Business inference method and related apparatus

Abstract

A business inference method, applied to the technical field of artificial intelligence (AI). During business inference, business inference is performed in a logic graph on the basis of business data, wherein the logic graph comprises judgment nodes indicating conditions that business objects need to satisfy and conclusion nodes indicating business inference results. Thus, a related conclusion node can be determined by determining a judgment node satisfied by the business data, so as to obtain a business inference result. In addition, the value of a business object in the judgment node is obtained from a target field corresponding to a database, wherein the target field is determined by means of an AI model on the basis of the business object and reference knowledge thereof, thereby ensuring that the value of the business object in the logic graph can be accurately obtained when enterprise data is independently stored, thus ensuring that business inference can be automatically and reliably executed, and effectively improving the efficiency of business inference.

Description

A business reasoning method and related apparatus

[0001] This application claims priority to two Chinese patent applications filed with the State Intellectual Property Office on December 27, 2024, application number 202411977909.2, entitled "An Analysis and Decision-Making Method and Related Equipment for Enterprise Logic Graphs" and on March 28, 2025, application number 202510390246.2, entitled "A Business Reasoning Method and Related Device", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of artificial intelligence (AI) technology, and in particular to a business reasoning method and related apparatus. Background Technology

[0003] With the development of information technology, enterprises, schools, and other institutions often need to analyze and reason about large amounts of business data in their daily operations to obtain the results. For example, in a company's tax rate calculation scenario, the company may need to calculate the tax rate corresponding to each income from different sources. Another example is in a company or school's code review scenario, where the company or school needs to review whether the generated code is up to standard.

[0004] Taking enterprises as an example, in business reasoning scenarios, the internal and external business rules and logic of an enterprise often have clear and rigorous expressions, with explicit analytical steps and dependencies. The reasoning chain is long and complex, sometimes even involving complex analytical logic such as logical judgments and calculations. Therefore, current business reasoning scenarios usually rely on experts to complete them. For example, in the scenario of calculating enterprise tax rates, the tax rate corresponding to each revenue in the enterprise is analyzed and reasoned by specialized financial personnel within the enterprise.

[0005] Currently, complex business reasoning scenarios in enterprises heavily rely on expert experience, requiring experts to manually perform business reasoning, resulting in low efficiency and difficulty in meeting the business development needs of enterprises. Summary of the Invention

[0006] This application provides a business reasoning method that can effectively improve the efficiency of business reasoning in enterprises.

[0007] Firstly, a business reasoning method is provided for implementing business reasoning in enterprise scenarios. This method specifically includes: executing a device to obtain a business reasoning request, whereby the business reasoning request instructs the execution of business reasoning on business data. The business reasoning request may contain business data indicating the need for business reasoning, such as carrying business data or indicating the method for obtaining the business data.

[0008] Then, based on the business reasoning request, the execution device determines the first conclusion node corresponding to the business data in the logical graph. The logical graph includes multiple judgment nodes and multiple conclusion nodes. The judgment nodes indicate the conditions that the business object must satisfy, and the conclusion nodes indicate the business reasoning result, whereby the business data satisfies the conditions indicated by the judgment nodes related to the first conclusion node. That is, the execution device determines which conditions indicated by the judgment nodes in the logical graph the business data satisfies to determine the judgment nodes corresponding to the business data; thus, based on the judgment nodes corresponding to the business data, the execution device continues to determine the conclusion nodes related to these judgment nodes, thereby obtaining the first conclusion node corresponding to the business data in the logical graph.

[0009] The business data includes the value of the first business object, the value of the first business object satisfies the conditions indicated by the first judgment node related to the first conclusion node, the value of the first business object is obtained based on the data recorded under the first target field, and the first target field is determined in the database by an artificial intelligence AI model based on the first business object and the reference knowledge related to the first business object. The database is used to store the data in the business data.

[0010] Finally, the execution device outputs the business reasoning result indicated by the first conclusion node. That is, after determining the first conclusion node corresponding to the business data, the execution device outputs the business reasoning result indicated by the first conclusion node as the result of this business reasoning, thereby completing the business reasoning process.

[0011] In this solution, during business reasoning, reasoning is performed within a logical graph based on business data. This logical graph includes decision nodes indicating the conditions that business objects must meet and conclusion nodes indicating the results of the reasoning. Therefore, by determining the decision nodes that the business data satisfies, the relevant conclusion nodes can be determined, thus yielding the business reasoning results. Furthermore, the values ​​of business objects in the decision nodes are obtained from the corresponding target fields in the database. These target fields are determined using an AI model based on the business objects and their reference knowledge. This ensures that, even with independent storage of enterprise data, the values ​​of business objects in the logical graph can be accurately retrieved, thereby ensuring that business reasoning can be executed automatically and reliably, effectively improving the efficiency of business reasoning.

[0012] In one possible implementation, the first business object is determined from the set of business objects based on the conditions indicated by the first judgment node. The business objects in the set of business objects are extracted from the set of rules, which are used to guide the construction of the logic graph.

[0013] That is, the execution device can find the first business object in the condition indicated by the first judgment node in the business object set, and the business object set is pre-extracted based on the rule set to improve the accuracy of the execution device in determining the business object.

[0014] In one possible implementation, the reference knowledge related to the first business object includes at least one of the following: rules, business cases, expert knowledge, and terminology explanations used in the rule set to extract the first business object.

[0015] In one possible implementation, the conditions indicated by the second judgment node related to the first conclusion node include a second business object, and the value of the second business object is obtained by the AI ​​model processing the data recorded under the second target field in the database.

[0016] That is, the value of the second business object cannot be obtained directly by extracting the data recorded under the second target field. Instead, the AI ​​model needs to further process the data recorded under the second target field to obtain the value.

[0017] In this solution, for some decision nodes, the AI ​​model is used to process the data in the database to obtain the value of the business object. This can effectively obtain the data of the business object even when the database itself does not directly store the data of the business object, thus improving the feasibility of the solution.

[0018] In one possible implementation, the business data also includes business description text not stored in the database, and the conditions indicated by the third judgment node related to the first conclusion node include a third business object, the value of which is extracted from the business description text by an AI model.

[0019] That is, the execution device cannot retrieve the fields related to the third business object from the database, and therefore cannot obtain the value of the third business object from the database. Instead, it needs to extract the value of the third business object based on the business description text outside the database.

[0020] In one possible implementation, the logic graph records the value range of the third business object, and the value range is used to provide constraints for the AI ​​model when extracting the value of the third business object.

[0021] In other words, in order to enable the AI ​​model to extract accurate values ​​for the third business object, the device responsible for constructing the logic graph can obtain the value range of the third business object in advance and record the value range of the third business object in the logic graph, so as to provide constraints for the AI ​​model when extracting the value of the third business object.

[0022] In this solution, the execution device uses an AI model to extract relevant data from the descriptive text given in the business reasoning request, which ensures the smooth execution of the business reasoning process and improves the applicability of the solution.

[0023] In one possible implementation, the logic graph records the target tools to be invoked when processing the first decision node, and the target tools are determined according to the conditions indicated in the first decision node; wherein, the target tools include one or more of AI models, rule engines, and analysis operators.

[0024] In this solution, for different judgment nodes in the logic graph, by recording the tools corresponding to the judgment nodes, the execution device can quickly process the judgment nodes based on the corresponding tools, thereby improving the efficiency and accuracy of the execution device in processing judgment nodes.

[0025] In one possible implementation, when the target tool includes an AI model, a correspondence between the first business object and reference knowledge is also established in the logic graph. This correspondence is used to instruct the AI ​​model to process the first judgment node by referring to the reference knowledge during the business reasoning process.

[0026] In this solution, by retrieving relevant reference knowledge for business objects from the knowledge base, the AI ​​model can be provided with relevant reference knowledge when it is called to process judgment nodes during the business reasoning process. This makes it easier for the AI ​​model to understand and analyze the conditions in the judgment nodes, and helps to improve the accuracy of the AI ​​model's reasoning.

[0027] In one possible implementation, the reference knowledge includes multiple types of knowledge, and the correspondence is also used to indicate the reference priority of each type of knowledge among the multiple types of knowledge.

[0028] In one possible implementation, the logic graph is a tree-connected graph, and the logic graph includes multi-level decision nodes in a tree connection. Each decision node in the last level of the multi-level decision nodes is connected to a conclusion node, and the decision nodes associated with the conclusion node include all decision nodes on the path from the root node of the logic graph to the conclusion node.

[0029] In one possible implementation, the business data includes scenario description data, and the logical graph is determined from multiple constructed logical graphs based on the scenario description data. Each logical graph has a corresponding scenario label, which is used to indicate the business scenario to which the logical graph is applicable.

[0030] In a second aspect, a business reasoning apparatus is provided, comprising modules for executing the business reasoning method in the first aspect or any possible implementation thereof.

[0031] Thirdly, a business inference apparatus is provided, comprising: a processor and a memory; the memory is used to store computer instructions, which, when executed by the processor, cause the business inference apparatus to perform the methods of any of the above aspects.

[0032] Fourthly, a computer-readable storage medium is provided that stores instructions which, when executed on a computer, cause the computer to perform the methods of any of the above aspects.

[0033] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, enable the computer to perform the methods described above.

[0034] In a sixth aspect, a chip is provided, the chip including a processor and a communication interface for communicating with a module other than the chip, the processor for running computer programs or instructions such that a device on which the chip is mounted can perform the methods of any of the above aspects.

[0035] In a seventh aspect, a computing device is provided, the computing device including a business inference device of the third aspect or a chip of the sixth aspect, wherein the business inference device or the chip in the computing device is used to implement the operational steps of the method of any of the above aspects.

[0036] Eighthly, a computing device cluster is provided, comprising at least one computing device, wherein any one computing device is used to run a computer program or instructions, such that the computing device cluster can perform the methods of any of the above aspects. Alternatively, some or all of the computing devices are used together to run a computer program or instructions, such that the computing device cluster can perform the methods of any of the above aspects.

[0037] Based on the implementation methods provided in the above aspects, this application can be further combined to provide more implementation methods. The beneficial effects of the second to eighth aspects mentioned above can be referred to the introduction of the first aspect above, and will not be repeated here. Attached Figure Description

[0038] Figure 1 is a schematic diagram of a system architecture 100 provided in this application;

[0039] Figure 2 is a flowchart illustrating a business reasoning method provided in this application;

[0040] Figure 3A is a schematic diagram of a process for constructing a logic graph provided in this application;

[0041] Figure 3B is a schematic diagram of the logic graph in a tax rate reasoning scenario provided in this application;

[0042] Figure 4 is a schematic diagram of a system architecture for implementing logical graph construction provided in this application;

[0043] Figure 5 is a schematic diagram of establishing correspondences in a logic graph according to this application;

[0044] Figure 6 is a schematic diagram of a process for constructing a logic graph provided in this application;

[0045] Figure 7 is a schematic diagram of constructing a judgment node based on a business object according to this application;

[0046] Figure 8 is a schematic diagram of establishing a correspondence between business objects to obtain the actual values ​​of the business objects provided in this application;

[0047] Figure 9 is a schematic diagram of a tool for configuring and determining which node needs to be called, provided in this application;

[0048] Figure 10 is a schematic diagram of a method for processing different types of decision nodes provided in this application;

[0049] Figure 11 is a schematic diagram of the structure of a business reasoning device provided in this application;

[0050] Figure 12 is a schematic diagram of the structure of a computing device provided in this application;

[0051] Figure 13 is a schematic diagram of the structure of a computing device cluster provided in this application;

[0052] Figure 14 is a schematic diagram of another computing device cluster provided in this application;

[0053] Figure 15 is a schematic diagram of the structure of a chip provided in this application;

[0054] Figure 16 is a schematic diagram of the structure of a computer-readable storage medium provided in this application. Detailed Implementation

[0055] To facilitate understanding, some technical terms used in this application will be introduced below.

[0056] (1) Large Language Model

[0057] Large language models are deep learning models trained on massive amounts of text data that can generate natural language text or understand the meaning of language text. Large language models can handle various natural language tasks, such as text classification, question answering, and dialogue, and are an important pathway to artificial intelligence.

[0058] Specifically, large language models are a technology that has emerged in recent years. Because large language models undergo meticulous data engineering and training processes, their parameters have learned a wealth of existing natural language processing knowledge. This knowledge can now replace humans in many language-related tasks, such as having large language models write code or perform text summarization.

[0059] Currently, large language models are mainly composed of Transformer networks.

[0060] (2) Transformer network

[0061] Transformer networks are powerful sequence models, but the computation time and memory required increase quadratically with sequence length, significantly increasing the hardware's storage and computing power demands. Essentially, Transformer networks employ a self-attention mechanism. Self-attention is a mechanism that associates different positions within a single sequence to compute a representation of the same sequence, playing a crucial role in machine reading, abstract summarization, and image description generation.

[0062] Taking the Transformer network applied to natural language processing as an example, the Transformer network processes input data of arbitrary length and generates new feature representations of the input data, which are then converted into target words. The self-attention network layer in the Transformer network uses an attention mechanism to capture the relationships between all other words, thereby generating new feature representations for each word. The advantage of the Transformer network's self-attention network is that the attention mechanism can directly capture the relationships between all words in a sentence without considering word positions.

[0063] (3) Prompt

[0064] Prompts originated as an input format designed by researchers for downstream tasks. Their purpose is to help pre-trained models "recall" what they "learned" during pre-training, hence the name "cue word." For large language models, a prompt is the user's input, instructing the model on the task to be performed. A prompt can be a simple question, a longer text, or a set of instructions, depending on the user's specific needs. Generally, a prompt is a short text string that provides context and task-related information to help the model better understand the requirements and generate the correct output. For example, in question-answering tasks, a prompt might contain a description of the question or topic to help the large language model generate the correct answer. Furthermore, prompts are often designed by humans to help large language models better understand specific tasks or domains.

[0065] In this way, when a large language model generates content, it first processes the prompt and then outputs content based on its understanding of the prompt. The working principle of the large language model is to predict the probability of the next word appearing based on the preceding context of the user input, thereby generating the following text word by word. Therefore, differences in the user's input prompt directly affect the quality of the large language model's output. In some cases, even a difference of just a few words in the user's input prompt can result in significantly different content generated by the large language model.

[0066] (4) Rule Engine

[0067] A rules engine separates rules from hard code and transforms them into configuration files or configuration items. This allows business experts and other rule configuration personnel to write business decisions and form business rules without writing code, using predefined semantic modules.

[0068] A rule engine typically consists of a rule base, an inference engine, and an executor. The rule base stores business rules, the inference engine parses and executes the rules, and the executor performs corresponding operations based on the inference engine's results.

[0069] Please refer to Figure 1, which is a schematic diagram of a system architecture 100 provided in this application. As shown in Figure 1, in the system architecture 100, the execution device 110 can be implemented by at least one computing instance among physical hosts (computing devices), virtual machines, and containers. When the execution device 110 is implemented by a virtual machine or a container, the execution device 110 actually exists in the form of a cloud computing product and can provide cloud services.

[0070] Optionally, the execution device 110 can be used in conjunction with other computing devices, such as data storage devices, load balancers, etc.; the execution device 110 can be deployed on a single physical site or distributed across multiple physical sites.

[0071] In addition, the system architecture 100 also includes a data storage system 120, which is used to store business data.

[0072] Optionally, for persistent data storage, the data storage system 120 may be located external to the execution device 110, exchanging data with the execution device 110 via a network. Alternatively, if the execution device 110 is a physical host, the data storage system 120 may also be located internally to the execution device 110, such as implementing the data storage system 120 as a database system that exchanges data with the processor via a bus. With the data storage system 120 present, the execution device 110 can use the data in the data storage system 120 to perform business reasoning.

[0073] Optionally, the user can interact with the execution device 110 by operating the local device 101. The local device 101 can represent any computing device, such as a personal computer, computer workstation, smartphone, tablet, and laptop.

[0074] Local device 101 can interact with execution device 110 through a communication network of any communication mechanism / standard. The communication network can be a wide area network, a local area network, a point-to-point connection, or any combination thereof.

[0075] In one implementation, the execution device 110 is used to implement the business reasoning method provided in this application, thereby obtaining the business reasoning result.

[0076] Optionally, during the process of the execution device 110 implementing the business inference method, the local device 101 can provide the execution device 110 with a business inference request, so that the execution device 110 can perform business inference based on the business inference request. Furthermore, after the execution device 110 executes the business inference method and obtains the business inference result, it can feed the business inference result back to the local device 101.

[0077] In another implementation, the execution device 110 and the local device 101 can also be the same device, that is, the execution device 110 can directly obtain the business inference request provided by the user.

[0078] It should be noted that the system architecture 100 provided in this application can be applied to any business reasoning scenario, such as the business reasoning scenarios of enterprises, schools, hospitals or research institutes, and this application does not limit the specific business reasoning scenario.

[0079] Please refer to Figure 2, which is a flowchart illustrating a business reasoning method provided in this application. As shown in Figure 2, the execution flow of the business reasoning method includes the following steps 201-203.

[0080] Step 201: Execute the device to obtain a business inference request. The business inference request is used to instruct the execution of business inference on the business data.

[0081] In this application, the execution device triggers the execution of the business inference process by obtaining a business inference request provided by the user. Specifically, the execution device may obtain the business inference request by receiving instructions issued by the user. For example, if a business analysis system runs on the execution device, the user can issue a business inference request to the execution device by clicking the "Business Inference" button on the system's interface. Alternatively, the execution device may also obtain the business inference request by receiving messages from other devices (such as the user's local device).

[0082] The business reasoning request may contain business data indicating the need for business reasoning, such as business data itself or instructions on how to obtain it. Furthermore, the requested business reasoning content is often closely related to the business data. Specifically, business data can include at least one of multimedia data such as text, images, audio, or video. The content of the business data is also related to the business reasoning scenario. For example, in a tax-related scenario such as corporate tax rate reasoning, the business data might include descriptive text about a specific revenue stream within the company, and the business reasoning request would request the reasoning of the tax rate for that revenue stream. Similarly, in a medical scenario such as disease reasoning, the business data might include descriptive text or images of symptoms, and the business reasoning request would request the reasoning of a disease based on those symptoms. And in a financial scenario such as corporate profit reasoning, the business data might include tables of various financial data within the company, and the business reasoning request would request the reasoning of the company's profit over a specific period.

[0083] Step 202: The execution device determines the first conclusion node corresponding to the business data in the logic graph based on the business reasoning request. The logic graph includes multiple judgment nodes and multiple conclusion nodes. The multiple judgment nodes are used to indicate the conditions that the business object needs to meet, and the multiple conclusion nodes are used to indicate the business reasoning result. The business data meets the conditions indicated by the judgment nodes related to the first conclusion node.

[0084] In this step, based on the business reasoning request, the execution device can perform business reasoning based on business data within the logic graph to determine the first conclusion node corresponding to the business data. Specifically, each conclusion node in the logic graph has associated judgment nodes, and each judgment node indicates the conditions that one or more business objects must satisfy. Furthermore, the value of any business object can be obtained based on the business data. Therefore, based on the business data, the execution device can obtain the value of the business object indicated by any judgment node. In this way, the execution device can determine which judgment nodes in the logic graph satisfy the conditions indicated by the business data, thus determining one or more judgment nodes corresponding to the business data. Then, based on the one or more judgment nodes corresponding to the business data, the execution device continues to determine the conclusion nodes related to these judgment nodes, thereby obtaining the first conclusion node corresponding to the business data in the logic graph.

[0085] For example, the business data includes the value of a first business object, and the value of the first business object satisfies the condition indicated by the first judgment node related to the first conclusion node. Furthermore, the value of the first business object is obtained based on the data recorded under the first target field, and the first target field is determined in the database using an AI model based on the first business object and related reference knowledge. The database is used to store the data in the business data.

[0086] In other words, business data typically consists of factual data generated during business operations, often including various types of data (such as cost data, revenue data, and profit data). This business data is pre-organized and stored in a database in a specific format (e.g., tables or graphs). To accurately retrieve the value of the first business object indicated in the first judgment node from the database, and to determine whether the business data satisfies the first judgment node, the execution device can determine the first target field in the database related to the first business object in the first judgment node. Then, based on the data under the first target field, the value of the first business object is obtained. Furthermore, the first target field related to the first business object needs to rely on the first business object itself and related reference knowledge, determined in the database through an AI model. That is, the first business object and the first target field may have semantic differences, making it difficult to directly determine the first target field based on the first business object itself. Therefore, it is necessary to combine the reference knowledge related to the first business object to determine the first target field.

[0087] For example, in some business scenarios (such as taxation or finance within the financial sector), logic graphs are typically constructed based on pre-defined rules (such as tax regulations or financial indicator calculation rules). Therefore, the conditions in the decision nodes of the logic graph are often expressed using specialized terminology within these pre-defined rules (e.g., the condition "whether the legal form of the director meets the requirements"). However, business data generated during business operations is often organized and stored in databases using simple, easily understood fields (e.g., "director's name," "age," "years of service"). Therefore, based on the conditions in the decision nodes, it is often difficult to directly determine the corresponding fields in the database for the business objects within those nodes. In this case, by combining the business objects in the decision nodes with the corresponding reference knowledge, and using an AI model to determine the target fields for the business objects in the database, the AI ​​model can better understand the meaning of the business objects through the reference knowledge, thereby accurately determining the corresponding target fields for the business objects. For example, regarding the business object "legal form of a director", the corresponding reference knowledge is a regulation, and the regulation indicates that the legal form of a director is a natural person. Therefore, based on the business object "legal form of a director" and its corresponding regulation, the AI ​​model can determine that the legal form of a director is essentially a natural person, and thus determine that the target field corresponding to the business object "legal form of a director" in the database is the "director's name" field.

[0088] Optionally, the first business object mentioned above can be determined from the set of business objects based on the conditions indicated by the first judgment node. The business objects in the set of business objects are extracted from the set of rules, which is used to guide the construction of the logic graph.

[0089] In other words, to accurately determine the business object in the first judgment node, the execution device can match the first business object mentioned by the first judgment node from multiple business objects included in the business object set based on the conditions indicated by the first judgment node, thereby achieving accurate extraction of the business object. Furthermore, the business objects in the business object set are extracted from the rule set used to guide the construction of the logic graph; therefore, based on the business object set, it is certain that the business objects included in each judgment node can be found, and it is guaranteed that the found business object is precisely the object whose value needs to be obtained in the judgment node.

[0090] Optionally, the reference knowledge related to the first business object includes at least one of the following: rules in the rule set used to extract the first business object (such as one or more regulations in the tax field used to extract the first business object), business cases, expert knowledge, and terminology explanations. Business cases refer to historical reasoning cases of the same or similar business reasoning scenarios, which can indicate the reasoning process for determining whether historical business data meets the conditions related to the judgment node. Expert knowledge can refer to expert interpretations of the conditions indicated by the judgment node, which can help the AI ​​model understand the conditions indicated by the judgment node. Terminology explanations are detailed semantic explanations of some professional terms appearing in the conditions indicated by the first judgment node, used to facilitate the AI ​​model's understanding of the concepts or keywords appearing in the conditions indicated by the first judgment node.

[0091] It should be noted that the above description uses the first judgment node as an example to illustrate how the execution device obtains values ​​for the first business object in the first judgment node. In practical applications, the execution device may obtain values ​​from the database for business objects in multiple judgment nodes in order to determine the conclusion node corresponding to the business data, and each judgment node may include one or more business objects. This application does not impose any specific limitations on this.

[0092] Furthermore, in the steps described above, the execution device can determine the business object in the decision node during business reasoning, thereby determining the target field corresponding to the business object in the database and obtaining the value of the business object. In some examples, determining the business object in the decision node and the target field corresponding to the business object in the database can also be performed in advance during the construction of the logic graph, and the mapping relationship between the business object in the decision node and the target field is recorded in the logic graph, so that the execution device can directly obtain the value of the business object from the database based on the mapping relationship during business reasoning.

[0093] Optionally, the aforementioned business data includes scenario description data. The logic graph used by the execution device when performing business inference is determined from multiple pre-constructed logic graphs based on the scenario description data. Each logic graph has a corresponding scenario label, which indicates the applicable business scenario. That is, when performing business inference, the execution device can combine the scenario labels corresponding to multiple pre-constructed logic graphs and select the corresponding logic graph based on the current actual business scenario, thereby ensuring that the logic graph suitable for the current business scenario is selected.

[0094] Step 203: Execute the device to output the business reasoning result indicated by the first conclusion node.

[0095] After determining the first conclusion node corresponding to the business data, the execution device can output the business reasoning result indicated by the first conclusion node as the result of this business reasoning, thereby completing the business reasoning process.

[0096] In summary, in the business reasoning method provided in this application, the execution device searches for the judgment nodes satisfied by the business data in the logic graph related to the business reasoning request, then determines the conclusion nodes related to these judgment nodes (i.e. the conclusion nodes corresponding to the business data in the logic graph), and finally outputs the business reasoning result indicated by the conclusion nodes.

[0097] In this solution, during business reasoning, reasoning is performed within a logical graph based on business data. This logical graph includes decision nodes indicating the conditions that business objects must meet and conclusion nodes indicating the results of the reasoning. Therefore, by determining the decision nodes that the business data satisfies, the relevant conclusion nodes can be determined, thus yielding the business reasoning results. Furthermore, the values ​​of business objects in the decision nodes are obtained from the corresponding target fields in the database. These target fields are determined using an AI model based on the business objects and their reference knowledge. This ensures that, even with independent storage of enterprise data, the values ​​of business objects in the logical graph can be accurately retrieved, thereby ensuring that business reasoning can be executed automatically and reliably, effectively improving the efficiency of business reasoning.

[0098] The above embodiments, using the first judgment node as an example, illustrate the process of determining the first judgment node corresponding to business data in the logic graph during business reasoning. However, in practical applications, the execution device needs to determine which judgment nodes' conditions the business data satisfies, and then determine the conclusion node corresponding to the business data. That is, business data may satisfy the conditions indicated by multiple judgment nodes in the logic graph, and the execution device may determine in different ways that the business data satisfies the conditions indicated by each judgment node.

[0099] For example, the first conclusion node corresponding to the business data is also related to the second judgment node in the logical graph, and the conditions indicated by the second judgment node related to the first conclusion node include the second business object. The value of the second business object is obtained by the AI ​​model processing the data recorded under the second target field in the database.

[0100] In other words, the value of the second business object cannot be obtained directly by extracting the data recorded under the second target field. Instead, the AI ​​model needs to further process the data recorded under the second target field. For example, assuming the second business object is "profit," and the second target fields include "revenue" and "cost," the second business object "profit" can establish an indirect relationship with the "revenue" and "cost" fields. This allows the subsequent execution device to obtain the actual data of the second business object "profit" by subtracting the data under the "cost" field from the data under the "revenue" field.

[0101] Optionally, business data may also include business description text not stored in the database. That is, a portion of the business data is stored in the database, while the business description text can be provided directly with a business reasoning request. For example, the business description text might be a contract or a business plan provided when a user issues a business reasoning request. The conditions indicated by the third judgment node related to the first conclusion node include a third business object, the value of which is extracted from the business description text by an AI model.

[0102] That is, the execution device cannot retrieve the fields related to the third business object from the database, and therefore cannot obtain the value of the third business object from the database. Instead, it needs to extract the value of the third business object based on the business description text outside the database.

[0103] For example, suppose the third business object is the source of revenue, but the database does not record fields related to the source of revenue, and the business description text related to the business reasoning request includes "revenue A comes from country A", then the execution device can use an AI model to extract the actual data of the third business object as "country A" from the business description text.

[0104] Optionally, the logic graph records the value range of the third business object, which is used to provide constraints for the AI ​​model when extracting the value of the third business object.

[0105] In other words, in order to enable the AI ​​model to extract accurate values ​​for the third business object, the device responsible for constructing the logic graph (hereinafter, the example of the execution device being responsible for constructing the logic graph) can obtain the value range of the third business object in advance and record the value range of the third business object in the logic graph, so as to provide constraints for the AI ​​model when extracting the value of the third business object.

[0106] In this way, when the execution device uses the AI ​​model to extract values ​​of the third business object from the business description text, the extracted actual values ​​must be within the value range of the third business object, and cannot exceed it. For example, if the third business object is a yes / no type business object, then the value range of the third business object is yes or no; if the third business object is a categorical type business object, then the value range of the third business object is discrete values; if the third business object is a numeric type business object, then the value range of the third business object is a certain numeric range.

[0107] The value range of the third business object can be provided by the user. For example, the execution device can generate a prompt message to instruct the user to specify the value range of the third business object for which the relevant field cannot be retrieved; then, the user can specify the value range of the third business object by issuing a value constraint instruction. Alternatively, the user can predefine the value range of one or more business objects in a set of business objects, so that the execution device can obtain the value range of the third business object by searching for it in the set of business objects.

[0108] Furthermore, in the logic graph, different decision nodes often indicate different conditions. Therefore, for different decision nodes, the execution device may need to use different tools to determine whether the business data meets the conditions indicated by the decision node. In this case, the logic graph can pre-record the tools required to process the decision nodes, so that the execution device can call the relevant tools to process the decision nodes.

[0109] For example, the logic graph records the target tools to be invoked when processing the first decision node. The target tools are determined based on the conditions indicated in the first decision node. These target tools include one or more of AI models, rule engines, and analysis operators.

[0110] Therefore, when the execution device determines whether the business data meets the conditions indicated by the first judgment node, the execution device can call the logic graph to process the target tool recorded by the first judgment node, thereby determining that the aforementioned business data meets the conditions indicated by the first judgment node.

[0111] Optionally, if the target tool includes an AI model, a correspondence between the first business object and reference knowledge is also established in the logic graph. This correspondence is used to instruct the AI ​​model to process the first judgment node by referring to the reference knowledge during the business reasoning process.

[0112] In this way, by retrieving relevant reference knowledge for business objects in the knowledge base, the AI ​​model can be provided with relevant reference knowledge when it is called to process judgment nodes during the business reasoning process. This makes it easier for the AI ​​model to understand and analyze the conditions in the judgment nodes, and helps to improve the accuracy of the AI ​​model's reasoning.

[0113] Optionally, in some embodiments, the reference knowledge for the first business object includes multiple types of knowledge, and the correspondence between the first business object and the reference knowledge is further used to indicate the reference priority of each type of knowledge among the multiple types of knowledge. That is, the execution device retrieves multiple relevant types of knowledge as reference knowledge for the business object in the knowledge base. Furthermore, the execution device can assign a reference priority to each type of knowledge based on the degree of relevance between the multiple types of knowledge and the business object, or the newness or oldness of the multiple types of knowledge.

[0114] In this way, when the execution device calls the AI ​​model to process the first judgment node during the business reasoning process, it can specify the reference priority of each knowledge in the input prompt words of the AI ​​model, so that the AI ​​model can refer to each knowledge in a targeted manner to process the first judgment node.

[0115] For example, suppose the reference knowledge corresponding to a business object includes three types of knowledge: expert knowledge, a regulation, and a business case. The expert knowledge is the most recently added interpretation by the expert, the regulation is strongly relevant to the business object, and the business case is a case provided a long time ago. Therefore, for these three types of knowledge, the execution device can record expert knowledge and regulations as high reference priority, and the business case as low reference priority. In this way, during subsequent business reasoning, the execution device can include these three types of knowledge in the prompts and instruct the AI ​​model to prioritize the expert knowledge and regulation among these three types of knowledge.

[0116] In this solution, when a business object corresponds to multiple types of knowledge, by recording the reference priority of each type of knowledge, the AI ​​model can be instructed to refer to each type of knowledge based on the corresponding reference priority during the business reasoning process. This ensures that the AI ​​model can focus on referring to strongly related knowledge, thereby improving the accuracy of the AI ​​model's reasoning.

[0117] It should be noted that for each judgment node in the logic graph that needs to be processed by calling the AI ​​model, the execution device can select and record the corresponding reference knowledge for these judgment nodes based on the business objects in the judgment node, so that the corresponding reference knowledge can be referenced when calling the AI ​​model to process the judgment node.

[0118] The above describes the process of the execution device performing business reasoning. For ease of understanding, the process of the execution device constructing a logic graph will be described below. Of course, in some embodiments, the logic graph may also be constructed by other devices, and the execution device is responsible for acquiring the constructed logic graph to perform business reasoning. Specifically, please refer to Figure 3A, which is a schematic diagram of a logic graph construction process provided in this application. As shown in Figure 3A, the process of the execution device constructing the logic graph includes the following steps 301-304.

[0119] Step 301: Obtain business description data. Business description data is used to describe the judgment logic and conclusions in the business reasoning process.

[0120] In this application, the business description data may include multimedia data such as text, images, voice, or video provided by the user. When the business description data includes data in the form of images, voice, or video, the execution device can obtain the judgment logic or conclusion in the business reasoning process by converting the images, voice, or video into text.

[0121] Business description data describes the judgment logic for various business objects within a specific business scenario, and the conclusions drawn when these judgments are satisfied. For example, when the logic graph to be constructed is applied to a corporate tax rate reasoning scenario in the tax field, the business description data can explain how to infer the tax rate of a company's income based on its operating information. Similarly, when the logic graph to be constructed is applied to a disease reasoning scenario in the medical field, the business description data can explain how to infer the disease by combining patient symptom information. Furthermore, when the logic graph to be constructed is applied to a corporate profit reasoning scenario in the financial field, the business description data can explain how to infer the company's net profit by combining information such as revenue and expenses.

[0122] Step 302: Based on the business description data, a logic graph is constructed by analyzing the business objects involved in the business reasoning process and the judgment logic and conclusions related to the business objects. The logic graph includes judgment nodes and conclusion nodes. Judgment nodes are used to indicate the conditions that the business objects need to meet, and conclusion nodes are used to indicate the business reasoning results. The business corresponding to the business reasoning results meets the conditions indicated by the judgment nodes related to the conclusion nodes.

[0123] In this application, the judgment logic in the business reasoning process described by the business description data often involves multiple business objects, and these judgment logics are all implemented based on the business objects. These business objects can be, for example, specific information about a particular business (such as company revenue or company costs), or the status of a particular business (such as whether a company's shareholders are tax residents of a certain country). Therefore, the execution device analyzes the business objects involved in the business reasoning process, and, taking the business objects as the main body, determines the judgment logic and conclusions related to each business object based on the judgment logic of the business reasoning process described by the business description data. This allows the generation of judgment nodes based on the judgment logic related to each business object, enabling the judgment nodes to indicate the conditions that the business object must meet. In other words, this application uses business objects as a medium to configure judgment nodes based on business objects, thereby forming judgment nodes used to indicate business judgment logic.

[0124] A logic graph can include multiple decision nodes and multiple conclusion nodes. Each decision node indicates a condition that the business objects involved in the business reasoning process must satisfy; different decision nodes indicate different conditions. That is, decision nodes are actually used to indicate the decision logic in the business reasoning process, in order to determine which conditions a specific business object actually meets. It should be noted that a decision node can include one or more business objects and can indicate one or more conditions that each business object must satisfy. Different decision nodes can include the same or different business objects to indicate different conditions.

[0125] Furthermore, each conclusion node indicates the result of business reasoning. Each conclusion node may be associated with one or more decision nodes, and the decision nodes associated with each conclusion node are different.

[0126] Optionally, the business description data used to construct the logic graph can include multiple description data points, which are connected in a tree structure. This allows for the generation of a corresponding node (e.g., a judgment node or a conclusion node) for each description data point, maintaining the tree-like connections between the generated nodes in the logic graph. If a description data point describes judgment logic, the execution device can use an AI model to analyze the business object within that data point, generating conditions for that business object and thus a corresponding judgment node. For example, the execution device can generate prompts based on the description data, instructing the AI ​​model to analyze the given description data to identify the business object involved and generate the conditions that the business object must meet. By inputting the prompts into the AI ​​model, the execution device can obtain the conditions indicated by the judgment node. If a description data point describes a conclusion, the execution device can use an AI model to organize this conclusion and generate a corresponding conclusion node.

[0127] For example, the logic graph constructed by the execution device is a tree-connected graph, which includes multiple layers of decision nodes and one layer of conclusion nodes. Each layer of decision nodes in the multiple layers can include one or more decision nodes. Each decision node in the last layer of the multiple layers is connected to a conclusion node, and the decision nodes associated with the conclusion node include all decision nodes on the path from the root node of the logic graph to the conclusion node.

[0128] In other words, a logic graph is a tree-like connection diagram containing multiple nodes, with these nodes connected through a tree-like relationship. That is, the logic graph organizes objects in a parent-child hierarchical structure. Nodes in a logic graph can have parent nodes or child nodes. Furthermore, a parent node can have one or more child nodes, while each child node can only have one parent node. Among the multiple nodes included in the logic graph are two types: decision nodes and conclusion nodes. Any decision node can have either a parent node or a child node. Conclusion nodes are leaf nodes in the logic graph, do not have child nodes, and the parent node of a conclusion node must be a decision node.

[0129] For example, please refer to Figure 3B, which is a schematic diagram of a logic graph in a tax rate reasoning scenario provided by this application. As shown in Figure 3B, in the logic graph, the connection relationship between nodes is constructed in a tree-like connection graph, thereby constructing multiple layers of judgment nodes and one layer of conclusion nodes.

[0130] For example, in the first-level judgment node of the logic graph, there is only one judgment node "taxable entity is enterprise", and this judgment node is the root node, which means that the taxable entity targeted by this logic graph is an enterprise.

[0131] In the second-level decision node of the logic graph, there are two decision nodes, indicating the following two conditions respectively: "Revenue comes from country A" and "Revenue comes from country B". At this time, the business object of both decision nodes is the country from which the revenue comes.

[0132] Similarly, the third-level decision nodes in the logic graph include decision nodes such as "revenue from sales of goods," "revenue from donations received," and "revenue from providing services," which are used to represent the conditions related to the company's revenue sources. At this time, the business object of each decision node in the third-level decision nodes is the source of revenue.

[0133] In the fourth-level decision nodes of the logic graph, there are decision nodes such as "Amount is in range 1", "Amount is in range 2", "Amount is in range 3", and "Amount is in range 4", which are used to represent conditions regarding the size of the company's revenue amount. At this point, the business object of each decision node in the fourth-level decision nodes is the amount of revenue.

[0134] The last node in the logic graph is the conclusion node, which represents the specific tax rate of the company's revenue when the conditions indicated by the relevant judgment nodes are met.

[0135] For the first conclusion node, all judgment nodes from the root node of the logic graph to the first conclusion node are the judgment nodes related to that conclusion node. That is, for the reasoning path corresponding to the first conclusion node, this reasoning path passes through the judgment node "taxable entity is an enterprise", the judgment node "enterprise income comes from country A", the judgment node "sales revenue", and the judgment node "amount is in range 1" in sequence. Therefore, the business data matching the first conclusion node must meet the conditions indicated by all judgment nodes on the reasoning path. In other words, assuming that in a real business reasoning scenario, to reason about the tax rate corresponding to income A, if income A satisfies the following conditions: the taxable entity is an enterprise (i.e., income A is enterprise income), income A comes from country A, income A is sales revenue, and the amount of income A is in range 1, then it can be reasoned that the tax rate of income A is the tax rate value 1.

[0136] Step 303: Perform a retrieval in the database based on the business object in the judgment node to obtain the target field related to the business object. The target field is the object used in the database to organize data storage.

[0137] Since the decision nodes essentially indicate conditions for business objects, and in actual business reasoning scenarios, business objects often have corresponding values, the actual business reasoning process involves determining which conditions indicated by the decision nodes are met based on the actual values ​​of the business objects, thereby achieving business reasoning. Optionally, to clearly identify the business objects that need to be retrieved for each target field when retrieving the corresponding target field in the database, the execution device can determine the business objects from the set of business objects based on the conditions indicated by the decision nodes. The business objects in the set of business objects are extracted from the rule set used to guide the construction of the logical graph; for example, the business description data mentioned above is obtained based on the rule set.

[0138] Furthermore, the actual values ​​of business objects are business data acquired or generated by enterprises and other organizations during normal operations, and this business data is often organized and stored in a database in a certain way. Therefore, in order to quickly obtain the actual values ​​of business objects indicated in each decision node during actual business reasoning, the execution device can perform a retrieval in the database based on the business objects in each decision node to obtain the target fields related to the business objects. Specifically, business data in the database is organized and stored in the form of fields. For example, when an enterprise uses tables to store actual operating data, the "income" field stores the actual income data of the current enterprise; the "cost" field stores the actual cost data of the current enterprise; in this case, for the business object "income amount", the "income" field related to the business object "income amount" can be retrieved.

[0139] Specifically, during the retrieval process, the names of fields in the database can indicate the attributes of the data recorded under those fields. For example, if a field is named "income," "cost," or "profit," then semantic search or AI model retrieval can be used to find the semantically closest field for the business object as the relevant target field. Alternatively, if a field in the database has a corresponding semantic label that describes the attributes of the data recorded under that field, such as "Field 1 is used to record the company's revenue data," then semantic search or AI model retrieval can be used to find the semantically closest label for the business object, and the field corresponding to the retrieved semantic label can then be used as the relevant target field for the business object.

[0140] However, in some scenarios, the conditions in the decision nodes of the logic graph are expressed using specialized terminology within predefined rules, while business data generated during enterprise operations is often organized and stored in the database using simple and easily understood fields. In this case, the execution device can first obtain reference knowledge about the business objects in the decision nodes, and then, based on the business objects and their reference knowledge, use an AI model to retrieve the target fields corresponding to the business objects from the database.

[0141] Step 304: Establish the correspondence between business objects and target fields in the logic graph. The correspondence is used to indicate the value of the business object based on the data recorded under the target field.

[0142] Specifically, this step involves recording the correspondence between business objects and fields in the logic graph. This allows the execution device to directly retrieve the corresponding business data from the database based on the recorded correspondence during the business reasoning process, ensuring the smooth and efficient completion of the business reasoning process. Furthermore, by only recording the correspondence between business objects and fields, the business data recorded under each field can be updated over time without affecting the normal retrieval of business object values ​​during the business reasoning process, thus enhancing the applicability of the logic graph.

[0143] For example, during normal operations, enterprises can continuously update the business data under each field in the database. When business reasoning needs to be performed, based on the correspondence between the business objects and the target fields recorded in the logic graph, they can obtain the latest actual data of the business objects from the database, thereby enabling the business reasoning process to be completed based on the same logic graph at different times.

[0144] In practical implementation, the execution device can add edges to the logical graph to record the correspondence between business objects and fields in the database. For example, after determining the correspondence between a business object and a certain field, the execution device adds an edge to the logical graph for that business object, and this edge has an edge label to indicate that the business object has a correspondence with a certain field in the database, so that it can be identified later to retrieve the value of the business object from a certain field in the database.

[0145] To facilitate understanding, the following will explain how to establish the correspondence between business objects so that the execution device can obtain the actual value of the business object during the actual business reasoning process based on the correspondence.

[0146] Implementation Method 1: If the field used to record the actual value of the business object is retrieved from the database, a mapping relationship between the business object and the field is established.

[0147] For example, the business objects in the logical graph include a first business object, the aforementioned target field includes a first target field, and the correspondence between the business object and the target field includes a mapping relationship between the first business object and the first target field. Specifically, the mapping relationship between the first business object and the first target field indicates that the data recorded under the first target field is used as the value of the first business object.

[0148] Specifically, the first target field is a field that is semantically identical to the first business object, and the data recorded under the first target field is the actual data of the first business object. Therefore, the execution device can record the relationship between the first target field and the first business object as a mapping data retrieval relationship, indicating that the value of the first business object can be obtained by directly retrieving the data under the first target field.

[0149] For example, assuming the first business object is enterprise revenue and the first target field is the "income" field, a mapping relationship can be established between the "enterprise revenue" business object and the "income" field, so that the execution device can obtain the value of the "enterprise revenue" business object by retrieving the data under the "income" field.

[0150] Implementation Method 2: If the field used to record the actual value of the business object is not found in the database, but the retrieved field records data that can be processed to obtain the actual value of the business object, then an indirect association relationship is established between the business object and the field.

[0151] For example, the business object in the logic graph includes a second business object, the aforementioned target field includes a second target field, and the correspondence between the business object and the target field includes the indirect association between the second business object and the second target field. The indirect association between the second business object and the second target field indicates how the value of the second business object is obtained by processing the data recorded under the second target field. Furthermore, the second target field may include one or more fields.

[0152] In other words, the data recorded under the second target field is not actually the actual data corresponding to the second business object, but the actual data corresponding to the second business object can be inferred based on the data recorded under the second target field. That is, the database does not directly record the actual data corresponding to the second business object, but it can indirectly infer the actual data corresponding to the second business object based on the data under other fields. For example, when the execution device retrieves target fields related to the second business object from the database using an AI model based on the second business object and its reference knowledge, the AI ​​model can analyze the actual meaning of the second business object based on its reference knowledge, and determine, by analyzing the meaning of each field in the database, that the value of the second business object requires processing the data under certain fields in the database.

[0153] For example, suppose the second business object is "profit", and the second target fields include "revenue" and "cost". In this case, the second business object "profit" can establish an indirect relationship with the "revenue" and "cost" fields, so that the subsequent execution device can obtain the actual data of the business object "profit" by subtracting the data under the "cost" field from the data under the "revenue" field.

[0154] It should be noted that since there is no direct semantic relationship between the second business object and the second target field, the execution device can retrieve the second target field for the second business object through an AI model. For example, if the execution device cannot retrieve a field that has a direct mapping relationship with the second business object, it can construct a prompt word and input it into the AI ​​model. This prompt word includes the second business object and the fields or semantic tags of fields existing in the database. The prompt word is used to instruct the AI ​​model to retrieve fields that have an indirect relationship with the second business object from a semantic perspective, so that the actual data of the second business object can be inferred based on these indirectly related fields.

[0155] In implementation method 3, if the fields related to the business object cannot be retrieved from the database, a correspondence between the business object and the AI ​​model is established.

[0156] For example, the business objects in the logic graph include a third business object, and no fields related to the third business object are found in the database. Then, the execution device can establish a correspondence between the third business object and the AI ​​model in the logic graph. This correspondence is used to indicate how to extract the value of the third business object from the input data of the logic graph using the AI ​​model.

[0157] In other words, the actual data of the third business object is not organized and stored in the database as fields, and the actual data of the third business object cannot be deduced based on the data recorded in the database. Therefore, the actual data of the third business object often needs to be extracted from the input data of the logic graph during the business reasoning process. That is, it can only be obtained from the user-provided input data (essentially the business data that needs to be used for business reasoning) when business reasoning is actually executed. Therefore, the execution device can establish a correspondence between the third business object and the AI ​​model to indicate that the value of the third business object needs to be extracted from the input data of the business reasoning process through the AI ​​model.

[0158] The above describes how to establish a correspondence between business objects and data sources in the logic graph during its construction, enabling rapid retrieval of business object values ​​during business reasoning. The following will describe how to configure corresponding processing tools for decision nodes in the logic graph during its construction, facilitating the rapid invocation of these tools during business reasoning to process the decision nodes.

[0159] For example, after constructing the decision nodes in the logic graph, the execution device can determine the target tools to be invoked during the business reasoning process by referring to the tool library based on the conditions indicated in the decision nodes. The tool library includes various tools that can be invoked. A target tool is a tool used to determine whether the input data of the logic graph satisfies the conditions indicated by the decision nodes. Target tools include one or more of AI models, rule engines, and analysis operators. AI models are used to understand and analyze semantic problems; rule engines are used to handle problems composed of logical or arithmetic expressions; and analysis operators are used to handle problems that depend on specific algorithms.

[0160] Furthermore, the execution device also records the target tool corresponding to the judgment node in the logic graph. In this way, when the execution device performs the business reasoning process based on the logic graph, it can call the target tool corresponding to the judgment node to perform the processing of the judgment node, so as to determine whether the input data of the logic graph meets the conditions indicated by the current judgment node.

[0161] Specifically, to determine the target tool corresponding to the judgment node, the execution device can pre-acquire functional descriptions of various tools to identify the functions that each tool can perform. Then, based on the conditions indicated by the judgment node and the functional descriptions of the various tools, the execution device determines the target tool that the judgment node needs to invoke. For example, the execution device inputs the conditions indicated by the judgment node and the functional descriptions of the various tools into an AI model, leveraging the semantic understanding capabilities of the AI ​​model to select the target tool capable of fulfilling the judgment of the conditions indicated by the current judgment node.

[0162] Generally, different tools are needed to determine whether input data meets different types of conditions. For example, a rule engine processes rules composed of logical and / or arithmetic expressions. Therefore, if the conditions indicated in a judgment node include preset rules composed of logical and / or arithmetic expressions (e.g., A > B, A ≠ C), the execution device can use the rule engine to determine whether the input data meets the preset rules to ensure the accuracy of the reasoning. Logical expressions are meaningful formulas that connect relational expressions or logical quantities using logical operators. Examples of logical operators include "OR", "AND", "NOT", "XOR", "equal to", "not equal to", "greater than", and "less than". Arithmetic expressions are formulas composed of numbers and arithmetic functions, such as "addition", "subtraction", "multiplication", "division", and "square".

[0163] For example, if the conditions indicated in the judgment node include those described in descriptive text (e.g., whether the shareholding ratio meets the requirements of XX regulations), it is often necessary to first understand the semantics expressed in the descriptive text before further determining whether the input data meets the conditions described in the descriptive text. Therefore, the execution device can leverage an AI model with strong semantic understanding capabilities to determine whether the input data meets the conditions described in the descriptive text in the judgment node.

[0164] For example, if the condition indicated in the judgment node is whether the input data meets certain analysis requirements, the execution device needs to use specialized analysis operators to process the input data in order to determine whether the input data meets the conditions indicated in the judgment node. For instance, suppose the input data of the logic graph is code to be reviewed, and the condition indicated by the judgment node is whether the code meets the syntax requirements. Then the execution device needs to use specialized syntax analysis operators to analyze the code to determine whether the code meets the syntax requirements.

[0165] It's important to note that a single decision node may contain multiple different types of conditions, each requiring different tools for analysis and processing. Therefore, a single decision node may correspond to multiple different tools. During business reasoning, for the same decision node, the execution device needs to invoke different tools to analyze and process the different conditions to determine whether the input data satisfies all the conditions indicated by the decision node. Furthermore, when a single decision node corresponds to multiple different tools, the execution device can further establish a correspondence between each part of the conditions within the decision node and the corresponding tool, thus clarifying which tool is responsible for processing each part of the conditions.

[0166] For example, suppose the condition indicated by the judgment node is: A is greater than B and A meets the requirements of XX regulations; in this case, the execution device can record that the tool corresponding to the condition "A is greater than B" is a rule engine, and the tool corresponding to the condition "A meets the requirements of XX regulations" is an AI model.

[0167] Furthermore, for each decision node in the logic graph, the execution device can select and record the corresponding tool for each decision node by analyzing the conditions within that node. For example, after determining the tool required for a certain decision node, the execution device adds an edge to that decision node in the logic graph, and this edge has an edge label to indicate that the decision node corresponds to one or more target tools, so that the tool required to be called when processing the decision node can be identified subsequently.

[0168] In this solution, by analyzing the conditions indicated in the judgment nodes, the tools required to process the judgment nodes are determined and recorded in the logic graph. This facilitates the rapid invocation of the corresponding tools to process the judgment nodes based on the recorded content during business reasoning, avoiding the need to analyze the conditions in the judgment nodes and select the processing tools later during business reasoning, thus effectively improving the efficiency of business reasoning.

[0169] Optionally, when the target tool includes an AI model, in order to improve the accuracy of the AI ​​model in processing decision nodes, the execution device can also perform a retrieval in the knowledge base based on the business object in the decision node to obtain reference knowledge related to the business object. The reference knowledge includes at least one of business cases, expert knowledge, preset rules, and terminology explanations.

[0170] Then, the execution device establishes a correspondence between business objects and reference knowledge in the logic graph to instruct the AI ​​model to process judgment nodes by referring to the reference knowledge during the business reasoning process.

[0171] Optionally, in some embodiments, the execution device may also generate scene tags corresponding to the logic graph based on the business description data, and add the logic graph and scene tags to the logic graph set. The logic graph set includes multiple constructed logic graphs. The scene tags corresponding to the logic graphs are used to indicate the business scenarios to which the logic graphs are applicable.

[0172] Specifically, the execution device can construct corresponding logic graphs for different business scenarios, resulting in a logic graph set containing multiple logic graphs. Furthermore, to facilitate the selection of appropriate logic graphs during business inference, the execution device generates scenario labels for each logic graph based on the business description data used to construct the logic graphs during the graph construction process. These scenario labels then represent the business scenario used by each logic graph. In this way, when performing business inference, the execution device can combine each scenario label in the logic graph set and select the corresponding logic graph based on the current actual business scenario, thereby ensuring the applicability of the logic graph-based business inference process across various scenarios.

[0173] To facilitate understanding, the business reasoning method provided in this application will be explained in detail below with specific examples.

[0174] Please refer to Figure 4, which is a schematic diagram of a system architecture for constructing a logical graph according to this application. As shown in Figure 4, the system architecture includes a logical layer, a semantic layer, and a physical layer. The logical layer stores the logical graph, which consists of two types of nodes: judgment nodes and conclusion nodes. The semantic layer records the business objects involved in the nodes of the logical graph. Furthermore, the semantic layer includes semantic analysis tools to perform semantic analysis on the conditions in the judgment nodes to determine the type of condition. The physical layer includes a knowledge base, a database, and a tool library.

[0175] Specifically, based on the business objects in the semantic layer, the execution device retrieves relevant reference knowledge from the knowledge base and value data related to the business objects from the database, enabling it to obtain the actual values ​​of the business objects during business reasoning. Based on semantic analysis tools (such as large language models) in the semantic layer, the execution device can analyze the types of conditions in the judgment nodes, thereby linking the judgment nodes to the tools required for business reasoning (such as AI models, rule engines, or analysis operators).

[0176] For example, please refer to Figure 5, which is a schematic diagram of establishing correspondences in a logic graph according to this application. As shown in Figure 5, for any decision node in the logic graph, a retrieval can be performed in the database based on the business objects included in the decision node to obtain the target fields related to the business objects. These target fields are used to directly or indirectly record the value data of the business objects. At this time, a correspondence between the business objects and the target fields in the database (such as the direct mapping relationship or indirect association relationship mentioned above) can be established in the logic graph so that the actual value of the business objects can be obtained based on these correspondences during the actual business reasoning process.

[0177] Furthermore, if the fields related to the business object cannot be retrieved in the database, a correspondence between the business object and the AI ​​model can be established so that the AI ​​model can be used to extract the actual value of the business object from the input data of the logic graph during the actual business reasoning process.

[0178] Furthermore, for any given decision node, the execution device can analyze the type of conditions within the decision node to establish a correspondence between the decision node and tools in the tool library, enabling the execution device to call the corresponding tool when processing the decision node. Optionally, to improve the accuracy of the AI ​​model in processing decision nodes, the execution device can also search the knowledge base based on the business objects involved in the conditions of the decision node, thereby establishing a correspondence between business objects and reference knowledge, so that reference knowledge to assist the AI ​​model in reasoning can be extracted when calling the AI ​​model to process the decision node.

[0179] Specifically, please refer to Figure 6, which is a schematic diagram of a logic graph construction process provided in this application. As shown in Figure 6, the process of constructing the logic graph includes the following steps 601-6011.

[0180] Step 601: Obtain business description data.

[0181] Step 601 is similar to step 201 above. Please refer to step 201 above for details, which will not be repeated here.

[0182] Step 602: Extract the business objects involved in the business reasoning process by analyzing the business description data.

[0183] The business description data can include multiple description data, which are connected in a tree structure. Therefore, for each description data, the execution device can extract one or more business objects from the description data to determine the subject in the judgment logic indicated by each description data.

[0184] Step 603: Construct judgment nodes and conclusion nodes based on business objects to obtain a logic graph.

[0185] For the extracted business objects and the content described by the description data of the business objects, the execution device can establish a corresponding judgment node or conclusion node for each description data to obtain a logic graph, and the nodes in the logic graph maintain a tree-like connection relationship (i.e., a tree-like connection relationship between multiple description data).

[0186] For example, please refer to Figure 7, which is a schematic diagram of constructing a judgment node based on a business object according to this application. As shown in Figure 7, assume that a certain descriptive data in the business description data is used to indicate whether the business meets the agreement's 0% tax rate requirement. For the specific content in the descriptive data, the business objects extracted by the execution device include: "country where the subsidiary is located", "shareholding ratio", and "whether the shareholder is a tax resident of country A". Furthermore, based on the extracted business objects, the execution device can generate the conditions that these business objects must meet, thereby forming judgment rules. The conditions that the business objects must meet are: "country where the subsidiary is located = country B", "shareholding ratio ≥ 25%", and "whether the shareholder is a tax resident of country A = yes". That is, the execution device can generate judgment rules composed of logical expressions based on these business objects, so that the rule engine can be used to implement the processing of the judgment node.

[0187] Generally, for any given decision node, the conditions generated by the execution device can be categorized into three types based on their degree of structure. For example, conditions in a decision node can be structured rules, semi-structured rules, or unstructured rules. If the conditions in a decision node are structured rules, it means the conditions are expressed through logical and / or arithmetic expressions; therefore, the decision node can rely on a pre-built rule engine for reasoning. If the conditions indicated by the decision node are unstructured rules, it means the conditions are often described through text; therefore, the decision node needs to utilize a large language model for reasoning. If the conditions indicated by the decision node are semi-structured rules, it means some conditions are expressed through logical and / or arithmetic expressions, while others are described through text; therefore, the decision node needs to utilize both a rule engine and a large language model for reasoning.

[0188] It should be noted that different decision nodes may have the same business object or different business objects, and no specific limitation is made here.

[0189] Step 604: Retrieve the corresponding data source for the business object in the judgment node in the database.

[0190] After extracting the business objects, the execution device can perform retrieval in the database based on the business objects, thereby retrieving the corresponding data source for each business object. The business data in the database is typically organized and stored in the form of fields. Furthermore, the names of the fields themselves or the semantic tags corresponding to the fields may indicate the attribute information of the data stored under those fields. Therefore, the execution device can use semantic retrieval to retrieve semantically related fields for the business objects, thereby finding the corresponding data source for each business object.

[0191] Step 605: Configure the mapping relationship between business objects and data sources.

[0192] For a business object that has been retrieved from a corresponding data source, the execution device can configure the mapping relationship between the business object and the data source in the logical graph. This allows the device to obtain the actual value of the business object based on the established mapping relationship during subsequent business reasoning. For example, for a business object that has been retrieved from a data source, the execution device can add an edge to the business object in the logical graph. This edge has an edge label to indicate that the business object has a mapping relationship with a certain field in the database.

[0193] Step 606: Configure an AI model-based data retrieval method for business objects for which the data source cannot be retrieved.

[0194] If, for certain business objects, the execution device cannot retrieve the corresponding data source from the database, it means that the values ​​of these business objects often need to be obtained from the input data of the logic graph during the actual business reasoning process. Therefore, the execution device can configure an AI model-based data retrieval method for business objects where the data source cannot be retrieved. This establishes a correspondence between the business object and the AI ​​model, instructing the AI ​​model to extract the actual values ​​of the business object from the input data of the logic graph. In other words, the AI ​​model-based data retrieval method means that the values ​​of the business object need to be extracted based on the AI ​​model.

[0195] Optionally, for business objects configured to retrieve data based on AI models, in order to facilitate the AI ​​model's understanding of the business object and accurately extract the actual value of the business object from the input data of the logic graph, the execution can also retrieve relevant reference knowledge of the business object in the knowledge base and establish a correspondence between the business object and the reference knowledge, so that the AI ​​model can extract the actual value of the business object by referring to the relevant reference knowledge of the business object.

[0196] For example, please refer to Figure 8, which is a schematic diagram of establishing a correspondence between business objects to obtain the actual values ​​of the business objects provided by this application. As shown in Figure 8, for the business objects "company monthly costs" and "company monthly revenue", the execution device can find the fields related to these two business objects in the tabular data of the database, and then establish a correspondence between these two business objects and the fields in the tabular data, so as to obtain the actual values ​​of the business objects from the tables based on the fields corresponding to the business objects. For the business object "whether the country where the subsidiary is located is in Europe", the execution device finds the fields related to this business object in the graph data of the database, and then establishes a correspondence between this business object and the fields in the graph data. For the business object "whether the qualifications of the directors meet the regulatory requirements", since the execution device cannot retrieve the data source related to this business object in the database, the execution device establishes a correspondence between this business object and the AI ​​model and reference knowledge (such as the regulations related to the appointment of directors), so as to extract the actual values ​​of the business objects based on the AI ​​model.

[0197] Step 607: Based on the type of conditions that the business object needs to meet, configure the tools to be called when processing judgment nodes.

[0198] After configuring the data retrieval method for business objects, the execution device can determine the types of conditions that each business object in the judgment node must meet, and configure the tools to be invoked when processing the judgment node based on the type of conditions. Specifically, for different types of conditions, the execution device often needs to use different types of tools to perform the processing to ensure that the condition judgment can be accurately achieved.

[0199] For example, please refer to Figure 9, which is a schematic diagram of the tools to be called by a configuration judgment node provided in this application. As shown in Figure 9, a judgment node may include one or more conditions. The execution device can determine the type of each condition, and thus determine the tools to be called by the judgment node. For example, judgment node 1 includes condition 1 and condition 2, where condition 1 is a semantic understanding type condition (e.g., the purpose of income A is XX), and condition 2 is an expression operation type condition (e.g., the amount of income A is greater than or equal to XX). Therefore, the execution device can establish a correspondence between the semantic understanding type condition 1 and the AI ​​model to instruct the business reasoning process to call the AI ​​model to determine whether the input data meets condition 1. Furthermore, the execution device can establish a correspondence between the expression operation type condition 2 and the rule engine to instruct the business reasoning process to call the rule engine to determine whether the input data meets condition 2, so as to ensure high accuracy of reasoning. In addition, for condition 3 in judgment node 2, condition 3 is specifically an algorithm analysis type condition (e.g., whether the code to be reviewed meets the code syntax requirements). At this time, the execution device can establish a correspondence between the algorithm analysis type condition 3 and a certain analysis operator, so as to facilitate the subsequent call of the analysis operator to implement the processing of judgment node 2.

[0200] It is worth noting that if the execution device is configured with an AI model for a certain judgment node, the execution device can further retrieve relevant reference knowledge from the knowledge base based on the conditions in that judgment node, so that the retrieved reference knowledge can be referenced when the AI ​​model is called to perform condition judgments, thereby improving the accuracy of the AI ​​model's judgment.

[0201] Step 608: Run the logic graph based on the trial data to obtain the trial results.

[0202] After configuring the data retrieval methods and processing tools for all decision nodes, the execution device runs the logic graph based on the trial calculation data provided by the user (i.e., the input data for testing) to determine whether the logic graph can be used to execute the normal business reasoning process.

[0203] For example, please refer to Figure 10, which is a schematic diagram of processing different types of judgment nodes provided in this application. As shown in Figure 10, for different types of judgment nodes, the corresponding processing method of the judgment node can be determined based on the degree of data structuring and rule structuring of the judgment node.

[0204] Specifically, if the business object in the judgment node is structured data (i.e., the business object has a direct mapping relationship with the data in the database), and the condition in the judgment node is a structured rule (i.e., the condition is represented by a rule composed of expressions), then the execution device can obtain the actual value of the business object through mapping and data retrieval, and perform rule reasoning through the rule engine to complete the processing of the judgment node.

[0205] If the business object in the judgment node is structured data and the condition in the judgment node is a semi-structured rule (i.e., the condition is represented by a rule composed of expressions and a text description), then the execution device can obtain the actual value of the business object by mapping data, and perform rule reasoning through the rule engine and semantic reasoning using an AI model to complete the processing of the judgment node.

[0206] If the business object in the judgment node is structured data and the condition in the judgment node is an unstructured rule (i.e., the condition is represented as a text description), then the execution device can obtain the actual value of the business object by mapping data retrieval and use an AI model to perform semantic reasoning, thereby completing the processing of the judgment node.

[0207] Similarly, if the business object in the determination node is semi-structured data (i.e., the business object has an indirect relationship with the data in the database), the execution device can combine mapping data retrieval with an AI model to perform semantic transformation on the data obtained from the mapping data retrieval (e.g., subtracting cost data from revenue data to obtain profit data) to obtain the actual value of the business object. If the business object in the determination node is unstructured data (i.e., the business object does not have a corresponding relationship with the data in the database), the execution device can extract data from the input data based on an AI model to obtain the actual value of the business object.

[0208] In general, in practical applications, how the execution device obtains the value of the business object is related to the degree of data structuring of the business object, and how the execution device processes the conditions in the judgment node is also related to the degree of rule structuring of the conditions in the judgment node.

[0209] Step 609: Determine if there are any abnormalities in the trial calculation results.

[0210] After running the logic diagram based on the trial calculation data, the execution device can obtain the corresponding trial calculation results. At this point, the execution device can determine whether there are any anomalies in the trial calculation results, and thus determine whether further adjustments to the logic diagram are needed.

[0211] Specifically, if the trial calculation result indicates that the execution device cannot derive a conclusion based on the trial calculation data, it means that there is a problem with the data retrieval method or processing tool configuration of the judgment node in the logic diagram. The execution device can generate an error message to prompt the user to actively adjust the configuration of the data retrieval method or processing tool of the judgment node.

[0212] If the trial calculation results in an incorrect conclusion derived by the execution device based on the trial calculation data, it indicates that the conditions in the decision nodes of the logic diagram may be incorrect, or that some reference knowledge configured for the decision nodes in the logic diagram is incorrect, or that the reference priority of the reference knowledge is incorrect. In this case, the execution device can also generate a trial calculation error message to prompt the user to actively adjust the conditions of the decision nodes or the configured reference knowledge.

[0213] Step 6010: Adjust the logic diagram.

[0214] In the event of abnormal trial results (such as trial results not matching actual business reasoning results, or trial results indicating that business results cannot be obtained), the execution device can adjust the logic graph based on the user's adjustment instructions, such as adjusting the conditions indicated by the judgment nodes in the logic graph, or adjusting the data source or processing tools configured for the judgment nodes in the logic graph.

[0215] Of course, in some cases, the execution device can also automatically detect and adjust abnormal locations in the logic graph using an AI model. For example, the execution device can input the constructed logic graph and the trial calculation results into the AI ​​model, instructing the AI ​​model to locate abnormal locations and causes in the logic graph based on the trial calculation results. Furthermore, based on the abnormal locations and causes indicated by the AI ​​model, the execution device then corrects the content in the logic graph according to the AI ​​model.

[0216] It should be noted that after the execution device adjusts the logic diagram, the execution device continues to execute step 608 to rerun the adjusted logic diagram based on the trial calculation data in order to determine whether the adjusted logic diagram is normal.

[0217] Step 6011: Output the logic diagram.

[0218] If the trial calculation result corresponding to the logic graph is normal, it means that the execution device has constructed a normal logic graph that can be used to realize business reasoning, so the execution device outputs the logic graph.

[0219] The method provided in this application has been described in detail above. Next, the device provided in this application for performing the above method will be described.

[0220] Please refer to Figure 11, which is a structural schematic diagram of a business reasoning device provided in this application. As shown in Figure 11, the business reasoning device includes: an acquisition module 1101, used to acquire a business reasoning request, which instructs the execution of business reasoning on business data; a processing module 1102, used to determine, based on the business reasoning request, a first conclusion node corresponding to the business data in a logic graph, the logic graph including multiple judgment nodes and multiple conclusion nodes, the multiple judgment nodes respectively indicating the conditions that the business object must satisfy, the multiple conclusion nodes respectively indicating the business reasoning result, and the business data satisfying the conditions indicated by the judgment nodes related to the first conclusion node; the processing module 1102 is also used to output the business reasoning result indicated by the first conclusion node; wherein, the business data includes the value of a first business object, the value of the first business object satisfying the conditions indicated by the first judgment node related to the first conclusion node, the value of the first business object being obtained based on the data recorded under a first target field, and the first target field being determined in a database based on the first business object and related reference knowledge through an artificial intelligence AI model, the database being used to store the data in the business data.

[0221] In one possible implementation, the first business object is determined from the set of business objects based on the conditions indicated by the first judgment node. The business objects in the set of business objects are extracted from the set of rules, which are used to guide the construction of the logic graph.

[0222] In one possible implementation, the reference knowledge related to the first business object includes at least one of the following: rules, business cases, expert knowledge, and terminology explanations used in the rule set to extract the first business object.

[0223] In one possible implementation, the conditions indicated by the second judgment node related to the first conclusion node include a second business object, and the value of the second business object is obtained by the AI ​​model processing the data recorded under the second target field in the database.

[0224] In one possible implementation, the business data also includes business description text not stored in the database, and the conditions indicated by the third judgment node related to the first conclusion node include a third business object, the value of which is extracted from the business description text by an AI model.

[0225] In one possible implementation, the logic graph records the value range of the third business object, and the value range is used to provide constraints for the AI ​​model when extracting the value of the third business object.

[0226] In one possible implementation, the logic graph records the target tools to be invoked when processing the first decision node, and the target tools are determined according to the conditions indicated in the first decision node; wherein, the target tools include one or more of AI models, rule engines, and analysis operators.

[0227] In one possible implementation, when the target tool includes an AI model, a correspondence between the first business object and reference knowledge is also established in the logic graph. This correspondence is used to instruct the AI ​​model to process the first judgment node by referring to the reference knowledge during the business reasoning process.

[0228] In one possible implementation, the reference knowledge includes multiple types of knowledge, and the correspondence is also used to indicate the reference priority of each type of knowledge among the multiple types of knowledge.

[0229] In one possible implementation, the logic graph is a tree-connected graph, and the logic graph includes multi-level decision nodes in a tree connection. Each decision node in the last level of the multi-level decision nodes is connected to a conclusion node, and the decision nodes associated with the conclusion node include all decision nodes on the path from the root node of the logic graph to the conclusion node.

[0230] In one possible implementation, the business data includes scenario description data, and the logical graph is determined from multiple constructed logical graphs based on the scenario description data. Each logical graph has a corresponding scenario label, which is used to indicate the business scenario to which the logical graph is applicable.

[0231] Both the acquisition module 1101 and the processing module 1102 can be implemented in software or in hardware. For example, the implementation of the processing module 1102 will be described below. Similarly, the implementation of the acquisition module 1101 can be referenced from the implementation of the processing module 1102.

[0232] As an example of a software functional unit, processing module 1102 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, or a container. Further, the aforementioned computing instance may be one or more. For example, module A may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed within the same region or in different regions. Further, the multiple hosts / virtual machines / containers used to run the code may be distributed within the same availability zone (AZ) or in different AZs, each AZ including one or more geographically proximate data centers. Typically, a region may include multiple AZs.

[0233] Similarly, multiple hosts / virtual machines / containers used to run this code can be distributed within the same Virtual Private Cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Communication between two VPCs within the same region, as well as between VPCs in different regions, requires a communication gateway to be set up within each VPC to enable interconnection between VPCs.

[0234] As an example of a hardware functional unit, the processing module 1102 may include at least one computing device, such as a server. Alternatively, the processing module 1102 may be implemented using a central processing unit (CPU), an application-specific integrated circuit (ASIC), or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), a data processing unit (DPU), a neural network processing unit (NPU), a system-on-chip (SoC), an offload card, an accelerator card, or any combination thereof.

[0235] The processing module 1102 includes multiple computing devices that can be distributed within the same region or in different regions. Similarly, the processing module 1102 can be distributed within the same Availability Zone (AZ) or in different AZs. Likewise, the processing module 1102 can be distributed within the same Virtual Private Cloud (VPC) or in multiple VPCs. These multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, GALs, DPUs, NPUs, SoCs, offloading cards, and accelerator cards.

[0236] Please refer to Figure 12, which is a schematic diagram of the structure of a computing device provided in this application. The computing device 1200 shown in Figure 12 can be used to execute the business reasoning method provided in this embodiment. As shown in Figure 12, the computing device 1200 includes: a bus 1202, a processor 1204, a memory 1206, and a communication interface 1208. The processor 1204, the memory 1206, and the communication interface 1208 communicate with each other via the bus 1202. The computing device 1200 can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memories in the computing device 1200.

[0237] Bus 1202 can be a Peripheral Component Interconnect Express (PCIe) bus, or an Extended Industry Standard Architecture (EISA) bus, a unified bus (Ubus or UB), a compute express link (CXL), a cache coherent interconnect for accelerators (CCIX), etc. Bus 1202 can be divided into address bus, data bus, control bus, etc. Among them, the unified bus is also called the Lingqu bus. For ease of representation, only one line is used in Figure 12, but this does not mean that there is only one bus or one type of bus. Bus 1202 can include a path for transmitting information between various components of computing device 1200 (e.g., memory 1206, processor 1204, communication interface 1208).

[0238] The processor 1204 can be implemented using a central processing unit (CPU), a microprocessor (MP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or a programmable logic device (PLD). The PLD can be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), a data processing unit (DPU), a system-on-chip (SoC), or any combination thereof.

[0239] The memory 1206 may include volatile memory, such as random access memory (RAM). The processor 1204 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0240] The memory 1206 stores executable program code, and the processor 1204 executes this executable program code to implement the functions of the aforementioned acquisition module and processing module, thereby realizing the aforementioned business reasoning method. That is, the memory 1206 stores instructions for executing the business reasoning method.

[0241] The communication interface 1208 uses transceiver modules such as, but not limited to, network interface cards and transceivers to enable communication between the computing device 1200 and other devices or communication networks.

[0242] It should be understood that the computing device 1200 of this application is used to execute the business reasoning method shown in Figures 2 to 9, and can correspond to the execution device in executing the method of this application. For the sake of brevity, it will not be described in detail here.

[0243] This application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smartphone.

[0244] Please refer to Figure 13, which is a schematic diagram of a computing device cluster provided in this application. As shown in Figure 13, the computing device cluster includes at least one computing device 1200. The memory 1206 of one or more computing devices 1200 in the computing device cluster may store the same instructions for executing business inference methods.

[0245] In some possible implementations, the memory 1206 of one or more computing devices 1200 in the computing device cluster may also store partial instructions for executing business inference methods. In other words, a combination of one or more computing devices 1200 can jointly execute instructions for executing business inference methods.

[0246] It should be noted that the memory 1206 in different computing devices 1200 within the computing device cluster can store different instructions, each used to execute a portion of the functions of the data processing device. That is, the instructions stored in the memory 1206 of different computing devices 1200 can implement the functions of one or more of the aforementioned acquisition and processing modules.

[0247] In some possible implementations, one or more computing devices in a computing device cluster can be connected via a network. This network can be a wide area network (WAN) or a local area network (LAN), etc. Figure 14 illustrates one possible implementation. Figure 14 is a schematic diagram of another computing device cluster structure provided in this application. As shown in Figure 14, in computing device cluster 1400, two computing devices 1200A and 1200B are connected via a network. Specifically, they are connected to the network through communication interfaces in each computing device. In this type of possible implementation, the memory 1206 in computing device 1200A stores instructions for executing the functions of the acquisition module. Simultaneously, the memory 1206 in computing device 1200B stores instructions for executing the functions of the processing module.

[0248] It should be understood that the functions of computing device 1200A shown in Figure 14 can also be performed by multiple computing devices 1200. Similarly, the functions of computing device 1200B can also be performed by multiple computing devices 1200.

[0249] It should be understood that the computing device 1200 or computing device cluster 1400 in this application may correspond to the business inference device in FIG11 of this application, and may correspond to the execution device that executes the business inference method in FIG2 of this application. Furthermore, the above and other operations and / or functions of each module in the computing device 1200 or computing device cluster 1400 are respectively for implementing the corresponding process of the business inference method in FIG2, and will not be described in detail here for the sake of brevity.

[0250] This application also provides a chip comprising a processing unit and a communication unit. The processing unit may be, for example, a processor, and the communication unit may be, for example, an input / output interface, pins, or circuits. The processing unit can execute computer execution instructions stored in a storage unit to cause the chip within the electronic device to perform the methods described in the above embodiments. Optionally, the storage unit may be an in-chip storage unit, such as a register or cache. Alternatively, the storage unit may be an external storage unit located within a wireless access device, such as a read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, such as random access memory (RAM).

[0251] Specifically, please refer to Figure 15, which is a schematic diagram of the structure of a chip provided in this application. The chip can be represented as a neural processing unit (NPU), a graphics processing unit (GPU), or a tensor processing unit (TPU). The following description will use an NPU 1500 as an example. The NPU 1500 is mounted as a coprocessor on the host CPU, and tasks are assigned by the host CPU. The core of the NPU is the arithmetic circuit 1503, which is controlled by a controller 1504 to retrieve matrix data from memory and perform multiplication operations.

[0252] In some implementations, the arithmetic circuit 1503 internally includes multiple processing engines (PEs). In some implementations, the arithmetic circuit 1503 is a two-dimensional pulsating array. The arithmetic circuit 1503 can also be a one-dimensional pulsating array or other electronic circuits capable of performing mathematical operations such as multiplication and addition. In some implementations, the arithmetic circuit 1503 is a general-purpose matrix processor.

[0253] For example, suppose we have an input matrix A, a weight matrix B, and an output matrix C. The arithmetic circuit retrieves the corresponding data of matrix B from the weight memory 1502 and caches it in each PE of the arithmetic circuit. The arithmetic circuit retrieves the data of matrix A from the input memory 1501 and performs matrix operations with matrix B. The partial result or the final result of the obtained matrix is ​​stored in the accumulator 1508.

[0254] Unified memory 1506 is used to store input and output data. Weight data is directly transferred to weight memory 1502 via Direct Memory Access Controller (DMAC) 1505. Input data is also transferred to unified memory 1506 via DMAC.

[0255] BIU stands for Bus Interface Unit, which is used for interaction between the AXI bus and the DMAC and the Instruction Fetch Buffer (IFB) 1509.

[0256] The Bus Interface Unit (BIU) 1510 is used by the instruction fetch memory 1509 to fetch instructions from external memory, and also by the memory access controller 1505 to fetch the original data of the input matrix A or the weight matrix B from external memory.

[0257] The DMAC is mainly used to move input data from external memory DDR to unified memory 1506, or to weight data to weight memory 1502, or to input data to input memory 1501.

[0258] The vector computation unit 1507 includes multiple processing units that further process the output of the computation circuit 1503 when needed, such as vector multiplication, vector addition, exponential operations, logarithmic operations, size comparisons, etc. It is mainly used for computation in non-convolutional / fully connected layers of neural networks, such as batch normalization, pixel-level summation, and upsampling of feature planes.

[0259] In some implementations, the vector computation unit 1507 can store the processed output vector in the unified memory 1506. For example, the vector computation unit 1507 can apply a linear function, or a nonlinear function, to the output of the computation circuit 1503, such as linear interpolation of feature planes extracted by a convolutional layer, or, for example, a vector of accumulated values, to generate activation values. In some implementations, the vector computation unit 1507 generates normalized values, pixel-level summed values, or both. In some implementations, the processed output vector can be used as activation input to the computation circuit 1503, for example, for use in subsequent layers of the neural network.

[0260] The instruction fetch buffer 1509 connected to the controller 1504 is used to store the instructions used by the controller 1504;

[0261] Unified memory 1506, input memory 1501, weighted memory 1502, and instruction fetch memory 1509 are all on-chip memories. External memory is proprietary to this NPU hardware architecture.

[0262] The processor mentioned above can be a general-purpose central processing unit, a microprocessor, an ASIC, or one or more integrated circuits used to control the execution of the above program.

[0263] It should be understood that the chip in Figure 15 of this application may correspond to the business inference device in Figure 11 of this application, or be deployed on the computing device 1200 or computing device cluster 1400 of this application. Furthermore, the chip in Figure 15 of this application may correspond to the execution device that executes the business inference method in Figure 2 of this application. The above-mentioned and other operations and / or functions of each module in the chip respectively implement the corresponding flow of the business inference method in Figure 2, which will not be elaborated here for the sake of brevity. In addition, in this application, the chip structure is not limited to the chip structure shown in Figure 15, and may include more or fewer hardware structures to implement the functions of the method shown in Figure 2.

[0264] Referring to Figure 16, which is a schematic diagram of the structure of a computer-readable storage medium provided in this application. This application also provides a computer-readable storage medium in which, in some embodiments, the method disclosed in Figure 2 can be implemented as computer program instructions encoded in a machine-readable format on a computer-readable storage medium or on other non-transitory media or articles of art.

[0265] Figure 16 schematically illustrates a conceptual partial view of an example computer-readable storage medium arranged according to at least some of the embodiments shown herein, the example computer-readable storage medium including a computer program for executing computer processes on a computing device. In one embodiment, the computer-readable storage medium 1600 is provided using a signal bearer medium 1601. The signal bearer medium 1601 may include one or more program instructions 1602 that, when executed by one or more processors, can provide the functionality or part of the functionality described above with respect to Figure 2.

[0266] In some examples, the signal carrying medium 1601 may include a computer-readable medium 1603, such as, but not limited to, a hard disk drive, a compact disc (CD), a digital video disc (DVD), a digital magnetic tape, a memory, ROM, or RAM, etc.

[0267] In some embodiments, the signal-bearing medium 1601 may comprise a computer-recordable medium 1604, such as, but not limited to, a memory, a read / write (R / W) CD, a R / W DVD, and so on. In some embodiments, the signal-bearing medium 1601 may comprise a communication medium 1605, such as, but not limited to, digital and / or analog communication media (e.g., fiber optic cables, waveguides, wired communication links, wireless communication links, and so on). Therefore, for example, the signal-bearing medium 1601 may be transmitted by a wireless communication medium 1605 (e.g., a wireless communication medium conforming to the IEEE 1202.X standard or other transmission protocols).

[0268] One or more program instructions 1602 may be, for example, computer-executable instructions or logical implementation instructions. In some examples, the computing device may be configured to provide various operations, functions, or actions in response to one or more program instructions 1602 conveyed to the computing device via a computer-readable medium 1603, a computer-recordable medium 1604, and / or a communication medium 1605.

[0269] It should be understood that the computer-readable storage medium 1600 in this application may be deployed on the business inference apparatus shown in FIG11, or on the computing device 1200 or computing device cluster 1400 of this application. In this way, the business inference apparatus, computing device 1200 or computing device cluster 1400 provided in this application implements the business inference method shown in FIG2 by reading one or more program instructions 1602 on the computer-readable storage medium 1600.

[0270] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the accompanying drawings of the device embodiments provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0271] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods of the various embodiments of this application.

[0272] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product.

[0273] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, computer instructions can be transferred from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0274] The above are merely specific embodiments of this application. Any variations or substitutions conceived by those skilled in the art based on the specific embodiments provided in this application should be covered within the protection scope of this application.

Claims

1. A business reasoning method, characterized in that, The method is executed by an execution device, and the method includes: Obtain a business inference request, the business inference request being used to instruct the performance of business inference on business data; Based on the business reasoning request, a first conclusion node corresponding to the business data in the logic graph is determined. The logic graph includes multiple judgment nodes and multiple conclusion nodes. The multiple judgment nodes are used to indicate the conditions that the business object needs to meet, and the multiple conclusion nodes are used to indicate the business reasoning result. The business data meets the conditions indicated by the judgment node related to the first conclusion node. Output the business reasoning result indicated by the first conclusion node; The business data includes the value of a first business object, the value of which satisfies the conditions indicated by the first judgment node related to the first conclusion node, the value of which is obtained based on the data recorded under the first target field, and the first target field is determined in the database by an artificial intelligence (AI) model based on the first business object and related reference knowledge, and the database is used to store the data in the business data.

2. The method according to claim 1, characterized in that, The first business object is determined from the set of business objects based on the conditions indicated by the first judgment node. The business objects in the set of business objects are extracted from the set of rules, which are used to guide the construction of the logic graph.

3. The method according to claim 2, characterized in that, The reference knowledge related to the first business object includes at least one of the following: rules, business cases, expert knowledge, and terminology explanations used to extract the first business object from the rule set.

4. The method according to any one of claims 1-3, characterized in that, The conditions indicated by the second judgment node related to the first conclusion node include a second business object. The value of the second business object is obtained by the AI ​​model processing the data recorded under the second target field in the database.

5. The method according to any one of claims 1-4, characterized in that, The business data also includes business description text not stored in the database. The conditions indicated by the third judgment node related to the first conclusion node include a third business object. The value of the third business object is extracted from the business description text by the AI ​​model.

6. The method according to claim 5, characterized in that, The logic graph records the value range of the third business object, and the value range is used to provide constraints for the AI ​​model when extracting the value of the third business object.

7. The method according to any one of claims 1-6, characterized in that, The logic graph records the target tools that need to be invoked when processing the first judgment node, and the target tools are determined according to the conditions indicated in the first judgment node; The target tools include one or more of AI models, rule engines, and analytics operators.

8. The method according to claim 7, characterized in that, When the target tool includes the AI ​​model, a correspondence between the first business object and the reference knowledge is also established in the logic graph. The correspondence is used to indicate that the AI ​​model is invoked during the business reasoning process to process the first judgment node by referring to the reference knowledge.

9. The method according to claim 8, characterized in that, The reference knowledge includes multiple types of knowledge, and the correspondence is also used to indicate the reference priority of each type of knowledge.

10. The method according to any one of claims 1-9, characterized in that, The logic graph is a tree-connected graph, and the logic graph includes multiple layers of decision nodes connected in a tree structure. Each decision node in the last layer of the multiple decision nodes is connected to a conclusion node, and the decision nodes related to the conclusion node include all decision nodes on the path from the root node of the logic graph to the conclusion node.

11. The method according to any one of claims 1-9, characterized in that, The business data includes scenario description data. The logic graph is determined from multiple constructed logic graphs based on the scenario description data. Each of the multiple logic graphs has a corresponding scenario label, which is used to indicate the business scenario to which the logic graph is applicable.

12. A business reasoning device, characterized in that, include: The acquisition module is used to acquire a business reasoning request, wherein the business reasoning request is used to instruct the execution of business reasoning on business data; The processing module is used to determine the first conclusion node corresponding to the business data in the logic graph based on the business reasoning request. The logic graph includes multiple judgment nodes and multiple conclusion nodes. The multiple judgment nodes are used to indicate the conditions that the business object needs to meet, and the multiple conclusion nodes are used to indicate the business reasoning result. The business data meets the conditions indicated by the judgment node related to the first conclusion node. The processing module is also used to output the business reasoning result indicated by the first conclusion node; The business data includes the value of a first business object, the value of which satisfies the condition indicated by the first judgment node related to the first conclusion node, the value of which is obtained based on the data recorded under the first target field, and the first target field is determined in the database by an AI model based on the first business object and related reference knowledge, and the database is used to store the data in the business data.

13. A computing device, characterized in that, The device includes a memory and a processor; the memory stores code, and the processor is configured to execute the code, wherein when the code is executed, the computing device performs the method as described in any one of claims 1 to 11.

14. A computing device cluster, characterized in that, It includes at least one computing device, each computing device including a processor and memory; The processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device to cause the cluster of computing devices to perform the operational steps of the method as described in any one of claims 1 to 11.

15. A chip system, characterized in that, The chip system includes a processor and a communication interface, the communication interface being used to communicate with modules outside the chip system, and the processor being used to execute the method as described in any one of claims 1 to 11.

16. A computer storage medium, characterized in that, The computer storage medium stores instructions that, when executed by the computer, cause the computer to perform the method according to any one of claims 1 to 11.

17. A computer program product, characterized in that, The computer program product stores instructions that, when executed by a computer, cause the computer to perform the method described in any one of claims 1 to 11.

Images