived transaction behavior text semantics associated with the initial transaction behavior text semantics; the derived transaction behavior text semantics is a non-constrained text semantics, and is obtained by repeating the online transaction monitoring log to which the initial transaction behavior text semantics belongs and then performing semantic mining; projecting the initial transaction behavior text semantics and the at least one derived transaction behavior text semantics into a target semantic knowledge relationship network respectively, to obtain the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection; using the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection, to obtain transaction behavior text linkage decision semantics; the transaction behavior text linkage decision semantics is used for online transaction task scheduling.

[0005] An embodiment of the present application provides an online transaction task scheduling device, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0006] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0007] This application can not only deeply mine the text semantics of the initial transaction behavior from the online transaction monitoring log, but also obtain the text semantics of the derived transaction behavior that is related to the initial semantics through repetition and re-mining technology, thereby greatly enriching the semantic information of the transaction behavior. Furthermore, by projecting this semantic information into the target semantic knowledge relationship network, the deep integration and structured representation of knowledge is achieved, providing a solid foundation for the generation of transaction behavior text linkage decision semantics. Finally, the use of this linkage decision semantics for online transaction task scheduling significantly improves the intelligence level and efficiency of scheduling, and provides new technical ideas and solutions for the optimization and upgrading of the online transaction environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of a multi-queue optimized online transaction task scheduling method provided in an embodiment of the present application.

[0009] Figure 2 A schematic diagram of the structure of an online transaction task scheduling device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0011] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually a class, and the number of objects is not limited. For example, the first object can be one or more. In addition, in this application, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0012] Figure 1 A multi-queue optimized online transaction task scheduling method is shown, which is applied to an online transaction task scheduling device. The method includes the following steps 101 to 104.

[0013] In a modern e-commerce environment, efficient scheduling of transaction tasks is crucial to ensure the smooth progress of the transaction process. In order to improve the efficiency and response speed of online transactions, this application implements intelligent task scheduling by deeply analyzing the text semantics of transaction behavior. The following is a specific application scenario example that details how this application is implemented. A large e-commerce platform is holding an annual promotion event, and the transaction volume is expected to surge during the event. In order to ensure that the transaction system can withstand the pressure of high concurrent access and ensure the accuracy and timeliness of transactions, the e-commerce platform decided to use steps 101-104 to optimize the processing flow of transaction tasks.

[0014] Step 101: Acquire the semantics of the initial transaction behavior text for queue task optimization scheduling; the semantics of the initial transaction behavior text is obtained by an online transaction task scheduling network performing semantic mining on an online transaction monitoring log under an online transaction project matching the online transaction task scheduling network.

[0015] First, the online transaction task scheduling device needs to obtain the initial transaction behavior text semantics for the queue task optimization scheduling. These initial transaction behavior text semantics come from the results of deep semantic mining of the transaction monitoring logs under the matching online transaction project by the online transaction task scheduling network.

[0016] Specifically, during the activities of the e-commerce platform, the transaction system will generate a large number of transaction monitoring logs, which record various transaction behavior information, such as users browsing products, adding to shopping carts, placing orders, and paying. The online transaction task scheduling device will capture these logs in real time, and use natural language processing technology (NLP) and machine learning algorithms to perform semantic mining on the logs, extract keywords and phrases that can represent transaction behaviors, and form the initial transaction behavior text semantics.

[0017] For example, through semantic mining, the online transaction task scheduling device can extract the text semantics of initial transaction behaviors such as "user A browsed product B" and "user C added product D to the shopping cart".

[0018] Step 102: Obtain at least one derivative transaction behavior text semantics that is related to the initial transaction behavior text semantics; the derivative transaction behavior text semantics is a non-constrained text semantics, and is obtained by repeating the online transaction monitoring log to which the initial transaction behavior text semantics belongs and then performing semantic mining.

[0019] Next, the online transaction task scheduling device needs to obtain the derived transaction behavior text semantics that are related to the initial transaction behavior text semantics. These derived transaction behavior text semantics are non-constrained text semantics, which are obtained by repeating the online transaction monitoring log to which the initial transaction behavior text semantics belongs and then performing semantic mining.

[0020] The purpose of paraphrasing is to increase the diversity and richness of semantics, so as to more comprehensively capture different aspects of transaction behavior. For example, for the initial semantics of "user A browsed product B", after paraphrasing, it may be "user A viewed the details of product B" or "user A browsed product B".

[0021] By performing semantic mining on the restated text again, the online transaction task scheduling device can extract more derivative transaction behavior text semantics, such as "User A expressed interest in product B", "User A viewed the price and evaluation of product B", etc.

[0022] Step 103: project the initial transaction behavior text semantics and the at least one derived transaction behavior text semantics into the target semantic knowledge relationship network respectively, and obtain the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection.

[0023] After obtaining the semantics of the initial transaction behavior text and the derived transaction behavior text, the online transaction task scheduling device projects these semantics into a pre-built target semantic knowledge relationship network. The target semantic knowledge relationship network is a complex network containing various transaction behaviors and their relationships, which is used to represent the association and dependency between different transaction behaviors.

[0024] The projection process is to convert text semantics into nodes and edges in the knowledge network. For example, the semantics of "user A browsed product B" can be projected into a node in the knowledge network, representing a browsing behavior, and connected to the nodes representing user A and product B.

[0025] Through projection, the online transaction task scheduling device can transform text semantics into structured knowledge representation, which is convenient for subsequent analysis and processing.

[0026] Step 104: Utilize the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection to obtain transaction behavior text linkage decision semantics; the transaction behavior text linkage decision semantics is used to perform online transaction task scheduling.

[0027] Finally, the online transaction task scheduling device generates transaction behavior text linkage decision semantics using the initial transaction behavior text semantics and derived transaction behavior text semantics that complete the knowledge projection. The transaction behavior text linkage decision semantics is used to guide the scheduling of online transaction tasks.

[0028] Specifically, the online transaction task scheduling device analyzes the associations and dependencies between different nodes and edges in the knowledge relationship network and identifies potential patterns and rules between transaction behaviors. For example, the online transaction task scheduling device can find patterns such as "users usually place orders within a period of time after browsing products" or "users often complete payment within a short period of time after adding products to the shopping cart."

[0029] Based on these patterns and rules, the online transaction task scheduling device will generate corresponding linkage decision semantics, such as "when user A browses product B for more than a certain period of time, prioritize the order task" or "when user C adds product D to the shopping cart, immediately prepare the payment task."

[0030] The transaction behavior text-linked decision semantics will be applied to transaction task scheduling in real time to guide the priority allocation and processing order of tasks, thereby optimizing the transaction process and improving transaction efficiency and user experience.

[0031] It can be seen that through the implementation of the above steps, this application can deeply analyze the text semantics of transaction behaviors, dig out the potential patterns and rules between transaction behaviors, and generate corresponding linkage decision semantics to guide the scheduling of transaction tasks. This intelligent scheduling method can significantly improve the processing efficiency and response speed of online transactions, and provide strong support for online transaction systems such as e-commerce platforms. In high-concurrency scenarios such as promotional activities, the application of this application is particularly critical, and can effectively ensure the stability of the transaction system and the satisfaction of the user experience.

[0032] It can be understood that when introducing in depth the initial transaction behavior text semantics, the derived transaction behavior text semantics, the initial transaction behavior text semantics after completing the knowledge projection, the derived transaction behavior text semantics after completing the knowledge projection, and the transaction behavior text linkage decision semantics, these semantic features can be specifically displayed by constructing feature vectors.

[0033] (1) Semantic feature vector of initial transaction behavior text

[0034] For the initial transaction behavior text semantics of "user A browsed product B", its feature vector can be expressed as: [user ID: A, product ID: B, behavior type: browse, timestamp: t1, weight: 1.0].

[0035] Among them, user ID and product ID are key entities in text semantics, behavior type indicates the type of transaction behavior, timestamp records the time when the behavior occurs, and weight reflects the importance of the semantics in the overall transaction behavior.

[0036] (2) Semantic feature vector of derivative transaction behavior text

[0037] For the derived transaction behavior text semantics of "User A expressed interest in product B" obtained from the repetition of "User A browsed product B", its feature vector can be expressed as: [User ID: A, Product ID: B, Behavior type: expressed interest, Timestamp: t2, Weight: 0.8].

[0038] Compared to the original semantics, the behavior type and timestamp of the derived semantics may be different, and the weights may also be adjusted according to the importance of the semantics after retelling.

[0039] (3) Initial transaction behavior text semantic feature vector after knowledge projection

[0040] When the semantics of "user A browsed product B" is projected into the target semantic knowledge relationship network, its feature vector may be expanded to: [user ID: A, product ID: B, behavior type: browse, timestamp: t1, weight: 1.0, associated users: [C, D], associated products: [E, F], associated behaviors: [add to cart, place order]].

[0041] Among them, the feature vector adds dimensions such as associated users, associated products and associated behaviors, which are used to represent other node information connected to the semantic node in the knowledge relationship network.

[0042] (4) Complete the semantic feature vector of the derivative transaction behavior text after knowledge projection

[0043] Similarly, after projection, the feature vector of the derived semantics “user A expressed interest in product B” may be expanded to: [user ID: A, product ID: B, behavior type: expressing interest, timestamp: t2, weight: 0.8, associated users: [G, H], associated products: [I, J], associated behaviors: [view reviews, collect]].

[0044] (5) Transaction behavior text linkage decision semantic feature vector

[0045] Finally, based on the semantic features after the above projection, the feature vector of the generated transaction behavior text linkage decision semantics "When user A browses product B for more than a certain time, the order task is prioritized" can be expressed as: [trigger condition: user A browses product B for more than a certain time, decision behavior: prioritize order tasks, priority: high, related user ID: A, related product ID: B, decision timestamp: t3, decision validity period: 60 minutes].

[0046] This feature vector includes multiple dimensions such as trigger conditions, decision behavior, priority, related user ID, related product ID, decision timestamp, and decision validity period, which are used to comprehensively describe the content and execution requirements of the linkage decision semantics.

[0047] Through the above examples of feature vectors, we can clearly see the specific differences and connections in the numerical representation of the initial transaction behavior text semantics, the derived transaction behavior text semantics, the initial transaction behavior text semantics after knowledge projection, the derived transaction behavior text semantics after knowledge projection, and the transaction behavior text linkage decision semantics. These feature vectors not only contain the key information of text semantics, but also expand the association information with other nodes in the knowledge relationship network, as well as the decision information used to guide the scheduling of transaction tasks.

[0048] It is worth mentioning that step 103 is the core invention of this application. Its technical solution aims to effectively project the initial transaction behavior text semantics and the derived transaction behavior text semantics into a pre-built target semantic knowledge relationship network, so as to obtain a structured knowledge representation rich in association information. First, the construction of the target semantic knowledge relationship network is clarified. The target semantic knowledge relationship network is a complex semantic structure that contains various transaction behaviors and their associations and dependencies. It is like a huge graph, with nodes representing entities such as transaction behaviors, users, and commodities, and edges representing the relationships between these entities. Next, the initial transaction behavior text semantics are projected. For each initial transaction behavior text semantic, such as "user A browsed commodity B", the online transaction task scheduling device will map it to the target semantic knowledge relationship network. Specifically, it is to find or create a node representing the behavior in the network, and connect it with the nodes of user A and commodity B to form an edge representing the browsing behavior. At the same time, this node and edge will also be given a certain weight to reflect the importance of the behavior in the overall transaction. Secondly, the derived transaction behavior text semantics are projected in the same way. The derived semantics are obtained from the repetition of the initial semantics, and they also contain important transaction behavior information. Therefore, the system will also project these derived semantics into the target semantic knowledge relationship network, create or find corresponding nodes, and connect them with relevant user and commodity nodes. Then, during the projection process, the online transaction task scheduling device will also use the associated information in the knowledge relationship network to enrich the projection results. For example, when projecting the semantics of "user A expressed interest in commodity B", the system will not only create a node representing the "express interest" behavior, but also find other behavior nodes related to the behavior, such as "add to shopping cart", "place an order", etc., according to the associated information in the knowledge relationship network, and establish connections between them. Finally, through this series of projection processing, the initial transaction behavior text semantics and the derived transaction behavior text semantics are effectively converted into structured knowledge representations. These representations not only contain the key information of the text semantics, but also expand the associated information with other nodes in the knowledge relationship network. Such knowledge representation provides strong support for subsequent transaction behavior analysis and decision making.

[0049] In addition, step 104 involves organically integrating the semantics of the initial transaction behavior text that has completed knowledge projection with the semantics of the derived transaction behavior text to generate the semantics of the transaction behavior text linkage decision. This process is the key to achieving efficient scheduling of online transaction tasks. First of all, it should be clear that the semantics of the initial transaction behavior text represents the most basic and direct semantic information in the transaction behavior, and it is the core content of the transaction behavior text. The derived transaction behavior text semantics is further mined and expanded on the basis of the initial semantics through methods such as knowledge projection, and it contains more implicit information and contextual connections related to the transaction behavior. In step 104, the two kinds of semantic information are organically combined. Specifically, the initial transaction behavior text semantics and the derived transaction behavior text semantics are deeply analyzed and compared to find out the connection points and complementarities between them. Then, using these connection points and complementarities, the two kinds of semantic information are integrated to form a more comprehensive and rich transaction behavior text linkage decision semantics. This integration process is not a simple superposition or splicing, but needs to take into account the logical relationship and semantic hierarchy between the two kinds of semantic information. It is necessary to ensure that the fused linkage decision semantics can accurately reflect the essential characteristics of the transaction behavior, while containing sufficient contextual information and implicit knowledge to support the subsequent online transaction task scheduling. Finally, the generated transaction behavior text linkage decision semantics will be used in the actual online transaction task scheduling. It will serve as an important basis for scheduling decisions, helping the system to understand and process transaction behavior texts more accurately and efficiently, thereby achieving more intelligent transaction task scheduling.

[0050] In summary, this application obtains the initial transaction behavior text semantics for queue task optimization scheduling, which is obtained by deep mining from the monitoring logs under the online transaction project that matches the online transaction task scheduling network, ensuring the accuracy and pertinence of the semantics. Furthermore, by obtaining the derived transaction behavior text semantics associated with the initial transaction behavior text semantics, and performing unconstrained repetition and re-mining of these semantics, the semantic information of the transaction behavior is greatly enriched, and the coverage and diversity of the semantics are improved. The initial and derived transaction behavior text semantics are projected into the target semantic knowledge relationship network, completing the deep integration and projection of knowledge, making the semantic information more structured and systematic. Finally, using these semantic information that completes the knowledge projection, the transaction behavior text linkage decision semantics is obtained, which provides strong decision support for online transaction task scheduling and significantly improves the intelligence level and efficiency of task scheduling.

[0051] In some preferred embodiments, the obtaining of at least one derivative transaction behavior text semantics associated with the initial transaction behavior text semantics includes: determining the online transaction monitoring log to which the initial transaction behavior text semantics belongs, and performing at least one content repetition operation on the online transaction monitoring log to obtain online transaction monitoring logs with different output states; and performing semantic mining on the online transaction monitoring logs with different output states respectively to obtain at least one derivative transaction behavior text semantics.

[0052] Based on this embodiment, the semantic information of the transaction behavior can be enriched and expanded by performing content repetition and semantic mining on the online transaction monitoring log to which the initial transaction behavior text semantics belongs, thereby providing more comprehensive and accurate decision support for subsequent online transaction task scheduling.

[0053] First, the initial transaction behavior text semantics are deeply mined from the monitoring logs under the online transaction projects that match the online transaction task scheduling network. These monitoring logs record various details and characteristics of transaction behaviors and are an important source of transaction behavior text semantics. Therefore, in order to obtain the derived transaction behavior text semantics that are related to the initial transaction behavior text semantics, it is first necessary to determine the online transaction monitoring logs to which these initial semantics belong.

[0054] After determining the online transaction monitoring logs to which the semantics of the initial transaction behavior text belongs, the next step is to perform content retelling on these logs. Content retelling is an important text processing technology that rewrites, reorganizes or transforms the original text to generate texts with different output states. These texts with different output states retain the core information of the original text while adding new expressions and contexts, thus providing more possibilities and diversity for subsequent semantic mining.

[0055] When performing content paraphrasing, a variety of paraphrasing strategies and methods can be used. For example, the sentences in the original log can be updated with synonyms, sentence structures, word order, etc. to generate paraphrased texts with different expressions. At the same time, the professional terms and specific expressions in the original log can be interpreted and expanded in combination with domain knowledge and contextual information to generate richer and more specific paraphrased texts. These paraphrased texts retain the core semantics of the original log while adding new semantic information and contextual connections, providing more comprehensive and in-depth materials for subsequent semantic mining.

[0056] After obtaining the online transaction monitoring logs with different output states, the next step is to perform semantic mining on these logs. Semantic mining is a technology that extracts useful information and knowledge from a large amount of text data. It extracts key information and semantic relationships in the text by performing word segmentation, part-of-speech tagging, syntactic analysis, entity recognition and other processing on the text. In the present invention, the online transaction task scheduling device uses semantic mining technology to perform in-depth analysis on the paraphrased online transaction monitoring logs to extract the derived transaction behavior text semantics that are related to the initial transaction behavior text semantics.

[0057] When performing semantic mining, the online transaction task scheduling device can use a variety of algorithms and models. For example, the online transaction task scheduling device can use a text representation method based on the bag-of-words model to convert the restated log text into a vector form, and extract the derived semantics related to the initial semantics by calculating the similarity between the vectors. At the same time, the online transaction task scheduling device can also use text representation methods based on deep learning, such as word embedding, text embedding, etc., to convert the restated log text into a vector representation in a high-dimensional space, and extract more complex and abstract derived semantics through models such as neural networks. These derived semantics retain the core features of the initial semantics while adding new semantic information and contextual connections, providing more comprehensive and accurate decision support for subsequent online transaction task scheduling.

[0058] In this way, by determining the online transaction monitoring logs to which the initial transaction behavior text semantics belong, and performing content retelling and semantic mining operations on these logs, at least one derivative transaction behavior text semantics associated with the initial semantics is successfully obtained. This process not only enriches the semantic information of the transaction behavior, improves the coverage and diversity of the semantics, but also provides more comprehensive and accurate decision support for subsequent online transaction task scheduling. At the same time, the content retelling and semantic mining technology used in this embodiment also has certain versatility and extensibility, and can be applied to other similar text processing and knowledge mining tasks.

[0059] In the next step, the text semantics of the derived transaction behavior is extracted through the debugged derived semantic mining network, and the debugging step of the derived semantic mining network includes: obtaining multiple online transaction monitoring log samples; for each online transaction monitoring log sample, performing at least one content repetition operation to obtain at least one involved transaction monitoring log corresponding to each online transaction monitoring log sample; performing semantic mining on each involved transaction monitoring log through the derivative semantic mining network to be debugged to obtain the involved transaction behavior semantics corresponding to each online transaction monitoring log sample; generating a feature derivative training error based on the semantic commonality coefficient between the involved transaction behavior semantics corresponding to each online transaction monitoring log sample; and performing multiple rounds of debugging on the derived semantic mining network to be debugged based on the feature derivative training error to obtain a debugged derivative semantic mining network.

[0060] It can be understood that the acquisition of the semantics of derivative transaction behavior text depends on the debugged derivative semantic mining network. The debugging process of the derivative semantic mining network aims to improve the network's semantic mining ability for the involved transaction monitoring logs by continuously optimizing the network parameters and structure, so as to obtain more accurate and richer semantics of derivative transaction behavior text.

[0061] First, the online transaction task scheduling device needs to obtain multiple online transaction monitoring log samples. These samples are the basis for derivative semantic mining network debugging. They contain various types of transaction behaviors and semantic information, and are important materials for network learning and optimization. By deeply analyzing and mining these samples, the online transaction task scheduling device can extract the core features and semantic relationships of transaction behaviors, providing strong support for subsequent network debugging.

[0062] Secondly, for each online transaction monitoring log sample, the online transaction task scheduling device needs to perform at least one content retelling operation. The purpose of this step is to generate involved transaction monitoring logs with different expressions and contexts by rewriting, reorganizing or transforming the original logs. These involved logs retain the core information of the original logs while adding new expressions and semantic information, providing more possibilities and diversity for subsequent semantic mining. The content retelling operation can adopt a variety of strategies and methods, such as synonym replacement, sentence transformation, word order adjustment, etc., to ensure that the generated involved logs have rich diversity and coverage.

[0063] Then, the online transaction task scheduling device performs semantic mining on each involved transaction monitoring log through the derivative semantic mining network to be debugged. This step is the core link of network debugging, which aims to extract derivative semantic information related to transaction behavior from the involved logs through network learning and optimization. When performing semantic mining, the online transaction task scheduling device can use a variety of algorithms and models, such as text representation methods based on bag-of-words models and text representation methods based on deep learning, to ensure that the extracted derivative semantic information is accurate and rich. At the same time, the online transaction task scheduling device also needs to evaluate and screen the quality of the extracted derivative semantic information to ensure that it meets actual needs and application scenarios.

[0064] Next, the online transaction task scheduling device needs to generate feature-derived training errors based on the semantic commonality coefficients between the semantics of the involved transaction behaviors corresponding to each online transaction monitoring log sample. This step is a key link in network debugging, which aims to evaluate the network's ability and accuracy in mining transaction behavior semantics by calculating the commonality of the derived semantics between the involved logs. The semantic commonality coefficient can be calculated using a variety of methods, such as cosine similarity based on the vector space model, Jaccard similarity based on semantic similarity calculation, etc. By calculating these commonality coefficients, the online transaction task scheduling device can obtain the errors and deficiencies of the network in mining transaction behavior semantics, providing powerful guidance for subsequent network debugging.

[0065] Finally, the online transaction task scheduling device needs to perform multiple rounds of debugging on the derivative semantic mining network to be debugged based on the feature-derived training error. This step is the final link of network debugging, which aims to improve the network's mining ability and accuracy of transaction behavior semantics by continuously optimizing network parameters and structure. When debugging the network, the online transaction task scheduling device can adopt a variety of optimization algorithms and strategies, such as gradient descent method, back propagation algorithm, etc., to ensure that the network can continuously learn and optimize. At the same time, the online transaction task scheduling device also needs to perform performance evaluation and testing on the debugged network to ensure its stability and reliability in practical applications. After multiple rounds of debugging and optimization, the online transaction task scheduling device can obtain a debugged derivative semantic mining network, which can accurately extract the derivative transaction behavior text semantics from the online transaction monitoring log, and provide strong support for subsequent online transaction task scheduling.

[0066] It should be noted that the acquisition of the semantics of the derivative transaction behavior text depends on the debugged derivative semantic mining network, and the debugging process of the derivative semantic mining network covers multiple steps and links, including obtaining online transaction monitoring log samples, performing content repetition operations, performing semantic mining, calculating semantic commonality coefficients, generating feature derivative training errors, and performing multiple rounds of network debugging. Through the orderly execution and continuous optimization of these steps and links, the online transaction task scheduling device can obtain a debugged derivative semantic mining network, which can accurately extract the semantics of the derivative transaction behavior text from the online transaction monitoring log, and provide strong support for the subsequent online transaction task scheduling. At the same time, the debugging process also has certain versatility and scalability, and can be applied to other similar text processing and knowledge mining tasks.

[0067] In some other optional embodiments, the generating of feature-derived training errors based on the semantic commonality coefficients between the semantics of the involved transaction behaviors corresponding to each online transaction monitoring log sample includes: determining, for any online transaction monitoring log sample among the multiple online transaction monitoring log samples, a first semantic commonality coefficient between the semantics of the involved transaction behaviors corresponding to the targeted online transaction monitoring log sample; determining a second semantic commonality coefficient between the semantics of the involved transaction behaviors corresponding to the targeted online transaction monitoring log sample and the semantics of the involved transaction behaviors corresponding to the remaining online transaction monitoring log samples; and generating feature-derived training errors based on the first semantic commonality coefficient and the second semantic commonality coefficient.

[0068] In actual application, this embodiment aims to evaluate the accuracy and consistency of the derived semantic mining network in extracting transaction behavior semantics by calculating and analyzing the commonality between the semantics involved in transaction behaviors, and then generate feature-derived training errors for network debugging.

[0069] First, each online transaction monitoring log sample is subjected to a content retelling operation to obtain at least one corresponding involved transaction monitoring log. These involved logs retain the core information of the original logs while adding new expressions and semantic information. By performing semantic mining on these involved logs, the online transaction task scheduling device can obtain the semantics of involved transaction behaviors corresponding to each online transaction monitoring log sample. These semantic information are the basis for the online transaction task scheduling device to subsequently calculate the semantic commonality coefficient and generate feature-derived training errors.

[0070] For any online transaction monitoring log sample among the multiple online transaction monitoring log samples, the online transaction task scheduling device first determines the first semantic commonality coefficient between the semantics of each involved transaction behavior corresponding to the targeted online transaction monitoring log sample. The purpose of this step is to evaluate the consistency and similarity of the derived semantics between different involved logs under the same log sample. By calculating the commonality coefficients between these involved semantics, the online transaction task scheduling device can obtain the accuracy and stability of the network when extracting the derived semantics between different involved logs under the same log sample. If the commonality coefficient is high, it means that the network has good consistency and accuracy when extracting these semantics; if the commonality coefficient is low, it means that there are large differences and errors in the network when extracting these semantics.

[0071] Secondly, the online transaction task scheduling device needs to determine the second semantic commonality coefficient between the semantics of the transaction behavior involved corresponding to the targeted online transaction monitoring log sample and the semantics of the transaction behavior involved corresponding to the remaining online transaction monitoring log samples. The purpose of this step is to evaluate the similarity and correlation of the derived semantics between different log samples. By calculating the commonality coefficient of the involved semantics between these different log samples, the online transaction task scheduling device can obtain the generalization ability and adaptability of the network in extracting derived semantics between different log samples. If the commonality coefficient is high, it means that the network can better extract and identify the common semantic features between different log samples; if the commonality coefficient is low, it means that the network has large deviations and limitations when extracting these semantics.

[0072] After obtaining the first semantic commonality coefficient and the second semantic commonality coefficient, the online transaction task scheduling device can generate a feature-derived training error based on these two coefficients. Specifically, the online transaction task scheduling device can perform weighted summation or average operation on the first semantic commonality coefficient and the second semantic commonality coefficient to obtain a comprehensive commonality coefficient index. Then, the online transaction task scheduling device can compare this comprehensive commonality coefficient index with a preset threshold. If the comprehensive commonality coefficient index is lower than the preset threshold, it means that the network has large errors and deficiencies when extracting derived semantics. At this time, the online transaction task scheduling device can generate a large feature-derived training error to guide the debugging and optimization of the network; if the comprehensive commonality coefficient index is higher than the preset threshold, it means that the network has good accuracy and consistency when extracting derived semantics. At this time, the online transaction task scheduling device can generate a smaller feature-derived training error to maintain the stability and generalization ability of the network.

[0073] Through the above process, the online transaction task scheduling device can obtain the feature-derived training error corresponding to each online transaction monitoring log sample. These error information reflects the accuracy and consistency of the network in extracting derived semantics between different log samples, and is an important basis for the online transaction task scheduling device to subsequently debug and optimize the network. By conducting in-depth analysis and mining of these error information, the online transaction task scheduling device can find the problems and deficiencies of the network in extracting derived semantics, and then perform targeted operations such as adjusting the network structure and optimizing parameters to improve the performance and accuracy of the network.

[0074] In addition, it is worth noting that when calculating the semantic commonality coefficient, the online transaction task scheduling device can use a variety of calculation methods and technical means. For example, the online transaction task scheduling device can use a vector space model to calculate the cosine similarity between the involved semantics; or use a semantic similarity calculation model to calculate the Jaccard similarity between the involved semantics, etc. The selection of these calculation methods and technical means depends on the specific application scenario and requirements, as well as factors such as the data resources and computing resources owned by the online transaction task scheduling device.

[0075] Therefore, through the orderly execution and continuous optimization of the above steps and links, the online transaction task scheduling device can obtain accurate feature-derived training error information, providing strong support for subsequent network debugging and optimization. At the same time, the process also has certain versatility and scalability, and can be applied to other similar text processing and knowledge mining tasks.

[0076] In an alternative embodiment, the derivative semantic mining network to be debugged is used to perform semantic mining on each involved transaction monitoring log to obtain the semantics of the involved transaction behavior corresponding to each online transaction monitoring log sample, including: performing semantic mining on each involved transaction monitoring log by the derivative semantic mining network to be debugged; and performing transaction scenario quantification on each text semantic coding mined by the derivative semantic mining network to be debugged to obtain the semantics of the involved transaction behavior corresponding to each online transaction monitoring log sample.

[0077] The core task of this embodiment is to conduct in-depth analysis of each involved transaction monitoring log through the derivative semantic mining network to be debugged, so as to extract the involved transaction behavior semantics closely related to each online transaction monitoring log sample. In this way, it not only covers the process of semantic mining using the derivative semantic mining network, but also involves the important step of quantifying the transaction scenario of the mined text semantic coding, so as to ensure that the obtained involved transaction behavior semantics can accurately reflect the actual characteristics of the transaction scenario.

[0078] First, the online transaction task scheduling device needs to conduct detailed semantic mining on each involved transaction monitoring log through the derivative semantic mining network to be debugged. This step is the basis for extracting the semantics of involved transaction behaviors. It requires the online transaction task scheduling device to use the powerful processing power of the network to mine deep semantic information from the text content of the involved logs. During the mining process, the network will conduct a comprehensive analysis of the vocabulary, sentence patterns, contextual relationships, etc. in the logs, so as to capture the key semantic features related to the transaction behavior. These features may include transaction type, transaction amount, transaction time, transaction parties, etc., which together constitute the core content of the semantics of involved transaction behaviors.

[0079] After completing the semantic mining, the online transaction task scheduling device obtains the text semantic coding of each transaction monitoring log involved. However, these codes are still abstract and high-dimensional, and it is difficult to directly use them to describe specific transaction scenarios. Therefore, the online transaction task scheduling device needs to proceed to the next step: quantify the transaction scenarios of the text semantic coding.

[0080] Transaction scenario quantification is a process of converting abstract semantic coding into specific numerical features. It requires the online transaction task scheduling device to design a set of reasonable quantitative indicators based on the actual needs of the transaction scenario, and apply these indicators to the text semantic coding, so as to obtain numerical results that can accurately reflect the characteristics of the transaction scenario. For example, the online transaction task scheduling device can design quantitative indicators such as transaction amount, transaction frequency, and transaction time distribution, and obtain the specific values ​​of these indicators by analyzing and processing the text semantic coding.

[0081] Take the transaction amount as an example. For example, the online transaction task scheduling device extracts the vocabulary and sentence patterns related to the transaction amount by analyzing the text semantic coding, and judges that the transaction amount belongs to the "medium level" based on this. In order to convert this abstract judgment into a specific numerical feature, the online transaction task scheduling device can set a numerical range of the transaction amount, such as 1,000 yuan to 5,000 yuan as "medium level". In this way, the online transaction task scheduling device converts the abstract semantic judgment "medium level" into a specific numerical range of "1,000 yuan to 5,000 yuan". In actual applications, the online transaction task scheduling device can further subdivide or adjust this numerical range as needed.

[0082] Similarly, for quantitative indicators such as transaction frequency and transaction time distribution, the online transaction task scheduling device can also use similar methods to process. For example, for transaction frequency, the online transaction task scheduling device can set a specific numerical range to represent "higher" or "lower" frequency; for transaction time distribution, the online transaction task scheduling device can count the number of transactions in different time periods and convert these numbers into specific numerical features.

[0083] After quantifying the transaction scenarios, the online transaction task scheduling device obtains the semantics of the transaction behaviors involved corresponding to each online transaction monitoring log sample. These semantics are presented in the form of numerical features, which can accurately and specifically describe the actual characteristics of the transaction scenario. For example, for a certain online transaction monitoring log sample, the online transaction task scheduling device may obtain the following semantics of the transaction behaviors involved: the transaction amount is 1,000 to 5,000 yuan (medium level), the transaction frequency is 5 to 10 times a day (high), and the transaction time is mainly distributed during the day (8:00 to 18:00), etc. These specific numerical features provide strong support for the subsequent transaction behavior analysis and task scheduling of the online transaction task scheduling device.

[0084] Therefore, the derivative semantic mining network to be debugged performs semantic mining on each involved transaction monitoring log, and performs transaction scenario quantification on the mined text semantic coding, and the online transaction task scheduling device obtains the semantics of the involved transaction behavior corresponding to each online transaction monitoring log sample. In this way, not only the powerful processing power of the derivative semantic mining network is fully utilized, but also the abstract semantic coding is converted into specific numerical features through transaction scenario quantification, thereby providing an accurate and specific transaction scenario description for the online transaction task scheduling device.

[0085] In an optional design idea, the initial transaction behavior text semantics and the at least one derived transaction behavior text semantics are projected into the target semantic knowledge relationship network to obtain the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection, including: the residual feature processing branch in the multi-head attention network is used to quantify the transaction scenario of the initial transaction behavior text semantics and the at least one derived transaction behavior text semantics to obtain the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection. Then, the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection are used to obtain the transaction behavior text linkage decision semantics, including: the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection are used to obtain the transaction behavior text linkage decision semantics through the semantic integration branch in the multi-head attention network.

[0086] It can be understood that this design idea is committed to effectively integrating the semantics of the initial transaction behavior text and at least one derived transaction behavior text into the target semantic knowledge relationship network, in order to obtain a richer and more accurate transaction behavior text semantic representation. This process not only involves an in-depth understanding and quantification of the initial and derived transaction behavior text semantics, but also covers the comprehensive use of these semantics to form transaction behavior text linkage decision semantics with decision-making guidance significance.

[0087] First, the online transaction task scheduling device uses the residual feature processing branch in the multi-head attention network to perform fine transaction scenario quantification on the initial transaction behavior text semantics and at least one derived transaction behavior text semantics. This step is the core, which requires the online transaction task scheduling device to convert the abstract text semantics into a numerical representation that can reflect the actual characteristics of the transaction scenario. In order to achieve this goal, the online transaction task scheduling device designs multiple quantitative indicators, such as transaction amount, transaction frequency, transaction time distribution, etc. These indicators can comprehensively cover the key characteristics of transaction behavior.

[0088] Taking the transaction amount as an example, the online transaction task scheduling device converts abstract amount descriptions (such as "large amount" and "small amount") into specific numerical ranges by analyzing the words and sentences related to the transaction amount in the text semantics and combining them with external transaction amount data. For example, "large amount" may be quantified as "more than 10,000 yuan" and "small amount" may be quantified as "less than 1,000 yuan". Such quantification processing enables the online transaction task scheduling device to more accurately grasp the amount characteristics of transaction behavior.

[0089] At the same time, the online transaction task scheduling device also adopts similar processing methods for quantitative indicators such as transaction frequency and transaction time distribution. By analyzing the relevant information in the text semantics and combining it with external transaction data, the online transaction task scheduling device converts these abstract text descriptions into specific numerical features. For example, the transaction frequency may be quantified as "occurring 3-5 times a day", and the transaction time distribution may be quantified as "mainly concentrated from 9 am to 5 pm".

[0090] After completing the transaction scenario quantification, the online transaction task scheduling device obtains the initial transaction behavior text semantics and derived transaction behavior text semantics that complete the knowledge projection. These semantics are presented in the form of numerical features, which can more accurately reflect the actual characteristics of the transaction scenario.

[0091] Next, the online trading task scheduling device uses the semantic integration branch in the multi-head attention network to comprehensively utilize the initial trading behavior text semantics and the derived trading behavior text semantics that have completed the knowledge projection to form the trading behavior text linkage decision semantics. This step is the key to generating the final decision semantics, which requires the online trading task scheduling device to organically integrate the semantics of multiple trading behavior texts and extract semantic features that are instructive for decision-making.

[0092] In the semantic integration branch, the online transaction task scheduling device adopts a multi-head attention mechanism, which allows the model to pay attention to other semantic information while processing the semantics of each transaction behavior text, thereby realizing the linkage between semantics. Through the processing of the multi-head attention mechanism, the model can capture the association and difference between the semantics of different transaction behavior texts, and then generate more comprehensive and accurate decision semantics.

[0093] Finally, the online transaction task scheduling device obtains the transaction behavior text linkage decision semantics, which is presented in the form of a numerical feature vector. This feature vector contains multiple dimensions, each of which represents a key feature related to the transaction behavior decision. For example, a possible feature vector dimension includes the size of the transaction amount, the frequency of transactions, whether the transaction time is concentrated, etc. For each dimension, the online transaction task scheduling device gives a specific numerical representation, for example, the transaction amount may be represented as "0.8" (indicating a large amount), the frequency of transactions may be represented as "0.6" (indicating a medium frequency), and whether the transaction time is concentrated may be represented as "0.9" (indicating a high concentration).

[0094] The above example is not only intuitive, but also convenient for subsequent calculations and processing. The online transaction task scheduling device can use this feature vector as input and apply it to various machine learning or deep learning models to conduct further transaction behavior analysis, prediction or decision-making. For example, the online transaction task scheduling device can use this feature vector to train a transaction behavior classifier to determine whether a certain transaction behavior is an abnormal behavior; or the online transaction task scheduling device can combine this feature vector with other transaction data to train a transaction amount prediction model to predict the size of future transaction amounts.

[0095] Most importantly, the transaction behavior text-linked decision semantics plays a vital role in the online transaction task scheduling device. It not only enriches the text semantic representation of transaction behavior, but also provides strong support for queue optimization and task scheduling functions.

[0096] In terms of queue optimization, the transaction behavior text linkage decision semantics, as a form of numerical feature vectors, can intuitively reflect the key characteristics of transaction behavior, such as transaction amount, transaction frequency, whether the transaction time is concentrated, etc. These characteristics are of great significance for determining the priority and scheduling order of transactions. For example, transactions with larger amounts may require higher priority to ensure timely processing and reduce potential risks; while transactions with concentrated time may require the scheduling system to optimize resources during peak hours to cope with the surge in transaction volume.

[0097] By leveraging transaction behavior text to link decision semantics, the online transaction task scheduling device can more accurately determine the urgency and importance of transactions, thereby optimizing the task arrangement in the queue. This can not only improve transaction processing efficiency, but also ensure that key transactions are processed in a timely manner, further improving the overall transaction quality and user experience.

[0098] In terms of task scheduling, transaction behavior text linkage decision semantics also plays an important role. Task scheduling refers to the reasonable arrangement and allocation of tasks according to the system resources and load conditions to achieve optimal system performance and resource utilization. By considering the key characteristics of transaction behavior, such as transaction frequency and transaction time distribution, the online transaction task scheduling device can perform task allocation and scheduling more intelligently.

[0099] For example, during peak trading periods, the system may face resource constraints. At this time, by using transaction behavior text linkage decision semantics, the scheduling device can prioritize tasks with large transaction amounts and high frequency to ensure that key businesses are not affected. At the same time, for tasks with relatively scattered transaction times, the scheduling device can process them during relatively idle resource periods to achieve balanced resource utilization.

[0100] In summary, transaction behavior text linkage decision semantics provides important support and guidance for the queue optimization and task scheduling functions of online transaction task scheduling devices. By deeply exploring and utilizing the key features of transaction behavior text semantics, the scheduling device can more accurately judge the priority and importance of transactions, optimize queue arrangement, and realize intelligent task allocation and scheduling. This not only improves the processing efficiency and quality of transactions, but also ensures that key businesses are processed in a timely manner, further improving the user's transaction experience and satisfaction.

[0101] In some optional design ideas, the debugging steps of the multi-head attention network include: obtaining a first initial transaction behavior text semantic sample, wherein the first initial transaction behavior text semantic sample is obtained by semantic mining of historical online transaction monitoring logs under historical online transaction projects; obtaining a second initial transaction behavior text semantic sample, wherein the second initial transaction behavior text semantic sample is obtained by semantic mining of online transaction monitoring logs with the same transaction scenario keywords as the historical online transaction monitoring logs under the historical online transaction projects; obtaining at least one first derivative transaction behavior text semantic sample that is related to the first initial transaction behavior text semantic sample, and obtaining at least one first derivative transaction behavior text semantic sample that is related to the second initial transaction behavior text semantic sample. a second derivative transaction behavior text semantic sample; inputting the first initial transaction behavior text semantic sample and the at least one first derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and outputting the first transaction behavior text semantic sample; inputting the second initial transaction behavior text semantic sample and the at least one second derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and outputting the second transaction behavior text semantic sample; generating an attention training error based on the difference between the first transaction behavior text semantic sample and the second transaction behavior text semantic sample; performing multiple rounds of debugging on the multi-head attention network to be debugged based on the attention training error to obtain a debugged multi-head attention network.

[0102] It can be understood that the debugging steps of the multi-head attention network can ensure that the network can effectively process and integrate the semantics of transaction behavior text. The debugging steps of the multi-head attention network not only involve in-depth mining of the semantics of the initial transaction behavior text, but also include a detailed analysis of the semantics of the derived transaction behavior text, and multiple rounds of debugging of the network based on these semantic samples.

[0103] First, the online transaction task scheduling device needs to obtain the first initial transaction behavior text semantic sample. This step is achieved by conducting in-depth semantic mining on the historical online transaction monitoring logs under the historical online transaction project. The historical online transaction monitoring logs record various details of the transaction, including transaction time, transaction amount, transaction parties, etc. Through semantic mining technology, the online transaction task scheduling device can extract text semantics related to transaction behavior from these logs to form the first initial transaction behavior text semantic sample. This sample is a text representation of historical transaction behavior, which contains key information of the transaction.

[0104] Next, the online transaction task scheduling device obtains the second initial transaction behavior text semantic sample. The method of obtaining this sample is slightly different from the first sample. The online transaction task scheduling device obtains the second initial transaction behavior text semantic sample by searching for online transaction monitoring logs with the same transaction scenario keywords as the historical online transaction monitoring logs under the historical online transaction project, and performing semantic mining on these logs. This design is to ensure that the online transaction task scheduling device has two initial transaction behavior text semantic samples with similar transaction scenarios but possibly different transaction details, so as to facilitate more comprehensive debugging of the network later.

[0105] After obtaining the initial transaction behavior text semantic samples, the online transaction task scheduling device also needs to obtain the derivative transaction behavior text semantic samples related to these samples. For the first initial transaction behavior text semantic sample, the online transaction task scheduling device finds at least one first derivative transaction behavior text semantic sample related to it; similarly, for the second initial transaction behavior text semantic sample, the online transaction task scheduling device also finds at least one second derivative transaction behavior text semantic sample related to it. These derivative transaction behavior text semantic samples are further supplements and extensions of the initial transaction behavior text semantic samples, and they provide more information about the transaction behavior.

[0106] Then, the online transaction task scheduling device inputs the first initial transaction behavior text semantic sample and at least one first derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and the network will output the first transaction behavior text semantic sample. Similarly, the online transaction task scheduling device also inputs the second initial transaction behavior text semantic sample and at least one second derivative transaction behavior text semantic sample into the network, and the network will output the second transaction behavior text semantic sample. These two output samples are the results obtained after the network processes and integrates the input samples, and they reflect the network's understanding and processing capabilities of the transaction behavior text semantics.

[0107] Next, the online transaction task scheduling device generates an attention training error based on the difference between the first transaction behavior text semantic sample and the second transaction behavior text semantic sample. This error is an indicator of how well the network performs when processing two similar but possibly different transaction behavior text semantic samples. If the network can accurately capture the difference between the two samples and give a reasonable output, the attention training error will be small; conversely, if the network is confused or wrong when processing the two samples, the attention training error will be large.

[0108] Finally, the online trading task scheduling device performs multiple rounds of debugging on the multi-head attention network to be debugged based on the attention training error. The purpose of debugging is to optimize the parameters and structure of the network so that it can be more accurate and effective in processing the semantics of the transaction behavior text. Through multiple rounds of debugging, the online trading task scheduling device can gradually reduce the attention training error and improve the performance of the network. Finally, the online trading task scheduling device will obtain a debugged multi-head attention network, which can better understand and process the semantics of the transaction behavior text, and provide more powerful support for subsequent transaction behavior analysis and decision-making.

[0109] In actual applications, the specific value of the attention training error will vary according to different application scenarios and data sets. For example, in a specific online transaction monitoring project, the online transaction task scheduling device may set the attention training error threshold to 0.1. If the attention training error after a round of debugging is less than this threshold, the online transaction task scheduling device will believe that the network has achieved good performance and can stop debugging; if the error is greater than the threshold, the online transaction task scheduling device will need to continue debugging the network until the error is reduced to below the threshold.

[0110] In addition, in order to more intuitively evaluate the performance of the network, the online trading task scheduling device can also combine the attention training error with some other evaluation indicators (such as accuracy, recall, etc.). This can not only evaluate the performance of the network more comprehensively, but also help the online trading task scheduling device better understand the performance of the network when processing different transaction behavior text semantics. For example, the online trading task scheduling device can find that the network has a large error when processing certain specific transaction behavior text semantics, so as to further optimize and improve the network in a targeted manner.

[0111] In this way, through carefully designed debugging steps and reasonable evaluation indicators, the online trading task scheduling device can effectively debug and optimize the multi-head attention network, so that it can better understand and process the semantics of trading behavior text. This not only improves the performance of the network, but also provides stronger support for subsequent trading behavior analysis and decision-making.

[0112] In some further preferred embodiments, the obtaining of the first initial transaction behavior text semantic sample comprises: obtaining a plurality of historical online transaction monitoring logs belonging to historical online transaction items; performing semantic mining on the plurality of historical online transaction monitoring logs belonging to historical online transaction items respectively, and performing feature integration on the mined semantic data to obtain the first initial transaction behavior text semantic sample.

[0113] It is worth noting that obtaining the semantic sample of the first initial transaction behavior text provides a basic and rich semantic representation for subsequent transaction behavior analysis, prediction or decision-making. In order to achieve this goal, this application adopts two main steps: first, obtaining multiple historical online transaction monitoring logs belonging to historical online transaction projects, and second, conducting in-depth semantic mining on these logs and integrating the features of the mined semantic data.

[0114] First, obtaining multiple historical online transaction monitoring logs belonging to historical online transaction items is a critical first step. These monitoring logs record various details of online transaction items, including but not limited to transaction time, transaction amount, transaction party information, transaction status, etc. They are a direct reflection of transaction behavior and contain rich transaction patterns and rules. In order to obtain these logs, it is usually necessary to extract them from the background database of the transaction system to ensure the integrity and accuracy of the data.

[0115] After obtaining a sufficient number of historical online transaction monitoring logs, the next step is to perform semantic mining on these logs. Semantic mining is a complex but crucial process that aims to extract meaningful semantic information from a large amount of text data. In this process, various natural language processing techniques may be used, such as word segmentation, part-of-speech tagging, named entity recognition, etc., to convert the original log text into structured semantic data. For example, for a log containing "User A transferred 1,000 yuan to user B at 3 pm on May 1, 2023", semantic mining can convert it into data containing structured fields such as "transaction time: 3 pm on May 1, 2023", "transaction amount: 1,000 yuan", "transaction parties: user A, user B", and "transaction type: transfer".

[0116] However, it is not enough to just obtain structured semantic data, because each log may only reflect one aspect of the transaction, while a complete transaction behavior often involves multiple aspects. Therefore, it is necessary to integrate the features of the mined semantic data to obtain a comprehensive text semantic representation of the transaction behavior. Feature integration is a process of fusing multiple single features into a comprehensive feature, which can be achieved through various feature fusion technologies, such as feature splicing, feature weighting, feature selection, etc. In this process, the correlation, complementarity and redundancy between different features may be taken into account to ensure that the integrated features can reflect the full picture of the transaction behavior to the greatest extent possible.

[0117] Take feature concatenation as an example. For example, if the online transaction task scheduling device has mined the two features of "transaction amount" and "transaction type" from the log, the process of feature integration is to concatenate these two features together to form a new feature vector. For example, for a transaction with an amount of 1,000 yuan and a transfer type, its feature vector can be expressed as [1000, transfer]. Of course, actual transaction behavior may involve more features, such as transaction time, transaction parties, transaction status, etc. These features can be concatenated in a similar way to form a richer feature vector.

[0118] In addition to feature concatenation, feature weighting is also a commonly used feature integration method. In some cases, different features may have different importance for the representation of transaction behavior. For example, when judging whether a transaction is an abnormal transaction, "transaction amount" may be more important than "transaction time". Therefore, in the process of feature integration, different weights can be assigned to different features to highlight their importance. For example, for the above feature vector [1000, transfer], the online transaction task scheduling device can assign it a weight [0.7, 0.3] to indicate that "transaction amount" is more important than "transaction type".

[0119] After completing the feature integration, the online transaction task scheduling device obtains the first initial transaction behavior text semantic sample. This sample is a comprehensive transaction behavior text semantic representation, which covers multiple aspects of the transaction behavior and is presented in the form of a numerical feature vector. This sample is not only intuitive, but also convenient for subsequent calculation and processing. For example, the online transaction task scheduling device can use this feature vector as input and apply it to various machine learning or deep learning models for further transaction behavior analysis, prediction or decision-making.

[0120] In practical applications, the acquisition of the first initial transaction behavior text semantic sample is only the first step in the entire transaction behavior analysis process. Next, the online transaction task scheduling device may need to further process and handle this sample to adapt to different analysis tasks. For example, if the online transaction task scheduling device wants to determine whether a transaction is an abnormal transaction, the online transaction task scheduling device can use the first initial transaction behavior text semantic sample as input to train an abnormal transaction detection model. This model can learn the differences in semantic features between normal transactions and abnormal transactions, and classify new transaction behaviors accordingly.

[0121] In addition, if the online transaction task scheduling device wants to predict the future transaction amount, the online transaction task scheduling device can combine the first initial transaction behavior text semantic sample with other transaction data (such as historical transaction amount, transaction time, etc.) to train a transaction amount prediction model. This model can learn the complex relationship between the transaction amount and various factors, and predict the future transaction amount based on this.

[0122] In this way, by obtaining historical online transaction monitoring logs, performing semantic mining and feature integration, the first initial transaction behavior text semantic sample is successfully obtained. This sample provides a basic and rich semantic representation for subsequent transaction behavior analysis, prediction or decision-making, and has broad application prospects and practical value. In practical applications, the online transaction task scheduling device can further process and handle this sample according to specific analysis tasks and requirements to obtain more accurate and useful analysis results.

[0123] In an exemplary embodiment, the multi-head attention network includes a plurality of residual feature processing branches with the same neural network configuration; the inputting of the first initial transaction behavior text semantic sample and the at least one first derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and outputting the first transaction behavior text semantic sample, comprises: inputting the first initial transaction behavior text semantic sample into the first residual feature processing branch to obtain a first behavior semantic residual vector sample; respectively inputting the at least one first derivative transaction behavior text semantic sample into the respective corresponding second residual feature processing branches to obtain a second behavior semantic residual vector sample; based on the first behavior semantic residual vector sample and the second behavior semantic residual The method comprises: inputting the second initial transaction behavior text semantic sample to the third residual feature processing branch to obtain a third behavior semantic residual vector sample; respectively inputting the at least one second derivative transaction behavior text semantic sample to the respective corresponding fourth residual feature processing branches to obtain a fourth behavior semantic residual vector sample; and obtaining a second transaction behavior text semantic sample based on the third behavior semantic residual vector sample and the fourth behavior semantic residual vector sample.

[0124] It can be understood that this embodiment involves how to effectively merge the initial transaction behavior text semantic sample with at least one derived transaction behavior text semantic sample to output a comprehensive and accurate transaction behavior text semantic sample. In order to complete this task, the online transaction task scheduling device introduces a multi-head attention network, which includes multiple residual feature processing branches with the same neural network configuration. This design enables the network to process multiple transaction behavior text semantic samples at the same time and extract key residual features from them, thereby generating a richer and more accurate transaction behavior text semantic representation.

[0125] First, the online transaction task scheduling device inputs the first initial transaction behavior text semantic sample into the first residual feature processing branch in the multi-head attention network. This branch will perform in-depth semantic analysis and feature extraction on the initial transaction behavior text semantic sample, and finally obtain a first behavior semantic residual vector sample. This residual vector sample represents the key features and potential information in the initial transaction behavior text semantic sample, and is an important basis for subsequent semantic fusion.

[0126] At the same time, the online transaction task scheduling device inputs at least one first derivative transaction behavior text semantic sample into the corresponding second residual feature processing branch. These branches perform similar semantic analysis and feature extraction on the derivative transaction behavior text semantic sample to obtain a series of second behavior semantic residual vector samples. These residual vector samples represent the key features and potential information in each derivative transaction behavior text semantic sample, and also provide important data support for subsequent semantic fusion.

[0127] After obtaining the first behavior semantic residual vector sample and the second behavior semantic residual vector sample, the online transaction task scheduling device performs semantic fusion based on these residual vector samples to generate the first transaction behavior text semantic sample. This step is the core, which requires the online transaction task scheduling device to organically fuse the key features and potential information in multiple residual vector samples to form a comprehensive and accurate transaction behavior text semantic representation. In order to achieve this goal, the online transaction task scheduling device may adopt various feature fusion technologies, such as feature splicing, feature weighting, feature selection, etc., to ensure that the fused semantic representation can maximize the overall picture of the transaction behavior.

[0128] After completing the generation of the first transaction behavior text semantic sample, the online transaction task scheduling device then inputs the second initial transaction behavior text semantic sample into the third residual feature processing branch in the multi-head attention network. This branch performs similar semantic analysis and feature extraction on the second initial transaction behavior text semantic sample to obtain a third behavior semantic residual vector sample. Similarly, the online transaction task scheduling device inputs at least one second derivative transaction behavior text semantic sample into the respective corresponding fourth residual feature processing branches to obtain a series of fourth behavior semantic residual vector samples.

[0129] Finally, the online transaction task scheduling device performs semantic fusion based on the third behavior semantic residual vector sample and the fourth behavior semantic residual vector sample to generate a second transaction behavior text semantic sample. This step is similar to generating the first transaction behavior text semantic sample, and also requires the online transaction task scheduling device to organically fuse the key features and potential information in multiple residual vector samples. Through this step, the online transaction task scheduling device can obtain an equally comprehensive and accurate transaction behavior text semantic representation, which provides strong support for subsequent transaction behavior analysis, prediction or decision-making.

[0130] In practical applications, the number of residual feature processing branches in the multi-head attention network can be flexibly configured according to actual needs. For example, if the online trading task scheduling device only needs to process one initial trading behavior text semantic sample and one derived trading behavior text semantic sample, then the online trading task scheduling device can only configure two residual feature processing branches. If the online trading task scheduling device needs to process more trading behavior text semantic samples, then the online trading task scheduling device can increase the number of residual feature processing branches accordingly to meet actual needs.

[0131] In addition, it is worth noting that the neural network configurations in the residual feature processing branches are the same, which means they have the same network structure and parameter settings. This design enables each residual feature processing branch to perform semantic analysis and feature extraction on the input transaction behavior text semantic samples in the same way, thus ensuring that the extracted residual vector samples are consistent and comparable.

[0132] In the residual feature processing branch, the online trading task scheduling device may use various deep learning techniques to extract key features and potential information from the semantic samples of the trading behavior text. For example, the online trading task scheduling device may use a convolutional neural network (CNN) to capture local features in the text, such as words and phrases; or use a recurrent neural network (RNN) to capture temporal features in the text, such as the order of trading behaviors. Regardless of the technology used, the goal of the online trading task scheduling device is to extract features and information that can accurately reflect the essence of the trading behavior.

[0133] In addition to deep learning technology, the online trading task scheduling device can also combine traditional natural language processing technology to further enrich the representation of the semantic samples of the trading behavior text. For example, the online trading task scheduling device can use word segmentation technology to segment the text into vocabulary units, and then use part-of-speech tagging technology to assign a part-of-speech tag (such as noun, verb, etc.) to each vocabulary. This information can be incorporated into the residual vector sample as an additional feature to enhance its representation ability.

[0134] The first transaction behavior text semantic sample and the second transaction behavior text semantic sample finally obtained are two comprehensive and accurate transaction behavior text semantic representations. They are presented in the form of numerical feature vectors, which contain multiple dimensions, each of which represents a key feature related to transaction behavior decision-making. For example, a possible feature vector dimension includes the size of the transaction amount, the frequency of transactions, whether the transaction time is concentrated, etc. For each dimension, the online transaction task scheduling device gives a specific numerical representation, such as the transaction amount may be represented as "0.8" (indicating a large amount), the frequency of transactions may be represented as "0.6" (indicating a medium frequency), and whether the transaction time is concentrated may be represented as "0.9" (indicating a high concentration).

[0135] In this way, by introducing a multi-head attention network and cleverly using its multiple neural networks to configure the same residual feature processing branch, the effective fusion of the initial transaction behavior text semantic sample and the derived transaction behavior text semantic sample is successfully achieved. The generated first transaction behavior text semantic sample and the second transaction behavior text semantic sample provide a comprehensive and accurate semantic representation for subsequent transaction behavior analysis, scheduling or decision-making.

[0136] In an optional embodiment, the method also includes: migrating the transaction behavior text linkage decision semantics to a semantic disaster recovery database; the semantic disaster recovery database is used to record the transaction behavior text semantics obtained by performing a semantic update operation on the initial transaction behavior text semantics mined by different online transaction task scheduling networks.

[0137] Based on this, the method also includes: obtaining the current transaction behavior text semantics under any online transaction project, wherein the current transaction behavior text semantics is obtained by semantic mining of the selected online transaction monitoring log by the corresponding online transaction task scheduling network; performing semantic matching in the semantic disaster recovery database according to the current transaction behavior text semantics to obtain a semantic matching result, and the semantic matching result is used for queue optimization and task scheduling.

[0138] Furthermore, each transaction behavior text semantic in the semantic disaster recovery database is respectively provided with a priori feature annotation; the method also includes: when the semantic matching result indicates the existence of transaction behavior text semantics that meet the semantic commonality requirements, determining the target transaction behavior text semantics based on the semantic matching result; based on the priori feature annotation of the target transaction behavior text semantics, performing a report query on the queue task optimization scheduling of the target transaction behavior text semantics to determine the queue task optimization scheduling mode to which the target transaction behavior text semantics belongs.

[0139] It can be understood that this application not only covers the basic transaction behavior text semantic processing, but also further involves migrating these processed semantic information to a special semantic disaster recovery database. The establishment of this database is intended to record and store the transaction behavior text semantics obtained after the update operation of the initial transaction behavior text semantics mined in different online transaction task scheduling networks. In short, it is a "warehouse" that backs up and stores the processed and updated transaction behavior text semantics, ensuring that these valuable data resources will not be lost due to various unexpected situations, and providing solid data support for subsequent semantic analysis and decision-making.

[0140] Based on this, the present application has been further expanded. It can obtain the text semantics of the current transaction behavior under any online transaction project. These semantics are obtained by the corresponding online transaction task scheduling network through in-depth semantic mining of the selected online transaction monitoring log. This process actually converts the text records in the transaction process into machine-understandable semantic representations, which makes subsequent analysis and optimization possible.

[0141] Next, the semantics of the current transaction behavior text are used to perform semantic matching in the semantic disaster recovery database mentioned above. The purpose of this step is to find the historical transaction behavior text semantics that is most similar or relevant to the current transaction behavior, so as to obtain a semantic matching result. The importance of this result is self-evident, and it will become the key basis for subsequent queue optimization and task scheduling. Through semantic matching, the online transaction task scheduling device can better understand the position and role of the current transaction behavior in the overall transaction model, and then make more reasonable and efficient scheduling decisions.

[0142] Furthermore, in order to improve the effect of semantic matching and application, the semantics of each transaction behavior text in the semantic disaster recovery database are carefully set with prior feature annotations. These annotations are additional descriptions and annotations of the transaction behavior text semantics, which may cover multiple dimensions such as the type, scale, frequency, and time distribution of the transaction. When the semantic matching results show that there are transaction behavior text semantics that meet the semantic commonality requirements, the method will determine a target transaction behavior text semantics based on the results. Subsequently, using the prior feature annotations of this target semantics, the method will perform a series of report queries on queue task optimization scheduling. The purpose of this step is to deeply explore the queue task optimization scheduling mode to which the target transaction behavior text semantics belongs, that is, to find the scheduling strategy and execution plan that best suits this type of transaction behavior.

[0143] In this way, not only the effective management and utilization of transaction behavior text semantics are achieved, but also these semantic information are closely integrated with the actual transaction task scheduling and optimization. First, through the establishment of a semantic disaster recovery database and semantic migration operations, the safe storage and effective backup of transaction behavior text semantics are ensured. Secondly, by obtaining the current transaction behavior text semantics and performing semantic matching, the learning and reference of historical transaction patterns are achieved. Finally, by using prior feature annotations for report query and scheduling mode determination, more refined and personalized optimization suggestions are provided for the actual transaction task execution. This series of steps together constitutes a complete and efficient transaction behavior text semantic processing and application framework, which provides strong support for the continuous optimization and performance improvement of online trading projects.

[0144] In some independent embodiments, the method further includes: in response to a data optimization instruction, performing data optimization processing on the semantic disaster recovery database.

[0145] It can be understood that this application not only covers the migration, matching and application of the semantics of transaction behavior text, but also further involves the data optimization processing of the semantic disaster recovery database itself. The design of this extended function is to ensure that the data in the semantic disaster recovery database is always kept up to date and accurate, thereby providing more reliable data support for the analysis and decision-making of transaction behavior.

[0146] First, the present application includes migrating the transaction behavior text linkage decision semantics to the semantic disaster tolerance database. This step is the basis for building and optimizing the semantic disaster tolerance database. Through the migration operation, the transaction behavior text semantics obtained after the semantic update of the initial transaction behavior text semantics mined by different online transaction task scheduling networks are recorded and stored in the semantic disaster tolerance database. In this way, the database becomes a "live" data source, constantly receiving and storing the latest transaction behavior text semantic information. Based on this, the present application further includes obtaining the current transaction behavior text semantics under any online transaction project, and performing semantic matching in the semantic disaster tolerance database based on this semantics to obtain the semantic matching result. This step is the key to transaction behavior analysis and decision-making. Through semantic matching, the historical transaction behavior text semantics that are most similar or relevant to the current transaction behavior can be found, thereby providing a strong basis for subsequent queue optimization and task scheduling. This process actually converts the text records in the transaction process into machine-understandable semantic representations, and through comparison and analysis with historical data, digs out the laws and patterns contained therein. Furthermore, in order to improve the effect of semantic matching and application, the semantics of each transaction behavior text in the semantic disaster tolerance database are carefully set with prior feature annotations. These annotations are additional descriptions and annotations of the transaction behavior text semantics, which may cover multiple dimensions such as the type, scale, frequency, and time distribution of the transaction. When the semantic matching result indicates that there is a transaction behavior text semantics that meets the semantic commonality requirements, the method will determine a target transaction behavior text semantics based on this result, and use the prior feature annotations of this target semantics to perform a report query on the queue task optimization scheduling. The purpose of this step is to deeply explore the queue task optimization scheduling mode to which the target transaction behavior text semantics belongs, that is, to find the scheduling strategy and execution plan that best suits this type of transaction behavior.

[0147] However, relying solely on the above steps is not enough to ensure the long-term effectiveness and accuracy of the semantic disaster recovery database. As time goes by and the transaction environment changes, the data in the database may gradually become outdated or inaccurate. Therefore, it is particularly important to perform data optimization processing on the semantic disaster recovery database in response to the data optimization instruction.

[0148] Data optimization processing involves comprehensive review, update and integration of the semantics of transaction behavior text stored in the database. First, it is necessary to conduct a comprehensive review of the data in the database to identify those data that are outdated, inaccurate or no longer representative. This step is the basis of data optimization processing, which requires the online transaction task scheduling device to conduct a detailed review and analysis of each piece of data in the database to ensure the accuracy and validity of the data. Secondly, for those outdated or inaccurate data that have been identified, timely update or deletion operations are required. This step is the core of data optimization processing, which is directly related to the quality and accuracy of the database. When updating data, the online transaction task scheduling device needs to ensure that the new data is accurate, reliable and representative; when deleting data, the online transaction task scheduling device needs to ensure that this operation will not have an adverse effect on the overall structure and integrity of the database. Finally, after completing the review and update of the data, the data in the database needs to be integrated and optimized. This step is the finishing work of data optimization processing, which aims to improve the performance and query efficiency of the database. When integrating data, the online transaction task scheduling device needs to consider factors such as data relevance, consistency, and redundancy; when optimizing data, the online transaction task scheduling device needs to adopt appropriate data structures and algorithms to improve query efficiency and response speed. Through the above data optimization processing steps, the online transaction task scheduling device can ensure that the data in the semantic disaster recovery database is always kept up to date and accurate. This not only helps to improve the accuracy and reliability of transaction behavior analysis and decision-making, but also helps to improve the performance and stability of the entire transaction system. Because an accurate and reliable semantic disaster recovery database can provide more accurate data support and decision-making basis for the transaction system, thereby helping the transaction system to better cope with various complex and changing transaction environments.

[0149] It can be seen that this application not only covers the migration, matching and application of the semantics of transaction behavior text, but also further involves the data optimization processing of the semantic disaster recovery database itself. This extended function is designed to ensure that the data in the semantic disaster recovery database is always kept up to date and accurate, thereby providing more reliable data support for the analysis and decision-making of transaction behaviors. Through comprehensive data review, update and integration optimization processing steps, the online transaction task scheduling device can ensure the quality and accuracy of the database, thereby improving the performance and stability of the entire transaction system.

[0150] In some other optional embodiments, the method further includes: determining a task scheduling migration strategy according to a queue task optimization scheduling mode to which the text semantics of the target transaction behavior belongs.

[0151] It can be understood that after determining the queue task optimization scheduling mode to which the target transaction behavior text semantics belongs, the next key step is to carefully formulate a task scheduling migration strategy based on this specific mode. This link is the essence of this application, and it has a decisive impact on the efficiency and response time of task scheduling. In the process of formulating the task scheduling migration strategy, the online transaction task scheduling device must comprehensively and deeply consider the uniqueness and requirements of the queue task optimization scheduling mode, while closely combining the actual situation and urgent needs of the current transaction system.

[0152] The first task is to clarify the basic strategies and guiding principles of task scheduling according to the queue task optimization scheduling mode corresponding to the text semantics of the target transaction behavior. It is worth noting that different queue task optimization scheduling modes often correspond to completely different scheduling strategies and principles. For example, some modes may place special emphasis on real-time processing and rapid response of tasks, while some modes may pay more attention to stable execution and high reliability of tasks. Therefore, when formulating task scheduling migration strategies, online trading task scheduling devices must take these factors into consideration to ensure that the formulated strategies can achieve perfect matching and consistency with the selected mode.

[0153] Next, the online trading task scheduling device needs to closely combine the actual situation and urgent needs of the current trading system to formulate and optimize the task scheduling migration strategy in a specific and detailed manner. This covers a series of key operations, such as determining the priority and execution order of tasks, allocating appropriate resources and servers to tasks, and setting reasonable timeout and retry mechanisms. In the process of formulating strategies, the online trading task scheduling device also needs to comprehensively consider the load and capacity status of the system to ensure that the formulated strategy will not have any adverse impact on the stability and performance of the system during implementation.

[0154] Finally, the online trading task scheduling device must conduct sufficient and rigorous testing and verification of the formulated task scheduling migration strategy to ensure its feasibility and effectiveness in the actual environment. This step is extremely important because only through actual testing and verification can the online trading task scheduling device accurately discover the problems and deficiencies in the strategy and make corrections and optimizations in a timely manner. During the testing and verification process, the online trading task scheduling device needs to pay close attention to multiple key indicators such as the strategy's response time, scheduling efficiency, and resource utilization to ensure that the formulated strategy can fully meet actual needs and requirements.

[0155] Designed in this way, this application can develop a task scheduling migration strategy with broad versatility based on the queue task optimization scheduling mode to which the text semantics of the target transaction behavior belongs. The task scheduling migration strategy can not only significantly reduce the scheduling response time and improve the efficiency and accuracy of task scheduling, but also provide a more intelligent and efficient solution for the task scheduling of online trading projects. At the same time, because this strategy has a certain degree of versatility, it can be widely used and promoted in different trading systems and scenarios, bringing significant improvements in performance and user experience to more online trading projects.

[0156] Further, Figure 2 Schematic diagram of the structure of an online transaction task scheduling device 200 provided in an embodiment of the present application. Figure 2 The online transaction task scheduling device 200 shown includes a processor 210. The processor 210 can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0157] Alternatively, if Figure 2 As shown, the online transaction task scheduling device 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0158] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0159] Alternatively, if Figure 2 As shown, the online transaction task scheduling device 200 may further include a transceiver 220. The processor 210 may control the transceiver 220 to interact with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices.

[0160] Optionally, the online transaction task scheduling device 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as processing modules) or devices deployed with the storage engine in the various methods of the embodiments of the present application, which will not be described here for brevity.

[0161] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities.

[0162] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0163] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0164] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0165] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0166] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the present application, all of which are within the protection of the present application.

Claims

1. A multi-queue optimized online transaction task scheduling method, characterized in that: The method is implemented by an online transaction task scheduling device, and the method includes: Acquire the semantics of the initial transaction behavior text for the queue task optimization scheduling to be performed; the semantics of the initial transaction behavior text is obtained by the online transaction task scheduling network through semantic mining of the online transaction monitoring log under the online transaction project matching the online transaction task scheduling network; Acquire at least one derivative transaction behavior text semantics associated with the initial transaction behavior text semantics; the derivative transaction behavior text semantics is a non-constrained text semantics, and is obtained by repeating the online transaction monitoring log to which the initial transaction behavior text semantics belongs and then performing semantic mining; Projecting the initial transaction behavior text semantics and the at least one derived transaction behavior text semantics into the target semantic knowledge relationship network respectively, to obtain the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection; The initial transaction behavior text semantics and the derived transaction behavior text semantics after the knowledge projection are used to obtain the transaction behavior text linkage decision semantics; the transaction behavior text linkage decision semantics is used to perform online transaction task scheduling; The obtaining of at least one derivative transaction behavior text semantics associated with the initial transaction behavior text semantics includes: Determine the online transaction monitoring log to which the semantics of the initial transaction behavior text belongs, and perform at least one content retelling operation on the online transaction monitoring log to obtain online transaction monitoring logs with different output states; wherein the content retelling operation rewrites, reorganizes or transforms the original text to generate texts with different output states; Performing semantic mining on the online transaction monitoring logs of the different output states respectively to obtain at least one derivative transaction behavior text semantics; The semantics of the derivative transaction behavior text is extracted by a debugged derivative semantic mining network, and the debugging steps of the derivative semantic mining network include: Get multiple online transaction monitoring log samples; For each online transaction monitoring log sample, at least one content retelling operation is performed to obtain at least one involved transaction monitoring log corresponding to each online transaction monitoring log sample; Through the derivative semantic mining network to be debugged, semantic mining is performed on each involved transaction monitoring log to obtain the semantics of the involved transaction behavior corresponding to each online transaction monitoring log sample; Generate feature-derived training errors based on the semantic commonality coefficients between the semantics of transaction behaviors involved in each online transaction monitoring log sample; The derived semantic mining network to be debugged is debugged for multiple rounds according to the feature-derived training error to obtain a debugged derived semantic mining network.

2. The method according to claim 1, characterized in that The generating of feature-derived training errors according to the semantic commonality coefficients between the semantics of the transaction behaviors involved in each online transaction monitoring log sample includes: For any one of the plurality of online transaction monitoring log samples, determining a first semantic commonality coefficient between semantics of transaction behaviors involved corresponding to the targeted online transaction monitoring log sample; Determine a second semantic commonality coefficient between the semantics of the involved transaction behavior corresponding to the targeted online transaction monitoring log sample and the semantics of the involved transaction behavior corresponding to the remaining online transaction monitoring log samples; A feature-derived training error is generated based on the first semantic commonality coefficient and the second semantic commonality coefficient.

3. The method according to claim 1, characterized in that The derivative semantic mining network to be debugged performs semantic mining on each involved transaction monitoring log to obtain the semantics of the involved transaction behavior corresponding to each online transaction monitoring log sample, including: Semantic mining is performed on each involved transaction monitoring log through the derivative semantic mining network to be debugged; The transaction scenarios are quantified for each text semantic code mined by the derivative semantic mining network to be debugged, so as to obtain the semantics of the involved transaction behaviors corresponding to each online transaction monitoring log sample.

4. The method according to claim 1, characterized in that The projecting the initial transaction behavior text semantics and the at least one derived transaction behavior text semantics into the target semantic knowledge relationship network to obtain the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection, comprises: performing transaction scenario quantification on the initial transaction behavior text semantics and the at least one derived transaction behavior text semantics respectively through the residual feature processing branch in the multi-head attention network to obtain the initial transaction behavior text semantics and the derived transaction behavior text semantics that complete the knowledge projection; The method of obtaining transaction behavior text linkage decision semantics by using the initial transaction behavior text semantics and the derived transaction behavior text semantics after the knowledge projection is completed includes: obtaining transaction behavior text linkage decision semantics by using the initial transaction behavior text semantics and the derived transaction behavior text semantics after the knowledge projection is completed through the semantic integration branch in the multi-head attention network; The debugging steps of the multi-head attention network include: Acquire a first initial transaction behavior text semantic sample, where the first initial transaction behavior text semantic sample is obtained by semantic mining of a historical online transaction monitoring log under a historical online transaction project; Acquire a second initial transaction behavior text semantic sample, where the second initial transaction behavior text semantic sample is obtained by semantic mining an online transaction monitoring log having the same transaction scenario keyword as a historical online transaction monitoring log under the historical online transaction project; Acquire at least one first derivative transaction behavior text semantic sample associated with the first initial transaction behavior text semantic sample, and acquire at least one second derivative transaction behavior text semantic sample associated with the second initial transaction behavior text semantic sample; Inputting the first initial transaction behavior text semantic sample and the at least one first derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and outputting the first transaction behavior text semantic sample; Inputting the second initial transaction behavior text semantic sample and the at least one second derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and outputting the second transaction behavior text semantic sample; generating an attention training error according to a difference between the first transaction behavior text semantic sample and the second transaction behavior text semantic sample; The multi-head attention network to be debugged is debugged for multiple rounds according to the attention training error to obtain a debugged multi-head attention network.

5. The method according to claim 4, characterized in that The obtaining of the first initial transaction behavior text semantic sample includes: Acquire multiple historical online transaction monitoring logs belonging to historical online transaction items; Semantic mining is performed on multiple historical online transaction monitoring logs belonging to historical online transaction projects respectively, and feature integration is performed on the mined semantic data to obtain a first initial transaction behavior text semantic sample.

6. The method according to claim 4, characterized in that The multi-head attention network includes a plurality of residual feature processing branches with the same neural network configuration; the inputting the first initial transaction behavior text semantic sample and the at least one first derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and outputting the first transaction behavior text semantic sample, comprises: Adding the first initial transaction behavior text semantic sample to the first residual feature processing branch to obtain a first behavior semantic residual vector sample; Inputting the at least one first derivative transaction behavior text semantic sample into the respective corresponding second residual feature processing branches to obtain a second behavior semantic residual vector sample; Obtaining a first transaction behavior text semantic sample based on the first behavior semantic residual vector sample and the second behavior semantic residual vector sample; The step of inputting the second initial transaction behavior text semantic sample and the at least one second derivative transaction behavior text semantic sample into the multi-head attention network to be debugged, and outputting the second transaction behavior text semantic sample, comprises: Adding the second initial transaction behavior text semantic sample to the third residual feature processing branch to obtain a third behavior semantic residual vector sample; Inputting the at least one second derivative transaction behavior text semantic sample into the respective corresponding fourth residual feature processing branches to obtain a fourth behavior semantic residual vector sample; Based on the third behavior semantic residual vector sample and the fourth behavior semantic residual vector sample, a second transaction behavior text semantic sample is obtained.

7. The method according to claim 1, characterized in that The method further comprises: Migrating the transaction behavior text linkage decision semantics to a semantic disaster recovery database; the semantic disaster recovery database is used to record the transaction behavior text semantics obtained by performing a semantic update operation on the initial transaction behavior text semantics mined by different online transaction task scheduling networks; The method further includes: obtaining the current transaction behavior text semantics under any online transaction project, wherein the current transaction behavior text semantics is obtained by semantic mining of the selected online transaction monitoring log by the corresponding online transaction task scheduling network; performing semantic matching in the semantic disaster recovery database according to the current transaction behavior text semantics, obtaining a semantic matching result, and using the semantic matching result for queue optimization and task scheduling; Prior feature annotations are set for each transaction behavior text semantic in the semantic disaster recovery database; the method also includes: when the semantic matching result indicates that there are transaction behavior text semantics that meet the semantic commonality requirements, determining the target transaction behavior text semantics according to the semantic matching result; based on the prior feature annotations of the target transaction behavior text semantics, performing a report query on the queue task optimization scheduling of the target transaction behavior text semantics to determine the queue task optimization scheduling mode to which the target transaction behavior text semantics belongs.

8. An online transaction task scheduling device, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 7.

Images