AI Application Center: Methods and Systems for Rapid Application Delivery and Full Lifecycle Management

Abstract

This invention discloses a method and system for rapid delivery and full lifecycle management of AI application centers, belonging to the field of computer software technology. The key technical points include: acquiring user requests; obtaining text vectors and semantic tags based on the user requests; determining the shard corresponding to the user request based on the semantic tags, with the shard obtained according to a preset semantic mapping matrix; and invoking the main contract based on the shard and a preset association matrix to enable the main contract to complete the user request. This invention provides a data foundation for rapid delivery of AI applications through a semantic mapping matrix, and through blockchain sharding, stores related data in the same shard. By using the access order between shards stored in the association matrix, when a user triggers access to relevant data in an AI application, the corresponding shard can be quickly located through the association matrix, and the required data can be quickly loaded from the corresponding shard, ensuring timely processing of user requests.

Description

Technical Field

[0001] This invention relates to the field of computer software technology, and more specifically to a method and system for rapid delivery and full lifecycle management of AI application centers. Background Technology

[0002] In the current era of booming AI applications, the rapid delivery and full lifecycle management methods of AI application centers are becoming increasingly diverse. The rapid delivery and full lifecycle management of AI application centers include stages such as development and building, rapid delivery, and operation and maintenance management. In the development and building stage, it is necessary to integrate heterogeneous data such as training data, configuration data, and third-party interface data. However, traditional methods often rely on manually writing mapping rules to achieve data alignment, which takes a long time and is prone to errors. Furthermore, in the rapid delivery stage, when traditional AI applications are deployed in a multi-node distributed environment, accessing data takes a long time. Therefore, existing technologies have shortcomings. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for rapid delivery and full lifecycle management of AI application centers, which reduces the time for data integration and access to shards through semantic mapping matrix and blockchain sharding design.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] This invention provides a method for rapid application delivery and full lifecycle management of AI application centers, including:

[0006] Get user request;

[0007] The text vector and semantic tags are obtained based on the user request;

[0008] The segment corresponding to the user request is determined based on the semantic tags, and the segment is obtained according to a preset semantic mapping matrix;

[0009] The main contract is invoked based on the fragments and the preset association matrix, so that the main contract can complete the user request.

[0010] As a further improvement of the present invention, the step of obtaining the preset semantic mapping matrix includes:

[0011] Historical data is preprocessed to obtain word units;

[0012] The word elements are converted into word vectors and positional encoding is added to obtain the input information;

[0013] Based on the input information and the preset model, the text vector corresponding to the historical data is obtained;

[0014] The preset semantic mapping matrix is ​​obtained based on the text vectors of the historical data.

[0015] As a further improvement of the present invention, the fragmentation is obtained according to a preset semantic mapping matrix, including:

[0016] An iterative operation is performed based on the preset semantic mapping matrix. The iterative operation includes calculating the similarity between all vectors in the preset semantic mapping matrix and each current center vector, determining the vector corresponding to each current center vector based on the similarity, calculating the mean value corresponding to each current center vector, determining whether the mean value reaches a preset termination condition, and if not, updating the current center vector until the preset termination condition is reached, outputting each current center vector and its corresponding vector to obtain each slice.

[0017] As a further improvement of the present invention, obtaining semantic tags based on the user request includes:

[0018] Based on the user request and the text convolutional neural network model, a global feature vector is obtained;

[0019] The main label is obtained based on the global feature vector and the preset main label library;

[0020] Based on the user request and the named entity recognition model, sub-labels are obtained;

[0021] The semantic tag is obtained based on the main tag and the sub-tag.

[0022] As a further improvement of the present invention, determining the fragment corresponding to the user request based on the semantic tag includes:

[0023] Determine the corresponding decision tree model based on the main label;

[0024] Based on the sub-label and the intermediate node in the decision tree model, determine the corresponding leaf node;

[0025] A preset association matrix is ​​determined based on the leaf nodes;

[0026] The fragment corresponding to the user request is determined based on the preset association matrix.

[0027] As a further improvement of the present invention, the step of obtaining the preset correlation matrix includes:

[0028] Determine the corresponding association rules based on the preset business scenarios;

[0029] Based on the corresponding fragments in the association rules, the preset association matrix corresponding to each scene is obtained.

[0030] As a further improvement of the present invention, the main contract is invoked according to the fragmentation and the preset association matrix, so that the main contract completes the user request, including:

[0031] The access order of each fragment is determined based on the fragmentation and the preset association matrix;

[0032] The main contract is invoked so that it iterates through each shard in turn to complete the user request.

[0033] As a further improvement of the present invention, the main contract is invoked according to the fragmentation and the preset association matrix, so that the main contract completes the user request, including:

[0034] The subcontract to be invoked is determined based on the fragmentation and the preset association matrix;

[0035] The main contract is invoked so that it sends instructions to each of the subcontracts to be invoked, and each subcontract to be invoked obtains the verification result according to the corresponding proxy node and sends it to the main contract so that the main contract can complete the user request.

[0036] As a further improvement of the present invention, each of the subcontracts to be invoked corresponds to at least one shard, each shard corresponds to one proxy node, and the main contract corresponds to one proxy node.

[0037] This invention provides an AI application center application rapid delivery and full lifecycle management system, comprising:

[0038] The acquisition module is used to acquire user requests;

[0039] The identification module obtains text vectors and semantic tags based on the user request;

[0040] The positioning module is used to determine the segment corresponding to the user request based on the semantic tags, wherein the segment is obtained according to a preset semantic mapping matrix;

[0041] The scheduling module invokes the main contract based on the shards and a preset association matrix, so that the main contract can complete the user request.

[0042] In the rapid delivery phase, this invention uses a semantic mapping matrix to semantically encode heterogeneous data. Furthermore, during this phase, blockchain sharding is used to store related data from the semantic mapping matrix in the same shard. The access order between shards stored through the association matrix ensures that during the operation and maintenance phase, when a user triggers access to relevant data in an AI application, the corresponding shard can be quickly located using semantic tags and the association matrix. The main contract then rapidly loads the required data from the corresponding shard, ensuring timely processing of user requests. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the method steps of the present invention;

[0044] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0045] Figure 3 This is a schematic diagram of the response steps based on the main contract;

[0046] Figure 4 This is a schematic diagram of the response steps based on subcontracts and proxy nodes. Detailed Implementation

[0047] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof.

[0048] Identical parts are indicated by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, while the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific part, respectively.

[0049] The term "and / or" in the following text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0050] like Figure 1 As shown in the embodiments of this application, a method for rapid delivery and full lifecycle management of AI application center applications is provided, including:

[0051] Get user request;

[0052] Obtain text vectors and semantic tags based on user requests;

[0053] The segment corresponding to the user request is determined based on the semantic tags, and the segment is obtained according to the preset semantic mapping matrix;

[0054] The main contract is invoked based on the fragments and a pre-defined association matrix to enable the main contract to fulfill the user's request.

[0055] The AI ​​application center's rapid delivery and full lifecycle management includes stages such as development and building, rapid delivery, and operation and maintenance management. The development and building stage includes model training and data integration, which lay the foundation for subsequent delivery. The rapid delivery stage, also known as the data deployment stage, is used in this embodiment to perform sharding and deploy corresponding sub-contracts and agent nodes based on the shards. The operation and maintenance management stage is used after the AI ​​application is launched to respond to user requests and monitor performance and anomalies during the response process.

[0056] Specifically, the method provided in this embodiment is applied to the operation and maintenance management phase. First, it obtains the text vector from the user request. Then, based on a preset semantic mapping matrix, it obtains the semantic tags corresponding to the text vector. Finally, it determines the corresponding shard based on the semantic tags, retrieves the data required by the user through the main contract, and returns it to the user. A shard is a sub-unit obtained by splitting the data stored in the blockchain. Each shard stores a portion of the data in the blockchain, and the data in all shards together constitute the complete data stored in the blockchain. The main contract is a smart contract code module deployed on the blockchain in bytecode form. The main contract retrieves the data required by the user based on the shard corresponding to the user request and the preset association matrix.

[0057] This embodiment uses a semantic mapping matrix to achieve semantic encoding of heterogeneous data. During the rapid delivery phase, blockchain sharding is used to store related data in the same shard. Then, the access order between shards stored by the association matrix ensures that during the operation and maintenance phase, when a user triggers a call for relevant data in the AI ​​application, the corresponding shard can be quickly located through semantic tags and the association matrix, and the main contract can quickly load the required data from the corresponding shard, ensuring that user requests are processed in a timely manner.

[0058] Furthermore, this embodiment provides a step for obtaining a preset semantic mapping matrix, including:

[0059] Historical data is preprocessed to obtain word units;

[0060] Transform the word into a word vector and add positional encoding to obtain the input information;

[0061] Based on the input information and the preset model, the text vector corresponding to the historical data is obtained;

[0062] Based on the text vectors of historical data, a pre-defined semantic mapping matrix is ​​obtained.

[0063] The step of constructing the preset semantic mapping matrix is ​​completed in the development and construction phase. The final preset semantic mapping matrix is ​​stored in the database. The preset model in this embodiment can be a Transformer encoder. This embodiment does not limit this. The Transformer encoder is located in the BERT model, which also includes an input layer.

[0064] Specifically, the first step is to acquire historical data. For example, historical data can include heterogeneous data such as user data, configuration parameters, and third-party interfaces (structured tables, unstructured logs) from the past year, ensuring that the historical data covers the main business scenarios. However, this embodiment does not limit the specific time frame. Next, the historical data is initially categorized according to business type to form a domain corpus. For example, for e-commerce scenarios, business types can include refund-related, product consultation-related, and logistics-related categories. Initial categorization can be achieved by keyword retrieval for each text. If a text contains more than a threshold number of keywords of a certain type, it is classified into that category. For example, if the refund category corresponds to keywords such as refund, return, refund only, and refund amount, and a text containing "refund" and "refund only" is retrieved, then that text is classified into the refund category. This embodiment does not limit the threshold. Furthermore, the keyword retrieval method provided in this embodiment is merely an example; those skilled in the art can choose other methods for initial categorization, and this embodiment does not impose any restrictions on this. The purpose of this initial categorization to obtain the domain corpus is to remove irrelevant information and ensure work efficiency. If a text cannot be classified into any category, it is considered an irrelevant variable and needs to be removed, not participating in subsequent steps. After obtaining the domain corpus, the text within the corpus is preprocessed. This preprocessing includes converting non-text data from heterogeneous datasets into text strings, removing punctuation, and word segmentation. For example, the text "Goods damaged, apply for refund only" becomes "Goods / damaged / apply / refund only" after preprocessing. The initial classification and preprocessing steps are performed using external scripts or tools and are not part of the BERT model's internal processing.

[0065] The preprocessed text is then input into the BERT model. The text first reaches the input layer, which converts the preprocessed text into lexical units recognizable by the BERT model, such as "[CLS]goods[SEP]damaged[SEP]application[SEP]refundonly[SEP]", where "[CLS]" is a sentence-level feature marker and "[SEP]" is a separator. Each lexical unit is then mapped to a corresponding word vector, and positional encoding is added to each word vector to obtain the input information. In the BERT model, sine and cosine functions or rotational positional embeddings are commonly used for encoding; this embodiment does not impose such restrictions. The purpose of positional encoding is to endow the Transformer encoder, which originally could not distinguish the order of lexical units, with sequential understanding capabilities. The input information is then fed into the preset model, where the Transformer encoder extracts the corresponding semantic feature vector for each text based on an attention mechanism. Each semantic feature vector contains the contextual information of all lexical units corresponding to that text; the semantic feature vector is the text vector. Finally, a semantic mapping matrix is ​​obtained based on each text vector, where each row represents a text vector.

[0066] This embodiment establishes a semantic mapping matrix based on historical data. Compared with the existing technology of manually writing mapping rules and integrating historical data, it improves the efficiency of data processing. Furthermore, the semantic mapping matrix can transform historical data into semantic feature vectors and form a mapping relationship between text vectors and semantic tags by combining associated tags. Thus, when a new request comes in, the corresponding semantic tag can be quickly located, and the fragment corresponding to the user request can be determined based on the semantic tag, so as to quickly respond to the user request.

[0067] Furthermore, this embodiment provides a step for obtaining fragments including:

[0068] The iterative operation is performed based on the preset semantic mapping matrix. The iterative operation includes calculating the similarity between all vectors in the preset semantic mapping matrix and each current center vector, determining the vector corresponding to each current center vector based on the similarity, calculating the mean corresponding to each current center vector, determining whether the mean has reached the preset termination condition, if not, updating the current center vector, until the preset termination condition is reached, and outputting each current center vector and its corresponding vector to obtain each slice.

[0069] Specifically, before the first iteration, multiple center vectors need to be initialized as center vectors in the first iteration. That is, in the first iteration, calculating the similarity with each current center vector specifically involves calculating the similarity with each initialized center vector. For example, multiple text vectors can be randomly selected from a preset semantic mapping matrix as initial center vectors, with the number of selected vectors equal to the preset number of shards. This embodiment does not limit the number of shards. For example, the number of shards can be determined based on the business type, or further, based on the business scenario, and ultimately determined according to the business scenario. For example, refund-related businesses can include multiple business scenarios such as refunds for damaged goods and refunds for incorrect addresses, or it can be determined based on data type, such as order number, product information, and user information. The preset number of shards should be less than the upper limit of the number of shards to ensure that adding shards later does not affect system operation. This upper limit is specifically determined by various factors such as the overall system architecture, network bandwidth, and node performance. If the number of shards exceeds this upper limit, it can easily lead to system congestion and performance degradation.

[0070] The first iteration then proceeds. For each remaining text vector in the pre-defined semantic mapping matrix, its similarity to each current center vector is calculated. These remaining text vectors are those not initially selected as center vectors in the pre-defined semantic mapping matrix. Next, each remaining text vector is grouped with its highest-similar center vector. For example, assuming there are only two text vectors A and B, and two center vectors C and D, with A having 80% similarity to C and 50% to D, and B having 70% similarity to C and 60% to D, A and B are grouped with their highest-similar center vector C. In this case, the vectors corresponding to center vector C are A and B. For center vectors that are grouped and contain only themselves, their similarity to each other center vector is calculated, and they are grouped with the group corresponding to their highest-similar center vector. For example, in the case of D, since only one center vector C remains, D is directly grouped with the group corresponding to C. Therefore, the vectors corresponding to center vector C are D, A, and B.

[0071] Next, the mean value corresponding to each current center vector is calculated. For example, for center vector C, its mean value is a vector obtained by averaging A, B, C, and D element by element. Then, it is determined whether the mean value has reached a preset termination condition. The preset termination condition is that the change between each center vector and its corresponding mean value is less than a preset threshold. This embodiment does not limit the specific calculation method or the preset threshold. For example, the change value can be calculated based on distance. The preset threshold is that the distance is less than 0.5. In this case, the termination condition is that the distance between each center vector and its corresponding mean value is less than 0.5. If the preset termination condition is reached, each current center vector is output. Each center vector corresponds to a slice, and the slice stores the vector corresponding to that center vector.

[0072] If the preset termination condition is not met, the current center vector is updated. The update steps are as follows: First, identify the center vector whose corresponding mean change is greater than or equal to a preset threshold. Then, replace this center vector with its corresponding mean. Next, identify the unreplaced initial center vectors and use them, along with the replaced center vectors, as the updated center vectors. That is, the current center vector in the second iteration is the center vector updated in the first iteration. The remaining vectors at this point are the text vectors in the preset semantic mapping matrix that were not selected as initial center vectors and the replaced initial center vectors. The above steps are repeated until the preset termination condition is met. For example, if for center vector C, its corresponding mean is A, B, C, and D, and the change between E and C does not meet the preset termination condition, then E is used instead of C as the updated center vector.

[0073] For each center vector in the final output, if it belongs to the initial center vector, its corresponding slice stores its corresponding vector and itself; if it is the mean generated during a certain iteration, it does not belong to the preset semantic mapping matrix, and the slice corresponding to this center vector only stores its corresponding vector, not itself. However, based on each center vector and its corresponding vector, an index table is generated. The index table records the slice corresponding to each center vector, the specific value of each center vector, and the position of its corresponding vector in the preset semantic mapping matrix. Finally, this index table is stored in the database for subsequent retrieval.

[0074] Furthermore, the preset semantic mapping matrix in this embodiment is not fixed. During the operation and maintenance phase, new data will be continuously collected. The newly collected data will be temporarily stored in an independent cache area in the system. Every preset time interval, the corresponding text vector needs to be generated based on the data in the cache area. The steps for generating text vectors are the same as those for generating text vectors corresponding to historical data, and will not be elaborated here. This embodiment does not limit the length of the preset time interval. Finally, the text vectors are added to the preset semantic mapping matrix to update the semantic mapping matrix. Also, the text vectors generated from the newly collected data need to be added to the shards. Specifically, for each text vector generated from the newly collected data, the index table needs to be called to calculate its similarity to each center vector in the index table. If the highest similarity is greater than or equal to the preset similarity, the text vector is added to the shard corresponding to the center vector with the highest similarity. After adding to the shard, the mean of all vectors in that shard needs to be recalculated. Then, this mean is used as the updated center vector of that shard, and the index table is updated based on the updated center vector and the vectors included in the shard. This embodiment does not limit the value of the preset similarity; if the highest similarity is greater than or equal to the preset similarity, the text vector is added to the shard corresponding to the center vector with the highest similarity. If the similarity is less than the preset similarity, a new shard needs to be added to store the text vector. Since there is only one text vector in the shard at this time, the text vector is used as the center vector corresponding to the shard, and the index table is updated. Since each shard in this embodiment corresponds to a different business type or business scenario, the vectors stored between each shard are significantly different. Therefore, this embodiment does not consider the case where the highest similarity corresponds to multiple center vectors. If such a case occurs, it can be considered whether there is an error in determining the business type or business scenario when determining the number of shards, and the shard selection and update process can be backtracked and analyzed to correct the error. This embodiment does not impose any restrictions on this.

[0075] This embodiment is based on a blockchain sharding design. By iteratively updating the central vector and verifying the convergence of the mean, it can maximize the semantic similarity of vectors in the same shard. This allows different types of data to be stored in different shards, ensuring that user requests can be directly routed to the corresponding shard after entering the system. In other words, when the main contract accesses a shard, it does not need complex judgments but directly matches the corresponding shard based on semantic tags, avoiding the waste of resources when traversing shards and ensuring efficient processing of user needs.

[0076] Furthermore, this embodiment provides a method for obtaining semantic tags based on user requests, including:

[0077] Based on the user request and the text convolutional neural network model, a global feature vector is obtained;

[0078] The main label is obtained based on the global feature vector and the preset main label library;

[0079] Based on the user request and the named entity recognition model, sub-labels are obtained;

[0080] Based on the main tag and child tags, semantic tags are obtained.

[0081] Specifically, the first step is to obtain the corresponding text vector based on the user request. The specific steps are the same as those for obtaining the text vector corresponding to historical data, and will not be repeated here. After obtaining the text vector corresponding to the user request, it is input into a text convolutional neural network model. A global feature vector is obtained through convolution and pooling operations. The global feature vector is then mapped to a vector of a preset dimension using a fully connected layer. The probability value corresponding to each dimension is obtained through an activation function. The value of the preset dimension is determined based on the category label. For example, a preset main label library stores multiple category labels. Each category label corresponds to multiple preset main labels and their corresponding vectors. This vector should have the same dimension as the global feature vector. The category label is obtained by clustering analysis of multiple preset main labels. For example, if the preset main labels "Returned Goods" and "Unsubscribed Members" are clustered into one category, then this category label can be recorded as "Returned Goods". Therefore, the category label corresponding to the text vector of the user request can be determined based on the probability value. The process of classifying multiple preset main labels based on clustering analysis and determining the category label based on the preset main labels included in each category is a technical means that can be implemented by those skilled in the art, and will not be repeated here. Then, based on the category label corresponding to the text vector, the vector corresponding to each preset main label in the category label is obtained, and the similarity between each vector and the global feature vector is calculated. The preset main label corresponding to the highest similarity is taken as the main label corresponding to the text vector. This embodiment does not restrict the calculation method of similarity. For example, cosine similarity can be used.

[0082] Next, based on the text corresponding to the user request, proper nouns in the text, such as the shopping platform name, business terms, and numerical types, are tagged using a Named Entity Recognition (NER) model. The identified proper nouns are then arranged to obtain the sub-labels corresponding to the vector. The NER model is an existing model, and this embodiment will not elaborate on it. Finally, the main label and sub-labels are combined to obtain the semantic label corresponding to the user request.

[0083] This embodiment determines the main tag through a text convolutional neural network model and a preset main tag library to ensure that the main tag meets the user's core needs. Then, the NER model accurately identifies proper nouns (such as platform, product, address) in the text, so that the sub-tags reflect the details of the user's needs. Finally, the main tag and sub-tags form semantic tags, enabling the system to accurately identify the corresponding business scenario based on the semantic tags, and achieve efficient response and accurate execution of user requests.

[0084] Furthermore, this embodiment provides a method for determining the fragment corresponding to a user request based on semantic tags, including:

[0085] Determine the corresponding decision tree model based on the main label;

[0086] Determine the corresponding leaf node based on the sub-label and the intermediate node in the decision tree model;

[0087] The pre-defined association matrix is ​​determined based on the leaf nodes;

[0088] The shard corresponding to a user request is determined based on a pre-defined association matrix.

[0089] Specifically, in this embodiment, each preset main label corresponds to a decision tree model, and the preset main label is the root node in the decision tree model. Therefore, after obtaining the main label corresponding to the user request, the corresponding decision tree model can be accurately called according to the correspondence between the main label and the root node. Then, according to the information corresponding to the sub-labels, the intermediate nodes are passed sequentially, with each intermediate node corresponding to one piece of information in the sub-label, until the leaf node is reached. The leaf node is the corresponding business scenario. Finally, the preset association matrix corresponding to the leaf node is called, which is the preset association matrix corresponding to the user request. The association matrix records the shards corresponding to the request and the access order of the shards. Among them, the models used in this embodiment, such as the text convolutional neural network model, NER model, decision tree model, and BERT model, are all trained models. Training the model is a conventional technique for those skilled in the art, and determining the information and position corresponding to each intermediate node in the decision tree model based on the training is also a conventional technique for those skilled in the art. This embodiment will not elaborate on this.

[0090] This embodiment uses the main label as the root node, and then sequentially passes through intermediate nodes through the information corresponding to the sub-labels, finally locating the scene corresponding to the unique leaf node. This ensures the clarity of the recognition logic, and the paths for business scenarios determined based on the decision tree model are all traceable. When the recognition result is abnormal, such as when the leaf node cannot be located, the abnormal intermediate node can be directly located and an error can be reported, so as to correct the model in a timely manner. Compared with black box models such as neural networks, the transparency of the decision tree model can reduce debugging costs. When a new business scenario needs to be added, only intermediate nodes and leaf nodes need to be added to the decision tree, without reconstructing the entire model. In contrast, models such as neural networks require retraining a large number of samples and cannot quickly respond to adjustments in business rules.

[0091] Furthermore, this embodiment provides a step for obtaining a preset correlation matrix, including:

[0092] Determine the corresponding association rules based on the preset business scenarios;

[0093] Based on the corresponding fragments in the association rules, a preset association matrix is ​​obtained for each scene.

[0094] Specifically, during the development and construction phase, association rules are determined based on the specific pre-set business scenarios. For example, in the business scenario of a user submitting a refund request, the steps to be performed are: verifying the order validity, verifying the product status, confirming the refund address, and executing the refund operation. Verifying the order validity means verifying whether the order exists, i.e., determining whether it is a forged order number. Then, verifying the product status, such as determining whether the product belongs to the category of products not eligible for refund. Finally, confirming the refund address based on the user information and executing the refund operation, the association rule is: verify order validity → verify product status → confirm refund address.

[0095] In the rapid delivery phase, assuming that fragmentation is based on data type, order number, product information, and user information are located in separate fragments. , and In this case, the association rule is obtained as follows: → → Based on the sharding involved in this rule, the correlation matrix corresponding to this request scenario can be obtained as follows: .

[0096] Each row and each column represents a partition, because and There is a connection between them, and the arrow direction is... → Therefore, the element in the first row and second column of the association matrix is ​​1, and similarly, the element in the second row and third column is 1. Thus, for each business scenario, its corresponding association matrix can be obtained, which illustrates the shards that need to be accessed when responding to the request corresponding to that scenario.

[0097] In this embodiment, an association matrix is ​​set up. After obtaining a user request, the corresponding business scenario can be determined based on the user request. Then, the corresponding association rules are called according to the business scenario. Finally, the main contract is called according to the corresponding shards and order in the association matrix so that the main contract can complete the user request.

[0098] like Figure 2 As shown in the figure, this application embodiment provides an AI application center application rapid delivery and full lifecycle management system, including:

[0099] The acquisition module is used to acquire user requests;

[0100] The identification module obtains text vectors and semantic tags based on user requests;

[0101] The positioning module is used to determine the segment corresponding to the user request based on the semantic tags. The segment is obtained according to the preset semantic mapping matrix.

[0102] The scheduling module calls the main contract based on the sharding and the preset association matrix so that the main contract can complete the user request.

[0103] Specifically, the system provided in this embodiment includes a request access layer, a main contract, sharding, subcontracts, and a database, such as... Figure 2 As shown, the request access layer includes the aforementioned acquisition module, identification module, location module, and scheduling module. When a user request arrives at the request access layer, the acquisition module first receives the user request and sends it to the identification module. The identification module first obtains the text vector based on the BERT model and then obtains the semantic label based on the text convolutional neural network model and the NER model. The semantic label is then sent to the location module. The location module determines the specific business scenario using the semantic label and the decision tree model, and then determines the corresponding association matrix based on the business scenario. The location module then sends an instruction to the scheduling module, causing the scheduling module to send the corresponding association matrix to the main contract. This allows the main contract to complete the user request according to the shards and access order recorded in the association matrix. Specifically, the main contract can send instructions to subcontract 1, subcontract 2, ..., subcontract N simultaneously through the proxy node, causing the subcontracts to call the corresponding shards. However, this embodiment does not limit this. In practical applications, the main contract only calls some subcontracts through the proxy node. Since each association rule corresponds to a different subcontract, it is possible that subcontract 1 in one association rule corresponds to two shards (shards...). and In another association rule, subcontract 2 corresponds to only one shard. , Figure 2 Subcontract 1, Subcontract 2, ... Subcontract N in the shard are sets of subcontracts included in different association rules. Each shard corresponds to a proxy node, such as shards. The corresponding proxy node M. In this embodiment, the preset semantic mapping matrix and association matrix are stored in the database.

[0104] Furthermore, this embodiment provides a step of invoking the main contract based on sharding and a preset association matrix to enable the main contract to complete the user request, including:

[0105] The access order of each shard is determined based on the sharding and the preset association matrix;

[0106] The main contract is invoked so that it iterates through each shard in turn to fulfill the user request.

[0107] For example, such as Figure 3 As shown, using the aforementioned business scenario of a user submitting a refund request, after the main contract receives the association matrix and determines the sharding and access order, it first needs to access the shard. Check if the user's order for which a refund is requested is valid. If so, access the segment. Determine whether the product is not eligible for a refund. If not, confirm the refund address based on the user information and send an instruction to the refund portal to complete the refund process.

[0108] This embodiment uses the main contract to sequentially traverse each shard to complete the user request. Compared to existing technologies that require traversing a large amount of data for each access without sharding, this embodiment quickly accesses data and completes the user request based on the determined shards and access order, improving access efficiency. However, this embodiment further considers that since the execution of blockchain smart contracts is serial, meaning that a contract can only process one instruction stream at a time, in this embodiment, when all shard access depends on the main contract, each access step is executed in series. In this case, if one access step fails, it needs to start over from the first step. For example, if the access... Network lag can cause excessively long access times, while accessing... During the process, The refund details may change due to adjustments in the refund policy. To ensure the accuracy of the entire refund process, it is necessary to start over from the first step.

[0109] Therefore, further in this embodiment, sub-contracts are set up in the above system, each sub-contract corresponding to at least one shard. The sub-contracts are used to split a complete association rule into multiple sub-rules. For example, the above... → → It can be broken down into → and , → and All are subcontracts, and a proxy node is set up for each shard. The proxy node is the entry point for the shard and stores frequently accessed data in the shard. A corresponding proxy node is set up for the main contract to execute the instructions sent by the main contract. Both the main contract and the subcontracts are smart contract code modules, deployed on the blockchain in bytecode form. Therefore, communication between the proxy node, the main contract, and the subcontracts is all at the code level, consisting of function calls.

[0110] Furthermore, this embodiment provides a step of invoking the main contract based on sharding and a preset association matrix to enable the main contract to complete the user request, including:

[0111] The subcontract to be invoked is determined based on the fragmentation and the preset association matrix;

[0112] The main contract is invoked so that it sends instructions to each subcontract to be invoked, and each subcontract to be invoked obtains the verification result according to the corresponding proxy node and sends it to the main contract so that the main contract can complete the user request.

[0113] Since each preset association matrix corresponds to an association rule, and each association rule can be divided into multiple sub-contracts, this embodiment refers to the sub-contract in the association rule corresponding to the user request as the sub-contract to be called.

[0114] For example, such as Figure 4 As shown, Figure 4 The numbers ①-⑧ in the table represent the sequence of steps. Continuing with the business scenario of a user submitting a refund request, after the main contract receives the association matrix, it first sends an instruction to its corresponding proxy node 1. Then, proxy node 1 simultaneously sends instructions to both sub-contracts. → Upon receiving the instruction, simultaneously send and The corresponding proxy nodes 2 and 3 send instructions. Proxy nodes 2 and 3 first check their own stored data. If they cannot complete the verification of order validity or product status based on their own stored data, then the proxy nodes further access their corresponding shards. (In the subcontract) → Upon receiving the instruction, The same instruction was also received. A command is sent to proxy node 4 to check if the data it stores can identify the refund address. If not, the corresponding shard is accessed. The proxy node then returns the results of these accesses sequentially to the subcontract, proxy node 1, and the main contract. The main contract then integrates these results and determines whether to execute the refund operation based on the final outcome.

[0115] This embodiment achieves task division and decoupling through subcontracts, and multiple steps are processed in parallel by different subcontracts. This avoids the problem of a single point of failure causing the entire process to be paralyzed when accessing the main contract alone. Furthermore, by setting up proxy nodes to cache high-frequency data, the pressure of directly accessing shards is reduced, and query response is accelerated. Through the cooperation of subcontracts and proxy nodes, access efficiency in business scenarios can be improved.

[0116] The AI ​​application center application rapid delivery and full lifecycle management method and system provided in this application embodiment realizes semantic encoding of heterogeneous data through a semantic mapping matrix. In the rapid delivery stage, blockchain sharding is used to store related data in the same shard. Then, the access order between shards stored by the association matrix is ​​used to enable the corresponding shard to be quickly located through semantic tags and the association matrix when a user triggers the call for relevant data in the AI ​​application during the operation and maintenance management stage. The required data can then be quickly loaded from the corresponding shard through the main contract, ensuring that user requests are processed in a timely manner.

[0117] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An AI application center application rapid delivery and full life cycle management method, characterized in that, The method comprises the following steps: obtaining a user request; obtaining a text vector and a semantic label according to the user request; determining a corresponding shard of the user request according to the semantic label, wherein the shard is obtained according to a preset semantic mapping matrix; calling a main contract according to the shard and the preset association matrix, so that the main contract completes the user request; wherein the step of obtaining the preset association matrix comprises: determining a corresponding association rule according to a preset business scenario; obtaining the preset association matrix corresponding to each scenario according to the corresponding shard in the association rule; the association matrix comprises a plurality of shards corresponding to the scenario and a connection relationship of each shard; wherein calling the main contract according to the shard and the preset association matrix, so that the main contract completes the user request, comprises: determining an access order of each shard according to the shard and the preset association matrix; calling the main contract, so that the main contract sequentially traverses each shard to complete the user request. 2.The AI application center application rapid delivery and full life cycle management method of claim 1, wherein, The step of obtaining the preset semantic mapping matrix comprises: preprocessing historical data to obtain a word element; converting the word element into a word vector and adding position encoding to obtain input information; obtaining a text vector corresponding to the historical data according to the input information and a preset model; obtaining the preset semantic mapping matrix according to the text vector of the historical data. 3.The AI application center application fast delivery and full life cycle management method of claim 1, wherein, The shard is obtained according to the preset semantic mapping matrix, comprising: performing an iteration operation according to the preset semantic mapping matrix, wherein the iteration operation comprises: calculating the similarity of all vectors in the preset semantic mapping matrix and each current center vector, determining the vector corresponding to each current center vector according to the similarity, calculating the mean value corresponding to each current center vector, judging whether the mean value reaches a preset termination condition, updating the current center vector if the preset termination condition is not reached, and outputting each current center vector and its corresponding vector to obtain each shard. 4.The AI application center application fast delivery and full life cycle management method of claim 1, wherein, Obtaining a semantic label according to the user request comprises: obtaining a global feature vector according to the user request and a text convolutional neural network model; obtaining a main label according to the global feature vector and a preset main label library; obtaining a sub-label according to the user request and a named entity recognition model; obtaining the semantic label according to the main label and the sub-label. 5.The AI application center application fast delivery and full life cycle management method of claim 4, wherein, Determining the shard corresponding to the user request according to the semantic label comprises: determining a corresponding decision tree model according to the main label; determining a corresponding leaf node according to the sub-label and an intermediate node in the decision tree model; determining a preset association matrix according to the leaf node; determining the shard corresponding to the user request according to the preset association matrix. 6.The AI application center application fast delivery and full life cycle management method of claim 1, wherein, Calling the main contract according to the shard and the preset association matrix, so that the main contract completes the user request, comprises: determining a to-be-called sub-contract according to the shard and the preset association matrix; calling the main contract, so that the main contract sends an instruction to each to-be-called sub-contract, so that each to-be-called sub-contract obtains a verification result according to a corresponding proxy node and sends it to the main contract, so that the main contract completes the user request. 7.The AI application center application rapid delivery and full life cycle management method of claim 6, wherein, Each of the to-be-called sub-contracts corresponds to at least one shard, each shard corresponds to one proxy node, and the main contract corresponds to one proxy node.

8. An AI application center application rapid delivery and full life cycle management system, characterized in that, Comprise: An acquisition module is configured to acquire a user request; An identification module is configured to obtain a text vector and a semantic label according to the user request; A positioning module is configured to determine a shard corresponding to the user request according to the semantic label, wherein the shard is obtained according to a preset semantic mapping matrix; A scheduling module is configured to call a main contract according to the shard and a preset association matrix, so that the main contract completes the user request; Wherein, the step of obtaining the preset association matrix comprises: According to the corresponding association rule of the preset business scenario, the corresponding association rule of the preset business scenario is determined; According to the corresponding shard in the association rule, the preset association matrix corresponding to each scene is obtained; the association matrix comprises a plurality of shards corresponding to the scene and the connection relationship of each shard; Wherein, according to the shard and the preset association matrix, the main contract is called to make the main contract complete the user request, comprising: According to the shard and the preset association matrix, the access order of each shard is determined; The main contract is called to make the main contract traverse each shard in turn to complete the user request.

Images