Data processing method, apparatus, device, computer-readable medium, and program product

Abstract

Embodiments of the present disclosure disclose data processing methods, devices, equipment, computer readable media and program products. A specific embodiment of the method comprises: in response to currently performing data preprocessing on target data, preprocessing the target data using a first intelligent agent to obtain a real-time output target processing result, and based on the target processing result, performing the following generation steps: using a second intelligent agent, performing semantic reasoning on the target processing result to obtain reasoning content; in response to determining that the preprocessing on the target data is completed, generating display content corresponding to the reasoning content; in response to obtaining a next processing result, taking the next processing result as the target processing result and continuing to perform the generation steps. This embodiment is related to intelligent agents. In the process of performing data preprocessing by the first intelligent agent, parallel reasoning is performed by setting the second intelligent agent, which can greatly improve the processing efficiency of data.

Description

Technical Field

[0001] Embodiments of this disclosure relate to the field of intelligent agent technology, and more specifically to data processing methods, apparatus, devices, computer-readable media, and program products. Background Technology

[0002] Currently, with the continuous development of intelligent agents and their increasingly widespread application in various fields, utilizing multiple agents to process and analyze datasets within a target domain has become one of the main development directions. The typical approach for processing datasets and performing semantic analysis based on multiple agents is as follows: First, a first agent preprocesses the data in the dataset to obtain a preprocessed dataset. Then, after obtaining the preprocessed dataset, a second agent performs semantic analysis and inference on the preprocessed dataset to obtain the inference results.

[0003] However, when using the above method, the following technical problems often arise: The second agent remains idle until the first agent completes all data processing. This sequential execution method results in significant processing latency, often requiring users to wait for several minutes or even longer, severely compromising the timeliness of decision-making and leading to low data processing efficiency. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure provide data processing methods, apparatuses, devices, computer-readable media, and program products to address the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide a data processing method, including: in response to currently performing data preprocessing on target data, using a first intelligent agent to preprocess the target data to obtain a target processing result output in real time, and based on the target processing result, performing the following generation steps: using a second intelligent agent to perform semantic reasoning on the target processing result to obtain reasoned content; in response to determining that the preprocessing of the target data is completed, generating display content corresponding to the reasoned content; in response to obtaining a next processing result, using the next processing result as the target processing result, and continuing to perform the above generation steps, wherein the next processing result is the processing result output at the next time time obtained by the first intelligent agent continuing to preprocess the target data.

[0007] Optionally, the above-mentioned use of a second intelligent agent to perform semantic reasoning on the target processing result to obtain reasoning content includes: based on the historical reasoning content generated by the second intelligent agent, using the second intelligent agent to perform semantic reasoning on the target processing result to obtain reasoning content, wherein the historical reasoning content is the reasoning result of semantic reasoning based on the historical processing result, and the historical processing result is the processing result after preprocessing the current time corresponding to the previous time; and the above-mentioned response to obtaining the next processing result, using the next processing result as the target processing result, and continuing to execute the above-mentioned generation steps includes: in response to obtaining the next processing result, using the next processing result as the target processing result, using the reasoning content as historical reasoning content, and continuing to execute the above-mentioned generation steps.

[0008] Optionally, the above-mentioned process of performing semantic reasoning on the target processing result based on the historical reasoning content generated by the second agent to obtain reasoning content includes: in response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition, occluding the historical reasoning content; and using the second agent to perform semantic reasoning on the target processing result based on the occluded reasoning content to obtain reasoning content.

[0009] Optionally, the above-mentioned use of the first intelligent agent to preprocess the target data to obtain the target processing result in real time includes: using the first intelligent agent to process the target data into data that conforms to the cache structure form corresponding to the key-value pair cache pool, thereby obtaining the target processing result; and after the above-mentioned use of the first intelligent agent to preprocess the target data to obtain the target processing result in real time, the above method further includes: writing the target processing result into the key-value pair cache pool, so that the second intelligent agent can obtain the target processing result by using a logical pointer pointing to the content written by the first intelligent agent.

[0010] Optionally, before the above-mentioned response to the current execution of data preprocessing for the target data, using the first agent to preprocess the target data to obtain the target processing result in real time, and before performing the following generation steps based on the target processing result, the above method further includes: obtaining the inference requirements corresponding to the target data; writing the prompt information corresponding to the inference requirements into a key-value pair cache pool for the second agent to learn the inference requirements.

[0011] Optionally, the above-mentioned use of the first intelligent agent to process the target data into data conforming to the cache structure form corresponding to the key-value pair cache pool, and to obtain the target processing result, includes: using the first intelligent agent to process the target data into data conforming to the cache structure form, and to obtain an initial processing result; performing a predefined structure verification on the initial processing result to obtain a verification result; and determining the initial processing result as the target processing result in response to the verification result indicating that the initial processing result conforms to the predefined structure.

[0012] Optionally, the method further includes: in response to the above verification result indicating that the above initial processing result does not conform to the above predefined structure, instructing the first agent to re-preprocess the above initial processing result to obtain a reprocessing result; in response to the above reprocessing result corresponding to the verification result indicating that it conforms to the above predefined structure, determining the above reprocessing result as the target processing result.

[0013] Optionally, the above-mentioned use of the second intelligent agent to perform semantic reasoning on the target processing result based on the occluded reasoning content to obtain reasoning content includes: using the second intelligent agent to perform the following determination steps: performing semantic reasoning on the target processing result based on the occluded reasoning content to obtain current reasoning content; generating predictive reasoning content corresponding to the next prediction processing result based on the target processing result and the current reasoning content; and determining the current reasoning content and the predictive reasoning content as reasoning content.

[0014] Optionally, the above-mentioned response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition, and performing error content masking on the historical reasoning content, includes: in response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition, performing the following processing steps: extracting contextual reasoning content related to the target processing result from the historical reasoning content as error content; constructing error masking information for the error content; and masking the historical reasoning content according to the error masking information.

[0015] Optionally, before generating the display content corresponding to the inference content in response to determining that the preprocessing of the target data is completed, the method further includes: determining whether there is information related to the end of preprocessing in the target processing result; determining the inference confidence level corresponding to the inference content in response to determining that there is such information; and determining that the preprocessing of the target data is completed in response to the inference confidence level satisfying the confidence level condition.

[0016] Optionally, the first intelligent agent and the second intelligent agent each have corresponding short-term memory buffers; and the preprocessing of the target data using the first intelligent agent to obtain a real-time output target processing result includes: retrieving first associated data related to the target data from a long-term memory database; obtaining context data corresponding to the target data from the short-term memory buffer corresponding to the first intelligent agent; fusing the first associated data and the context data to obtain first fused data; and preprocessing the target data using the first intelligent agent based on the first fused data to obtain a real-time output target processing result.

[0017] Optionally, the above method further includes: storing the aforementioned context data in a short-term memory buffer corresponding to the second agent; and performing semantic reasoning on the target processing result using the second agent to obtain reasoning content, including: retrieving second associated data related to the target processing result from the long-term memory database; fusing the context data in the short-term memory buffer corresponding to the second agent with the second associated data to obtain second fused data; and performing semantic reasoning on the target processing result using the second agent based on the second fused data to obtain reasoning content.

[0018] Optionally, the above-mentioned data fusion of the first associated data and the context data to obtain the first fused data includes: determining whether there is a data conflict between the first associated data and the context data; in response to determining that there is a data conflict, determining the associated sub-data that conflicts in the first associated data and the context sub-data that conflicts in the context data; filtering out the sub-data with the earliest storage time from the associated sub-data and the context sub-data; and fusing the associated sub-data that does not conflict in the first associated data, the context sub-data that does not conflict in the context data, and the sub-data with the earliest storage time to obtain the first fused data.

[0019] Optionally, after the above response to determining that the preprocessing of the target data has been completed and generating the display content corresponding to the inference content, the above method further includes: generating reusable empirical data based on the inference content as long-term memory data; generating reusable empirical data based on the obtained target processing results to obtain each long-term memory data; and storing the obtained long-term memory dataset in the long-term memory database.

[0020] Secondly, some embodiments of this disclosure provide a data processing apparatus, including: a first execution unit configured to, in response to currently performing data preprocessing on target data, utilize a first intelligent agent to preprocess the target data to obtain a target processing result output in real time, and based on the target processing result, perform the following generation steps: utilize a second intelligent agent to perform semantic reasoning on the target processing result to obtain reasoned content; in response to determining that the preprocessing of the target data is completed, generate display content corresponding to the reasoned content; and a second execution unit configured to, in response to obtaining a next processing result, use the next processing result as the target processing result and continue to execute the above generation steps, wherein the next processing result is the next time-output processing result obtained by the first intelligent agent continuing to preprocess the target data.

[0021] Optionally, the first execution unit can be configured to: based on the historical reasoning content generated by the second agent, utilize the second agent to perform semantic reasoning on the target processing result to obtain reasoning content, wherein the historical reasoning content is the reasoning result of semantic reasoning based on the historical processing result, and the historical processing result is the processing result after preprocessing corresponding to the previous time. Furthermore, the first execution unit can be configured to: in response to obtaining the next processing result, use the next processing result as the target processing result, use the reasoning content as historical reasoning content, and continue executing the above generation steps.

[0022] Optionally, the first execution unit can be configured to: in response to detecting that the semantics of the data corresponding to the above target processing result satisfies the semantic inversion condition, perform erroneous content occlusion on the above historical reasoning content; and use the above second agent to perform semantic reasoning on the above target processing result based on the occluded reasoning content to obtain the reasoning content.

[0023] Optionally, the first execution unit can be configured to: use the first intelligent agent to process the target data into data conforming to the cache structure of the key-value pair cache pool, thereby obtaining the target processing result. The first execution unit can also be configured to: write the target processing result into the key-value pair cache pool, so that the second intelligent agent can obtain the target processing result using a logical pointer pointing to the content written by the first intelligent agent.

[0024] Optionally, the first execution unit can be configured to: obtain the inference requirements corresponding to the target data; and write the prompt information corresponding to the inference requirements into a key-value pair cache pool for the second agent to learn the inference requirements.

[0025] Optionally, the first execution unit may be configured to: use the first intelligent agent to process the target data into data conforming to the cache structure form to obtain an initial processing result; perform a predefined structure verification on the initial processing result to obtain a verification result; and, in response to the verification result indicating that the initial processing result conforms to the predefined structure, determine the initial processing result as the target processing result.

[0026] Optionally, the first execution unit may be configured to: in response to the above verification result indicating that the above initial processing result does not conform to the above predefined structure, instruct the first agent to re-preprocess the above initial processing result to obtain a reprocessing result; and in response to the above reprocessing result corresponding to the verification result indicating that it conforms to the above predefined structure, determine the above reprocessing result as the target processing result.

[0027] Optionally, the first execution unit may be configured to: utilize the second agent described above to perform the following determination steps: based on the occluded reasoning content, perform semantic reasoning on the target processing result to obtain the current reasoning content; based on the target processing result and the current reasoning content, generate the prediction reasoning content corresponding to the next prediction processing result; and determine the current reasoning content and the prediction reasoning content as the reasoning content.

[0028] Optionally, the first execution unit can be configured to: in response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition, perform the following processing steps: extract the contextual reasoning content related to the target processing result from the historical reasoning content as the error content; construct error mask information for the error content; and mask the historical reasoning content according to the error mask information.

[0029] Optionally, the first execution unit may be configured to: determine whether there is information related to the end of preprocessing in the above target processing result; in response to determining that there is, determine the inference confidence corresponding to the above inference content; and in response to the above inference confidence satisfying the confidence condition, determine that the preprocessing of the above target data is completed.

[0030] Optionally, the first intelligent agent and the second intelligent agent each have corresponding short-term memory buffers; and the first execution unit can be configured to: retrieve first associated data related to the target data from the long-term memory database; obtain context data corresponding to the target data from the short-term memory buffer corresponding to the first intelligent agent; perform data fusion on the first associated data and the context data to obtain first fused data; and preprocess the target data using the first intelligent agent based on the first fused data to obtain a real-time output target processing result.

[0031] Optionally, the first execution unit can be configured to: store the aforementioned context data in the short-term memory buffer corresponding to the second agent. The first execution unit can also be configured to: retrieve second associated data related to the target processing result from the long-term memory database; fuse the context data in the short-term memory buffer corresponding to the second agent with the second associated data to obtain second fused data; and, based on the second fused data, use the second agent to perform semantic reasoning on the target processing result to obtain reasoned content.

[0032] Optionally, the first execution unit may be configured to: determine whether there is a data conflict between the first associated data and the context data; in response to determining that there is a data conflict, determine the associated sub-data that conflicts in the first associated data and the context sub-data that conflicts in the context data; filter out the sub-data with the earliest storage time from the associated sub-data and the context sub-data; and perform data fusion on the associated sub-data that does not conflict in the first associated data, the context sub-data that does not conflict in the context data, and the sub-data with the earliest storage time to obtain the first fused data.

[0033] Optionally, the first execution unit can be configured to: generate reusable empirical data based on the above reasoning content as long-term memory data; generate reusable empirical data based on the obtained target processing results to obtain various long-term memory data; and store the obtained long-term memory dataset in the above-mentioned long-term memory database.

[0034] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0035] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0036] Fifthly, some embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0037] The above embodiments of this disclosure have the following beneficial effects: Through the data processing methods of some embodiments of this disclosure, by setting a second intelligent agent to perform parallel reasoning during the data preprocessing process of the first intelligent agent, the data processing efficiency can be greatly improved. Specifically, the reason for the low processing efficiency of related data is that the second intelligent agent is idle before the first intelligent agent completes all data processing work. This serial execution method has a large processing delay, often requiring the user to wait for several minutes or even longer, seriously damaging the timeliness of decision-making and resulting in low data processing efficiency. Based on this, the data processing method of some embodiments of this disclosure firstly, in response to the current execution of data preprocessing for target data, utilizes the first intelligent agent to preprocess the target data to obtain a real-time output target processing result, and based on the target processing result, performs the following generation steps: First, utilizing the second intelligent agent, semantic reasoning for the target processing result can be performed efficiently and accurately to obtain reasoned content. Here, during the real-time data preprocessing of the target data using the first intelligent agent, and after the first intelligent agent outputs the target processing result, the second intelligent agent can be instructed to perform semantic reasoning on the target processing result in real time. That is, during the preprocessing of the target data, the second model is simultaneously instructed to perform effective semantic reasoning on the real-time output processing results, achieving parallel execution between data preprocessing and semantic reasoning and avoiding significant processing delays. The second step, in response to the completion of preprocessing of the target data, accurately generates the display content corresponding to the reasoned content. Finally, in response to obtaining the next processing result, this next processing result is used as the target processing result, and the above generation steps continue. Here, the next processing result is the processing result output at the next time step obtained by the first agent after further preprocessing the target data. In this process, the first agent also outputs the processing result at the next time step during the data preprocessing of the target data. This next time step processing result can be used as the target processing result, simultaneously instructing the second model to perform effective semantic processing, thus achieving parallel execution of data preprocessing and semantic reasoning. In summary, by obtaining the processing results during the data preprocessing of the target data by the first intelligent agent, and simultaneously instructing the second model to perform parallel semantic reasoning, the efficiency of data processing can be greatly improved. This allows the first intelligent agent to perform data preprocessing while the second intelligent agent performs semantic reasoning based on the output of the first intelligent agent. Attached Figure Description

[0038] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0039] Figure 1 This is a schematic diagram illustrating an application scenario of a data processing method according to some embodiments of the present disclosure; Figure 2 These are flowcharts of some embodiments of the data processing method according to this disclosure; Figure 3 These are flowcharts of other embodiments of the data processing methods according to this disclosure; Figure 4 These are schematic diagrams illustrating the structure of some embodiments of the data processing apparatus according to this disclosure; Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0040] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0041] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0042] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0043] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0044] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0045] Before performing any of the operations involving the collection, storage, or use of user personal information (e.g., target data) disclosed in this disclosure, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, informing personal information subjects, and obtaining prior authorization and consent from personal information subjects.

[0046] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0047] Figure 1 This is a schematic diagram illustrating an application scenario of a data processing method according to some embodiments of the present disclosure.

[0048] exist Figure 1 In this application scenario, firstly, in response to the current data preprocessing of target data 102, electronic device 101 can utilize a first intelligent agent 103 to preprocess the target data 102, obtaining a real-time output target processing result 104. Based on the target processing result 104, the following generation steps are performed: First, using a second intelligent agent 105, semantic reasoning is performed on the target processing result 104 to obtain reasoned content 106. In this application scenario, target data 102 can be operational data. Target processing result 104 can be operational processing result. Reasoned content 106 can be operational reasoned content. Second, in response to determining that the preprocessing of target data 102 is complete, display content 107 corresponding to the reasoned content 106 is generated. In this application scenario, display content 107 can be operational display content. Finally, in response to obtaining the next processing result, electronic device 101 can use the next processing result as the target processing result and continue to execute the above generation steps. The next processing result is the processing result of the first intelligent agent 103 continuing to preprocess the target data 102 and outputting it at the next time step.

[0049] It should be noted that the aforementioned electronic device 101 can be either hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the electronic device is software, it can be installed in the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0050] It should be understood that Figure 1 The number of electronic devices shown is merely illustrative. Any number of electronic devices can be used depending on the implementation requirements.

[0051] Continue to refer to Figure 2The flowchart 200 illustrates some embodiments of a data processing method according to the present disclosure. The data processing method includes the following steps: Step 201: In response to the current data preprocessing of the target data, the first agent is used to preprocess the target data to obtain the real-time output target processing result, and based on the target processing result, the following generation steps are performed: Step 2021: Using the second intelligent agent, perform semantic reasoning on the above target processing results to obtain the reasoning content.

[0052] In some embodiments, the entity executing the above data processing method (e.g.) Figure 1The electronic device 101 shown can utilize a second intelligent agent to perform semantic reasoning on the aforementioned target processing results, obtaining the reasoned content. The target data can be data currently undergoing preprocessing or requiring further reasoning. In practice, the target data can be data from various domains. For example, the target data can be logistics data or operational data. In practice, the target data can be data currently being processed within a massive dataset. A massive dataset can be a dataset with a data volume exceeding the target value (a threshold representing whether the data is massive). That is, the target data is data within a massive dataset that the first intelligent agent is currently processing or processing. Here, the massive dataset can be a dataset uploaded by the target object (e.g., a user) that requires data processing and semantic reasoning. The current data preprocessing of the target data can be data preprocessing currently being performed or data preprocessing of the target data that was just performed. Data preprocessing can be a series of cleaning, transformation, and organization processes performed on the raw data. In practice, data preprocessing can include, but is not limited to, at least one of the following: data cleaning, data integration, and key-value pair extraction. The first intelligent agent can be an intelligent agent specifically designed to process the input data. Specifically, the first agent here acts as the "data cleaning object." This first agent, specifically responsible for data preprocessing, primarily ensures data quality and usability by building a reliable data foundation through automated cleaning, transformation, and evaluation. The real-time output of the target processing result can be the processing content generated by the first agent during the processing of the target data. In practice, the target processing result can be the complete processing result after data preprocessing of the target data, or it can be a partial processing result during the data preprocessing process. A partial processing result indicates that the first agent has not yet completely preprocessed the target data. That is, there may still be other preprocessed results. The target processing result can be the preprocessing result output by the first agent at the current time. For example, for target data including statement 1 and statement 2, the corresponding target processing result can be the overall preprocessing result after the first agent preprocesses statements 1 and 2, or it can be a partial processing result after the first agent preprocesses statement 1. That is, it indicates that the first agent has not yet processed statement 2. The second agent can be an agent specifically used for semantic reasoning and data analysis of the preprocessed data. Specifically, the second intelligent agent here plays the role of a "data analysis object." Based on the preprocessing results, this second intelligent agent is mainly used for business logic reasoning and decision support, transforming high-quality data into actionable business insights. It should be noted that the semantic reasoning content corresponding to the semantic reasoning of the target processing results can be based on the reasoning requirements.That is, a semantic reasoning task can be set up for the target processing result. After the second agent obtains the target processing result, it performs semantic reasoning based on the task content corresponding to the semantic reasoning task. The reasoning content can be the reasoning result after semantic reasoning based on the target processing result. In other words, the reasoning content is the task output result (i.e., the required result) corresponding to the semantic reasoning task.

[0053] It should be noted that the first and second agents can each be deployed with corresponding large language models to perform data preprocessing and semantic reasoning, respectively. The large language model corresponding to the first agent can be trained on a preprocessed training dataset. Preprocessed training data can include: unprocessed data, preprocessing requirements, and processed data. The large language model corresponding to the second agent can be trained on a semantic reasoning dataset. Semantic reasoning data can include: input data and semantic reasoning content. Both the first and second agents can use the same large model structure, and can be trained periodically.

[0054] In specific domains, the corresponding massive datasets can be datasets specific to those domains. For example, for the operations domain, the corresponding massive dataset is a massive e-commerce dataset. The corresponding target data can be user transaction data, and the corresponding data preprocessing can include: handling missing and outlier values, data standardization, and feature construction. For example, for the approximately 8% of key fields missing in user transaction data (e.g., purchase amount, region), nearest neighbor interpolation is used to supplement the region information, and the order amount is filled in by combining the average value of similar products, avoiding sample bias caused by direct deletion. Outlier orders (e.g., single purchase exceeding the average by 5 times) are identified through box plots, and business rules are used to determine whether they are genuine high-value orders (e.g., luxury goods purchases) or data errors, removing only obviously erroneous data rather than simply deleting it. The raw transaction data is transformed into three core indicators: "Recency of Recent Purchase," "Frequency of Purchase," and "Monetary Amount," and Z-Score standardization is used to eliminate differences in units of measurement, making data from different dimensions comparable. Based on behavioral data such as browsing time and search keywords, derived tags such as "price sensitivity" and "category preference" are constructed. For example, users who "frequently search for discount keywords but have a low order rate" can be categorized as a high-price-sensitive group. The corresponding semantic reasoning can be based on the pre-processed target data to segment users and identify their value.

[0055] For example, in the logistics field, the target data might be multi-sourced and heterogeneous pharmaceutical cold chain data. Corresponding data preprocessing could include: data cleaning, spatiotemporal alignment, temperature control threshold verification, and timeliness verification. Semantic reasoning could involve analyzing the root causes of temperature control anomalies, or real-time optimization and cost control.

[0056] As an example, firstly, a first preprocessing prompt word is generated to preprocess the target data. Then, the first preprocessing prompt word is input into a first agent to obtain the real-time output of the target processing result. Next, semantic reasoning prompt information based on the target processing result is obtained. Then, a first reasoning prompt word is generated to perform semantic reasoning on the target processing result based on the semantic reasoning prompt information. Finally, the first reasoning prompt word is input into a second agent to obtain the reasoned content.

[0057] In some optional implementations of certain embodiments, the aforementioned executing agent may utilize a second intelligent agent to perform semantic reasoning on the aforementioned target processing result to obtain reasoned content, including the following steps: Based on the historical reasoning content generated by the second agent, semantic reasoning is performed on the target processing result using the second agent to obtain the reasoning content. The historical reasoning content is the reasoning result based on the historical processing result, which is the processing result after preprocessing the previous time corresponding to the current time. The historical reasoning content can be the reasoning content generated under the condition of the previous time corresponding to the current time (i.e., the historical time preceding the current time). In practice, the current time can be the time when the target data is preprocessed to generate the target processing result. The previous time can be the time before the current time when the first agent output the processing result (i.e., the historical processing result). Here, the processing result output by the first agent at the previous time can be the processing result output before the target processing result after the first agent preprocessed the target data, or it can be the processing result after the first agent preprocessed the data of the previous processing. The data of the previous processing can be data processed before the target data.

[0058] In practice, historical reasoning content can be obtained by semantic reasoning using a second intelligent agent based on historical processing results. The generation of historical reasoning content can also reference reasoning content from previous historical times.

[0059] As an example, firstly, a second reasoning prompt is generated to perform semantic reasoning on the target processing result based on historical reasoning content. Then, the second reasoning prompt is input into the second agent to obtain the reasoning content.

[0060] In some optional implementations of certain embodiments, the execution entity can, based on the historical reasoning content generated by the second intelligent agent, utilize the second intelligent agent to perform semantic reasoning on the target processing result to obtain the reasoning content, including the following steps: The first step involves masking the historical inference content in response to the detection that the semantics of the target processing result satisfies the semantic inversion condition. The semantics of the target processing result can be the semantic content of the data presented by the target processing result. The semantic inversion condition can be a condition where a reasoning error exists in the historical inference content after the target processing result appears, causing a reversal of the semantic content of the reasoning. For example, the semantic inversion condition can be a condition that detects a semantic conflict between the semantics of the output processing result and the semantics of the historical inference content. In practice, semantic detection algorithms can be used to detect whether there is a semantic conflict between the semantics of the processing result and the semantics of the historical inference content. For example, the semantic detection algorithm can be a detection algorithm based on the Vector Space Model (VSM) and threshold determination, or a conflict detection algorithm based on semantic entropy. The masking of the error content can involve masking the inference sub-content in the historical inference content that conflicts with the semantics of the target processing result. In addition, the semantic inversion condition can be the presence of a semantic inversion signal in the data semantics. Semantic inversion signals can include explicit inversion and implicit inversion. For example, explicit inversion can be inversion words such as "but," "correction," and "not actually the case." Implicit reversals can be sudden changes in data polarity (for example, after a period of positive growth, there is a sudden sharp drop, or the data unit changes from "yuan" to "ten thousand yuan").

[0061] In practice, the executing agent can set up a parallel high-priority near-field to specifically monitor whether the semantics of the data corresponding to the above-mentioned target processing results satisfy the semantic inversion condition. In addition, if a semantic inversion is detected, the executing agent immediately sends a "logical interruption" signal to the second agent to stop the generation of inference prediction content.

[0062] As an example, firstly, the aforementioned execution entity can extract reasoning sub-content from historical reasoning content that conflicts with the semantics of the data corresponding to the target processing result, as erroneous content. Then, the aforementioned erroneous content is removed from the memory storing historical reasoning content, resulting in the removed reasoning content, which serves as the occluded reasoning content.

[0063] The second step involves using the aforementioned second agent to perform semantic reasoning on the result of the target processing based on the reasoning content after occlusion, thereby obtaining the reasoning content.

[0064] As an example, firstly, the aforementioned executing agent can generate a third inference prompt word to perform semantic inference on the target processing result based on the inference content after occlusion. Then, the third inference prompt word is input into the second agent to obtain the inference content.

[0065] Here, when a semantic reversal or intent drift is detected in the processing result of the upstream output, the executing agent needs to be able to quickly correct the reasoning result generated by the downstream agent (i.e., the second agent) based on incorrect assumptions, without having to restart the entire reasoning process, which greatly improves reasoning efficiency.

[0066] In some optional implementations of certain embodiments, the aforementioned execution entity may utilize a first intelligent agent to preprocess the aforementioned target data to obtain a real-time output target processing result, including the following steps: Using the first intelligent agent, the target data is processed into data conforming to the cache structure of the key-value pair cache pool, thus obtaining the target processing result. The key-value pair cache pool can be a cache pool that stores preprocessed data in key-value pair format. The cache structure can be the data structure format corresponding to the data stored in the key-value pair cache pool; that is, the cache structure can be in key-value pair format. For the key-value pair cache pool, the first intelligent agent can act as the writer, and the second intelligent agent can act as the reader.

[0067] As an example, firstly, the aforementioned executing agent can generate a second preprocessing prompt to preprocess the target data to produce a processing result in the form of key-value pairs. Then, the second preprocessing prompt is input into the first agent to obtain the target processing result.

[0068] In some optional implementations of certain embodiments, after the first intelligent agent is used to preprocess the target data to obtain the real-time output target processing result, the steps further include: The aforementioned executing entity can write the target processing result into the aforementioned key-value pair cache pool, so that the second agent can obtain the target processing result using a logical pointer pointing to the corresponding content written by the first agent. Specifically, the logical pointer pointing to the corresponding content written by the first agent can be a logical pointer specifically used to retrieve the processing result written by the first agent into the key-value pair cache pool. That is, by setting the logical pointer and the key-value pair cache pool, the second agent can obtain the processing result output by the first agent in real time, reducing transmission latency. In practice, the second agent can use the logical pointer to read the most recently written processing result in the corresponding video memory of the key-value pair cache pool to obtain the target processing result.

[0069] By setting logical pointers, a zero-copy mechanism can be implemented, eliminating the need for any data transfer time.

[0070] In addition, the second agent's attention mechanism monitors the latest output of the first agent in real time.

[0071] In some optional implementations of certain embodiments, the execution entity can utilize the first intelligent agent to process the target data into data conforming to the cache structure of the key-value pair cache pool, thereby obtaining the target processing result, including the following steps: The first step involves using the aforementioned first intelligent agent to process the target data into data conforming to the above cache structure, obtaining an initial processing result. This initial processing result can be a key-value pair format, requiring further structural verification. Specific implementation details are omitted here. It can be generated based on prompt word technology.

[0072] Here, by processing the target data into a cache-structured format, we can avoid outputting unstructured natural language nonsense by breaking the target data down into atomic units. For example, when processing data corresponding to "sales revenue," it must be generated sequentially.<Key:Sales> -><Value:1000> -><Unit:USD> This can be achieved by generating data in a cached structure, which facilitates semantic reasoning by a second agent.

[0073] The second step involves performing a predefined structure validation on the initial processing results to obtain the validation results. This predefined structure validation can involve verifying whether the initial processing results are key-value pairs and whether they contain redundant semantic content. The validation result can be one of the following: a result indicating that the initial processing results conform to the predefined structure, or a result indicating that the initial processing results do not conform to the predefined structure (e.g., a numeric field outputs text).

[0074] In practice, the initial processing results can be validated using a predefined structure based on the inserted Just-in-Time Validator.

[0075] The third step is to determine the initial processing result as the target processing result in response to the above verification result indicating that the initial processing result conforms to the above predefined structure.

[0076] Here, by setting a predefined structure check, the quality of the data input to the second agent can be guaranteed, avoiding the problem of error cascading and amplification.

[0077] In some optional implementations of certain embodiments, the steps further include: The first step involves informing the first agent that the initial processing result does not conform to the predefined structure, based on the aforementioned verification result. This reprocessing aims to adjust the structure and / or semantic content of the initial processing result to meet the requirements of the predefined structure. The reprocessing result can be the result of further processing by the first agent.

[0078] As an example, firstly, the aforementioned executing entity can generate a processing prompt word instructing the first intelligent agent to preprocess the target data again based on the initial processing result. Then, the processing prompt word is input into the first intelligent agent to obtain the result of the second processing.

[0079] The second step is to determine the target processing result as the result of the reprocessing if the verification result characterization of the above reprocessing result conforms to the above predefined structure.

[0080] It should be noted that if the verification result still does not conform to the predefined structure after processing, the first agent needs to be used again for preprocessing until the processing result conforms to the predefined structure.

[0081] In some optional implementations of certain embodiments, the execution entity may utilize the second intelligent agent to perform semantic reasoning on the target processing result based on the occluded reasoning content, thereby obtaining the reasoning content, including the following steps: Using the second intelligent agent described above, the following determination steps are performed: Sub-step 1: Based on the occluded inference content described above, perform semantic inference on the target processing result to obtain the current inference content. The current inference content can be inference content directly derived from the currently obtained content. That is, the current inference content does not include content used for future time prediction based on the currently obtained content.

[0082] As an example, firstly, a fourth reasoning prompt is generated, instructing the second agent to directly perform semantic reasoning based on the occluded reasoning content and the target processing result, without generating a prediction of the reasoning content in the future. Then, the fourth reasoning prompt is input into the second agent to obtain the current reasoning content.

[0083] Sub-step 2: Based on the aforementioned target processing result and the current inference content, generate the prediction inference content corresponding to the next prediction processing result. Here, the next prediction processing result is the processing result output by the first agent at the next time point. The next time point can be the next result output time after the current time. The prediction inference content can be the prediction of the next prediction processing result and the inference result of semantic reasoning based on the next prediction processing result.

[0084] It should be noted that the next prediction process can include multiple prediction results. That is, multiple prediction results represent multiple prediction contents, and represent multiple different prediction contents. Each prediction result has a corresponding confidence level. By generating multiple prediction results, the efficiency of inference can be improved in subsequent future reasoning processes.

[0085] Here, by generating predictive reasoning content, forward-looking logic construction can be achieved, enabling the second agent to perform speculative calculations while the first agent is processing data.

[0086] For example, while the first agent is still generating the ending symbol of the "revenue" field, the second agent has already predicted in advance, based on the context probability, that the next field may be "profit" or "cost", and has begun to pre-build the relevant analysis logic path in the video memory.

[0087] As an example, firstly, a prediction prompt word is generated, instructing the second agent to predict the processing result at the next time step based on the target processing result and the current inference content. Then, the prediction prompt word is input into the second agent to obtain the next predicted processing result. Next, a fifth inference prompt word is generated, which, based on the next predicted processing result and the current inference content, generates the inference prediction content for the next time step. Finally, the fifth inference prompt word is input into the second agent to obtain the predicted inference content corresponding to the next predicted processing result.

[0088] Sub-step 3: Determine the current reasoning content and the predicted reasoning content as the reasoning content.

[0089] As an example, the aforementioned executing entity can combine the current inference content and the predicted inference content to obtain combined inference content, which can then be used as the inference content.

[0090] In some optional implementations of certain embodiments, in response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition, the executing entity may perform erroneous content masking on the historical reasoning content, including the following steps: In response to the detection that the semantics of the data corresponding to the above target processing result satisfy the semantic inversion condition, the following processing steps are performed: Sub-step 1 involves extracting contextual reasoning content related to the target processing result from the aforementioned historical reasoning content, which is then used as error content. This contextual reasoning content can be reasoning content from the historical reasoning content that has a semantic relationship with the target processing result.

[0091] As an example, the aforementioned executing entity can utilize the semantic extraction function corresponding to the second intelligent agent to extract contextual reasoning content related to the aforementioned target processing result from the aforementioned historical reasoning content, and use it as error content.

[0092] Sub-step 2 involves constructing error mask information for the aforementioned erroneous content. This error mask information can be information used to mask the erroneous content. In practice, the error mask information can be a mask matrix. For example, it can be a correction mask matrix. That is, the correction mask matrix is ​​used to mask the tokens corresponding to the erroneous content. The matrix dimension of the correction mask matrix corresponds to the content dimension of the historical inference content. For the token range corresponding to the erroneous content in the historical inference content, elements in the matrix region corresponding to the token range in the correction mask matrix can be set to -∞ (representing masking the erroneous content), and elements in the matrix region outside the token range can be set to 0 (representing normal visibility). In other words, since the historical inference content is generated based on standard Scaled Dot-ProductAttention, by setting elements in the matrix region corresponding to the token range to -∞, the weights output by the Softmax function approach 0, thus mathematically "removing" the influence of the erroneous context on subsequent generation, achieving a logically lossless rollback.

[0093] In practice, the executing agent does not physically delete the erroneous content in the video memory. Instead, it modifies the attention mask matrix in the second agent and uses the error mask information to set the attention weights corresponding to the erroneous content to -∞ in order to mask it.

[0094] Sub-step 3: Based on the above error masking information, mask the error content of the above historical reasoning content.

[0095] As an example, the aforementioned execution entity can fuse the error mask information corresponding to the mask matrix with the error content corresponding to the historical reasoning content to achieve error content masking.

[0096] In some optional implementations of certain embodiments, the first agent and the second agent each have corresponding short-term memory buffers (STM buffers). The first agent maintains its own STM buffer, and the second agent also maintains its own STM buffer. These STM buffers can be buffers that store short-term input and output data of the agent. For example, for the first agent, the corresponding STM buffer can store the first agent's input and output information in the short term, so as to grasp the contextual semantic content and achieve coherence in data preprocessing. For the second agent, the corresponding STM buffer can store the second agent's input information (the preprocessed result) and output information (i.e., the reasoning content) in the short term, so as to grasp the contextual semantic content and achieve coherence in semantic reasoning.

[0097] It's important to note that the short-term memory buffer employs an LRU (Least Recently Used) eviction policy. For example, the short-term memory buffer can use a sliding window and LRU eviction strategy to maintain the session context of the 10,000 most recently accessed tokens. When the buffer is full, the least recently accessed context fragment is automatically evicted, retaining the most recently active dialogue content. The data in the short-term memory buffer is fused with the current input through positional encoding, enabling the agent to simultaneously focus on the current input data and historical context, achieving coherence in multi-turn dialogues.

[0098] Optionally, the aforementioned executing entity may utilize the first intelligent agent to preprocess the aforementioned target data to obtain the target processing result output in real time, including the following steps: The first step is to retrieve the first associated data related to the target data from the Long-Term Memory Knowledge Base (LTM) construction. The LTM database can be a hierarchical LTM knowledge base used to store persistent knowledge and experience across sessions. In practice, the LTM database can be stored using a vector database, where each memory entry is encoded as a high-dimensional vector (e.g., typically 768 or 1024 dimensions), supporting semantic similarity retrieval. The layers of the LTM database can include: a semantic layer, an instance layer, and a meta-cognitive layer. The semantic layer can store abstract business rules, analysis templates, decision-making patterns, etc. For example, the semantic layer might store data such as "trigger an alert when outliers exceed 3σ". The instance layer can store specific historical cases, successes, and lessons learned. For example, the instance layer might store data such as "in Q1 2024, a certain product's conversion rate dropped by 30% due to a pricing strategy error". The meta-cognitive layer can store the performance metrics, optimization strategies, and learning trajectories corresponding to each agent. The data stored in the metacognitive layer could be something like, "The second agent improves accuracy by 15% when processing time-series data." The first associated data related to the target data can be data that has a semantic relationship with the semantic content of the target data. This first associated data can include the K memory entries that have the closest semantic relationship with the target data. For example, K could be 5.

[0099] As an example, the aforementioned entity can utilize a hybrid retrieval strategy, combining keyword matching (BM25) and cosine similarity for memory retrieval, to retrieve first associated data related to the aforementioned target data.

[0100] The second step is to retrieve the context data corresponding to the target data from the short-term memory buffer corresponding to the first agent. The context data can be data that has a semantic relationship with the target data.

[0101] The third step is to fuse the aforementioned first associated data and the aforementioned contextual data to obtain the first fused data.

[0102] The fourth step involves using the first fused data and the first intelligent agent to preprocess the target data to obtain the real-time output target processing result.

[0103] In practice, during the preprocessing of target data based on the first fused data, a higher attention weight (e.g., a weight coefficient of 1.2) can be assigned to the first associated data in the first fused data in the attention mechanism of the Transformer corresponding to the first agent, so that the agent pays more attention to historical experience.

[0104] As an example, firstly, a third preprocessing prompt is generated, instructing the first agent to preprocess the target data based on the first fused data. Then, the third preprocessing prompt is input into the first agent to obtain the target processing result in real time.

[0105] In some optional implementations of certain embodiments, the steps further include: The aforementioned context data is stored in the short-term memory buffer corresponding to the second agent. In practice, after the first agent outputs the target processing result, key information in the context data used (e.g., data feature summaries, anomaly pattern identifiers, etc.) can be compressed and encoded to be transmitted to the short-term memory buffer corresponding to the second agent, so that the second agent can use it for semantic reasoning in the future.

[0106] Optionally, the aforementioned executing entity may utilize a second intelligent agent to perform semantic reasoning on the aforementioned target processing result to obtain the reasoned content, including the following steps: The first step is to retrieve second-related data from the aforementioned long-term memory database that is relevant to the results of the target processing. Details will not be elaborated further.

[0107] The second step involves fusing the context data in the short-term memory buffer corresponding to the second agent with the aforementioned second associated data to obtain the second fused data. Details will not be elaborated further.

[0108] The third step involves using the second fused data and the second intelligent agent to perform semantic reasoning on the results of the target processing, thereby obtaining the reasoned content. Details will not be elaborated further.

[0109] In some optional implementations of certain embodiments, the execution entity may perform data fusion on the first associated data and the context data to obtain first fused data, including the following steps: The first step is to determine whether there is a data conflict between the aforementioned first associated data and the aforementioned context data. A data conflict can be caused by a discrepancy between the content of the first associated data and the content of the context data.

[0110] As an example, the aforementioned executing entity can, based on the content conflict functionality of the first and second agents, utilize relevant content conflict algorithms to determine whether a data conflict exists between the aforementioned first associated data and the aforementioned context data. For instance, the content conflict algorithm could be a vector distance method with dynamic threshold calibration.

[0111] The second step, in response to the determination of a data conflict, involves identifying the conflicting associated sub-data within the aforementioned first associated data and the conflicting context sub-data within the aforementioned context data. Specifically, the semantics of the content corresponding to the associated sub-data conflict with the semantics of the content corresponding to the context sub-data.

[0112] In practice, conflicting related sub-data in the first related data and conflicting contextual sub-data in the context data can be extracted based on the content extraction algorithm corresponding to the intelligent agent.

[0113] The third step is to filter the earliest stored sub-data from the aforementioned associated sub-data and context sub-data. The associated sub-data has a corresponding storage time in the long-term memory database, while the context sub-data has a corresponding storage time in the short-term memory buffer.

[0114] Here, each long-term memory entry in the long-term memory database is accompanied by a version number and a timestamp (i.e., the storage time). When a conflict is detected between old and new memory data (e.g., the old memory data represents "Q4 sales decreased", while the new memory data represents "Q4 sales increased"), the memory with the earliest timestamp can be prioritized.

[0115] In addition, long-term memory entries are accompanied by a confidence score, which decays over time. For memory entries unverified for more than 6 months, their confidence score is automatically reduced by 20%, lowering their priority in subsequent retrievals. While running background tasks, the long-term memory database is periodically scanned to identify semantically conflicting pairs of memory entries. When a conflict is detected, a manual review or an automated adjudication mechanism based on statistical validation is triggered. Memory entries with confidence scores below a threshold (e.g., 0.3) or that have not been used for a long time can be automatically archived to cold storage, freeing up vector database space while retaining metadata for future recovery.

[0116] In summary, this approach can effectively address potential knowledge conflicts and outdated information issues in long-term memory data, ensuring the consistency and timeliness of long-term memory data.

[0117] Therefore, the following effects can be achieved by setting up a long-term memory database: Effect 1: The intelligent agent can remember the user's historical preferences and dialogue context to provide personalized and continuous service.

[0118] Effect 2: We can learn from past successes, avoid repeating mistakes, and improve the quality of decision-making.

[0119] Effect 3: The agent can automatically adjust its analysis strategy and reasoning path based on performance feedback from the long-term memory database.

[0120] Effect 4: As the system runs, the long-term memory database is continuously enriched, and the overall intelligence level of the system continues to improve.

[0121] The fourth step is to fuse the non-conflicting related sub-data in the first related data, the non-conflicting context sub-data in the above context data, and the earliest time sub-data to obtain the first fused data.

[0122] In some optional implementations of certain embodiments, after generating the display content corresponding to the inference content in response to determining that the preprocessing of the target data is complete, the method further includes: The first step is to generate reusable empirical data based on the above reasoning, which will serve as long-term memory data.

[0123] In practice, after the second agent completes the current semantic reasoning task, the performance can be evaluated. If the performance is significant (e.g., an accuracy improvement of over 10%), the reasoning experience corresponding to the reasoning content can be abstracted to obtain reusable experience data, which can then be stored in a long-term memory database for continuous learning. Long-term memory data can be memory entries subsequently stored in the long-term memory database.

[0124] The second step involves generating reusable empirical data based on the results of each target processing step, thus obtaining various long-term memory data. Details will not be elaborated further.

[0125] The third step is to store the obtained long-term memory dataset into the aforementioned long-term memory database.

[0126] Step 2022: In response to confirming that the preprocessing of the above target data is complete, generate the display content corresponding to the above reasoning content.

[0127] In some embodiments, in response to determining that preprocessing of the target data is complete, the executing agent can generate display content corresponding to the inference content. Here, completing the preprocessing of the target data indicates that the target processing result during the preprocessing of the target data by the first agent is the final output result. That is, completing the preprocessing of the target data indicates that the data preprocessing process for the target data has ended. The display content can be used to report and present the inference content. In practice, the display content is content displayed to the target object. The display format corresponding to the display content can be customized based on display requirements. For example, the display content can be in report form or in image form.

[0128] In practice, the aforementioned executing entity can utilize a third intelligent agent to generate the display content corresponding to the aforementioned reasoning content. This third intelligent agent can act as the reporting recipient. The third intelligent agent can process the input data based on the display requirements of the target object pair to obtain the display content according to those requirements. In practice, the third intelligent agent can be an agent deployed with a multimodal large model. In the context of the river, the third intelligent agent does not need to relearn the original data corresponding to the first intelligent agent. Instead, it directly receives the analysis conclusion stream generated by the second intelligent agent as input. Based on the reasoning content that the second intelligent agent has already structured and analyzed, the third intelligent agent can quickly generate the final decision recommendation (e.g., "It is recommended to immediately stop the deployment").

[0129] As another example, the aforementioned executing entity can substitute the reasoned content into the template corresponding to the display requirements to obtain the display content.

[0130] In some optional implementations of certain embodiments, prior to step 2022, the steps further include: The first step is to determine whether the target processing result contains information related to the end of preprocessing. This information can be semantic content related to the end of preprocessing and / or a terminator. For example, a terminator can be one of the following: an explicit paragraph terminator, or an object closure (e.g., "}" in JSON).

[0131] Specifically, after confirming its existence, the second agent can lock the corresponding generated reasoning content.

[0132] The second step, in response to the confirmed existence, is to determine the inference confidence level corresponding to the aforementioned inference content. This inference confidence level can be a confidence score corresponding to the inference content. The confidence score can be used to characterize the accuracy of the content corresponding to the inference. The higher the confidence score, the more accurate the inference. Inference confidence can also be used to determine whether implicit thinking is converted into explicit output. In practice, inference confidence can be a value between 0 and 1. For inference content that is a sequence of tokens, the length-normalized log probability corresponding to the token sequence can be used as the confidence score. The length-normalized log probability includes a length penalty factor (e.g., a value between 0.6 and 0.7).

[0133] The third step involves determining whether the preprocessing of the target data is complete, in response to the aforementioned inference confidence meeting the confidence condition. The confidence condition can be that the inference confidence is higher than a target confidence threshold. The target confidence threshold can be a threshold used to measure whether to convert implicit thinking into explicit output. In practice, the target confidence threshold can be a threshold set by 0. For example, the target confidence threshold can be 0.8.

[0134] In practice, after confirming that the reasoning confidence meets the confidence condition, the second agent can output the implicit reasoning content. Secondly, the temporary reasoning resources occupied by the implicit reasoning content can be released to free up resource space for the next data block.

[0135] From the user's perspective, we can see that the first intelligent agent is processing data, while the corresponding second intelligent agent is simultaneously displaying the analyzed and summarized content, achieving parallel output.

[0136] Step 203: In response to obtaining the next processing result, the next processing result is used as the target processing result, and the above generation steps are continued.

[0137] In some embodiments, in response to obtaining the next processing result, the executing entity may use the next processing result as the target processing result and continue to execute the above generation steps. The next processing result may be the output result of the first agent preprocessing the target data after the target processing result. That is, the next processing result is the processing result obtained by the first agent continuing to preprocess the target data at the next time. The processing time corresponding to the next processing result is the storage time corresponding to the target processing result. For example, the target processing result may be a token generated by preprocessing the target data at the current time, and the next processing result is the next token generated by preprocessing the target data at the next time.

[0138] Here, the second agent immediately performs a forward pass for each token generated by the first agent, thus enabling semantic reasoning.

[0139] In some optional implementations of certain embodiments, the above-mentioned response to obtaining the next processing result, using the next processing result as the target processing result, and continuing to execute the above-mentioned generation steps may include the following steps: In response to obtaining the next processing result, the next processing result is used as the target processing result, and the reasoning content is used as the historical reasoning content, and the above generation steps are continued. Details will not be elaborated further.

[0140] The above embodiments of this disclosure have the following beneficial effects: Through the data processing methods of some embodiments of this disclosure, by setting a second intelligent agent to perform parallel reasoning during the data preprocessing process of the first intelligent agent, the data processing efficiency can be greatly improved. Specifically, the reason for the low processing efficiency of related data is that the second intelligent agent is idle before the first intelligent agent completes all data processing work. This serial execution method has a large processing delay, often requiring the user to wait for several minutes or even longer, seriously damaging the timeliness of decision-making and resulting in low data processing efficiency. Based on this, the data processing method of some embodiments of this disclosure firstly, in response to the current execution of data preprocessing for target data, utilizes the first intelligent agent to preprocess the target data to obtain a real-time output target processing result, and based on the target processing result, performs the following generation steps: First, utilizing the second intelligent agent, semantic reasoning for the target processing result can be performed efficiently and accurately to obtain reasoned content. Here, during the real-time data preprocessing of the target data using the first intelligent agent, and after the first intelligent agent outputs the target processing result, the second intelligent agent can be instructed to perform semantic reasoning on the target processing result in real time. That is, during the preprocessing of the target data, the second model is simultaneously instructed to perform effective semantic reasoning on the real-time output processing results, achieving parallel execution between data preprocessing and semantic reasoning and avoiding significant processing delays. The second step, in response to the completion of preprocessing of the target data, accurately generates the display content corresponding to the reasoned content. Finally, in response to obtaining the next processing result, this next processing result is used as the target processing result, and the above generation steps continue. Here, the next processing result is the processing result output at the next time step obtained by the first agent after further preprocessing the target data. In this process, the first agent also outputs the processing result at the next time step during the data preprocessing of the target data. This next time step processing result can be used as the target processing result, simultaneously instructing the second model to perform effective semantic processing, thus achieving parallel execution of data preprocessing and semantic reasoning. In summary, by obtaining the processing results during the data preprocessing of the target data by the first intelligent agent, and simultaneously instructing the second model to perform parallel semantic reasoning, the efficiency of data processing can be greatly improved. This allows the first intelligent agent to perform data preprocessing while the second intelligent agent performs semantic reasoning based on the output of the first intelligent agent.

[0141] Further reference Figure 3 The diagram illustrates a flow 300 of another embodiment of the data processing method according to this disclosure. This data processing method includes the following steps: Step 301: Obtain the reasoning requirements corresponding to the target data mentioned above.

[0142] In some embodiments, the executing entity (e.g. Figure 1 The electronic device 101 shown can acquire the reasoning requirements corresponding to the aforementioned target data. These reasoning requirements can be the reasoning content regarding how to perform semantic reasoning on the target data. In practice, reasoning requirements can be pre-set based on the target data, representing what kind of semantic reasoning content the second intelligent agent needs to perform subsequently. For example, a reasoning requirement could be "We need to pay attention to sales figures, outliers, and year-on-year / month-on-month comparisons later."

[0143] Here, by generating reasoning requirements, the second agent can quickly learn about the corresponding task content of subsequent semantic reasoning tasks in advance.

[0144] Step 302: Write the prompt information corresponding to the above reasoning requirements into the key-value pair cache pool so that the second agent can learn the above reasoning requirements.

[0145] In some embodiments, the executing entity may write the prompt information corresponding to the inference requirement into a key-value pair cache pool for the second agent to learn the inference requirement. The prompt information corresponding to the inference requirement may be content prompting the second agent to learn the inference requirement in advance. In practice, the prompt information may be in text form or in encoded vector form.

[0146] Step 303: In response to the current data preprocessing of the target data, the first agent is used to preprocess the target data to obtain the real-time output target processing result, and based on the target processing result, the following generation steps are performed: Step 3031: Using the second intelligent agent, perform semantic reasoning on the above target processing results to obtain the reasoning content.

[0147] Step 3032: In response to confirming that the preprocessing of the target data has been completed, generate the display content corresponding to the reasoning content.

[0148] Step 304: In response to obtaining the next processing result, the next processing result is used as the target processing result, and the above generation steps are continued.

[0149] In some embodiments, the specific implementation of steps 303-304 and the resulting technical effects can be found in [reference needed]. Figure 2 Steps 201-202 in the corresponding embodiments will not be repeated here.

[0150] from Figure 3 It can be seen from this that, with Figure 2 Compared to the description of some corresponding embodiments, Figure 3In some corresponding embodiments, the data processing method process 300 can inform the second agent of the corresponding semantic reasoning task in advance, realize the inference engine warm-up in time, and quickly perform semantic reasoning processing after obtaining the processing result output by the first agent.

[0151] Further reference Figure 4 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a data processing apparatus, which are similar to... Figure 2 Corresponding to the method embodiments shown, this data processing device can be specifically applied to various electronic devices.

[0152] like Figure 4 As shown, a data processing device 400 includes a first execution unit 401 and a second execution unit 402. The first execution unit 401 is configured to, in response to currently performing data preprocessing on target data, utilize a first intelligent agent to preprocess the target data to obtain a real-time output target processing result, and based on the target processing result, perform the following generation steps: utilize a second intelligent agent to perform semantic reasoning on the target processing result to obtain reasoned content; in response to determining that the preprocessing of the target data is complete, generate display content corresponding to the reasoned content; the second execution unit 402 is configured to, in response to obtaining a next processing result, use the next processing result as the target processing result and continue executing the above generation steps, wherein the next processing result is the next time-time output processing result obtained by the first intelligent agent continuing to preprocess the target data.

[0153] In some optional implementations of certain embodiments, the first execution unit 401 may be further configured to: based on the historical reasoning content generated by the second agent, utilize the second agent to perform semantic reasoning on the target processing result to obtain reasoning content, wherein the historical reasoning content is the reasoning result of semantic reasoning based on the historical processing result, and the historical processing result is the processing result after preprocessing the current time corresponding to the previous time. Furthermore, the first execution unit 401 may be further configured to: in response to obtaining the next processing result, use the next processing result as the target processing result, use the reasoning content as historical reasoning content, and continue executing the above generation steps.

[0154] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: in response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition, to perform erroneous content occlusion on the historical reasoning content; and to use the second agent to perform semantic reasoning on the target processing result based on the occluded reasoning content to obtain the reasoning content.

[0155] In some optional implementations of certain embodiments, the first execution unit 401 may be further configured to: use the first intelligent agent to process the target data into data conforming to the cache structure form corresponding to the key-value pair cache pool, thereby obtaining the target processing result. Furthermore, the first execution unit 401 may be further configured to: write the target processing result into the key-value pair cache pool, so that the second intelligent agent can obtain the target processing result using a logical pointer pointing to the content written by the first intelligent agent.

[0156] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: obtain the inference requirements corresponding to the target data; and write the prompt information corresponding to the inference requirements into a key-value pair cache pool for the second agent to learn the inference requirements.

[0157] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: use the first intelligent agent to process the target data into data conforming to the cache structure form to obtain an initial processing result; perform a predefined structure verification on the initial processing result to obtain a verification result; and, in response to the verification result indicating that the initial processing result conforms to the predefined structure, determine the initial processing result as the target processing result.

[0158] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: in response to the verification result indicating that the initial processing result does not conform to the predefined structure, instruct the first agent to re-preprocess the initial processing result to obtain a reprocessing result; and in response to the verification result corresponding to the reprocessing result indicating that it conforms to the predefined structure, determine the reprocessing result as the target processing result.

[0159] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: utilize the second intelligent agent to perform the following determination steps: based on the occluded inference content, perform semantic inference on the target processing result to obtain the current inference content; based on the target processing result and the current inference content, generate the prediction inference content corresponding to the next prediction processing result; and determine the current inference content and the prediction inference content as the inference content.

[0160] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: in response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition, perform the following processing steps: extract the contextual reasoning content related to the target processing result from the historical reasoning content as error content; construct error mask information for the error content; and mask the historical reasoning content according to the error mask information.

[0161] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: determine whether there is information related to the end of preprocessing in the target processing result; in response to determining that there is, determine the inference confidence corresponding to the inference content; and in response to the inference confidence satisfying the confidence condition, determine that the preprocessing of the target data is completed.

[0162] In some optional implementations of certain embodiments, the first intelligent agent and the second intelligent agent each have corresponding short-term memory buffers; and the first execution unit 401 can be further configured to: retrieve first associated data related to the target data from a long-term memory database; obtain context data corresponding to the target data from the short-term memory buffer corresponding to the first intelligent agent; perform data fusion on the first associated data and the context data to obtain first fused data; and, based on the first fused data, use the first intelligent agent to preprocess the target data to obtain a real-time output target processing result.

[0163] In some optional implementations of certain embodiments, the first execution unit 401 may be further configured to: store the context data in the short-term memory buffer corresponding to the second agent. The first execution unit 401 may also be further configured to: retrieve second associated data related to the target processing result from the long-term memory database; fuse the context data in the short-term memory buffer corresponding to the second agent with the second associated data to obtain second fused data; and, based on the second fused data, use the second agent to perform semantic reasoning on the target processing result to obtain reasoned content.

[0164] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: determine whether there is a data conflict between the first associated data and the context data; in response to determining that there is a data conflict, determine the associated sub-data that conflicts in the first associated data and the context sub-data that conflicts in the context data; filter out the sub-data with the earliest storage time from the associated sub-data and the context sub-data; and perform data fusion on the associated sub-data that does not conflict in the first associated data, the context sub-data that does not conflict in the context data, and the sub-data with the earliest storage time to obtain the first fused data.

[0165] In some optional implementations of some embodiments, the first execution unit 401 may be further configured to: generate reusable empirical data as long-term memory data based on the above reasoning content; generate reusable empirical data based on the obtained target processing results to obtain each long-term memory data; and store the obtained long-term memory dataset in the long-term memory database.

[0166] It is understandable that the units described in the data processing device 400 are related to the reference. Figure 2 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the data processing apparatus 400 and the units contained therein, and will not be repeated here.

[0167] The following is for reference. Figure 5 It illustrates electronic devices suitable for implementing some embodiments of this disclosure (e.g., Figure 1 A schematic diagram of the structure of electronic device 101)500. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0168] like Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory 502 or a program loaded from a storage device 508 into a random access memory 503. The random access memory 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, the read-only memory 502, and the random access memory 503 are interconnected via a bus 504. An input / output interface 505 is also connected to the bus 504.

[0169] Typically, the following devices can be connected to the input / output interface 505: input devices 506 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 508 including, for example, magnetic tape, hard disk, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0170] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a read-only memory 502. When the computer program is executed by the processing device 501, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0171] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0172] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0173] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: in response to currently performing data preprocessing on target data, utilize a first intelligent agent to preprocess the target data to obtain a real-time output target processing result, and based on the target processing result, perform the following generation steps: utilize a second intelligent agent to perform semantic reasoning on the target processing result to obtain reasoned content; in response to determining that the preprocessing of the target data is complete, generate display content corresponding to the reasoned content; in response to obtaining a next processing result, use the next processing result as the target processing result and continue executing the above generation steps, wherein the next processing result is the next time-time output processing result obtained by the first intelligent agent continuing to preprocess the target data.

[0174] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0175] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0176] The units described in some embodiments of this disclosure can be implemented in software or in hardware. The described units can also be housed in a processor; for example, a processor may be described as including a first execution unit and a second execution unit. The names of these units do not necessarily limit the unit itself; for example, the second execution unit may also be described as "a unit that, in response to obtaining a next processing result, uses the next processing result as the target processing result and continues to execute the above-described generation steps."

[0177] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0178] Some embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements any of the data processing methods described above.

[0179] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A data processing method, comprising: In response to the current data preprocessing of the target data, the first agent preprocesses the target data to obtain a real-time output target processing result, and based on the target processing result, performs the following generation steps: Using a second intelligent agent, semantic reasoning is performed on the target processing result to obtain the reasoned content; In response to determining that the preprocessing of the target data is complete, display content corresponding to the inference content is generated; In response to obtaining the next processing result, the next processing result is used as the target processing result, and the generation step is continued. The next processing result is the processing result of the next time step output obtained by the first agent continuing to preprocess the target data.

2. The method according to claim 1, wherein, The step of using a second intelligent agent to perform semantic reasoning on the target processing result to obtain the reasoning content includes: Based on the historical reasoning content generated by the second intelligent agent, semantic reasoning is performed on the target processing result using the second intelligent agent to obtain reasoning content. The historical reasoning content is the reasoning result of semantic reasoning based on the historical processing result, and the historical processing result is the preprocessing result of the current time corresponding to the previous time. In response to obtaining the next processing result, the next processing result is used as the target processing result, and the generation step is continued, including: In response to obtaining the next processing result, the next processing result is used as the target processing result, and the reasoning content is used as the historical reasoning content, and the generation step is continued.

3. The method according to claim 2, wherein, The step of using the second agent to perform semantic reasoning on the target processing result based on the historical reasoning content generated by the second agent to obtain the reasoning content includes: In response to detecting that the semantics of the data corresponding to the target processing result satisfy the semantic inversion condition, the historical reasoning content is masked for errors. Using the second intelligent agent, semantic reasoning is performed on the target processing result based on the occluded reasoning content to obtain the reasoning content.

4. The method according to claim 1, wherein, The step of using a first intelligent agent to preprocess the target data to obtain the target processing result in real time includes: Using the first intelligent agent, the target data is processed into data conforming to the cache structure of the key-value pair cache pool, thereby obtaining the target processing result; and After the first intelligent agent preprocesses the target data to obtain the real-time output target processing result, the method further includes: The target processing result is written to the key-value pair cache pool so that the second agent can obtain the target processing result by using a logical pointer pointing to the corresponding content written by the first agent.

5. The method according to claim 1, wherein, Before the step of preprocessing the target data using a first agent in response to the current execution of data preprocessing for the target data to obtain a real-time output target processing result, and performing the following generation steps based on the target processing result, the method further includes: Obtain the reasoning requirements corresponding to the target data; The prompt information corresponding to the reasoning requirement is written into the key-value pair cache pool so that the second agent can learn the reasoning requirement.

6. The method according to claim 4, wherein, The step of using the first intelligent agent to process the target data into data conforming to the cache structure form corresponding to the key-value pair cache pool, and obtaining the target processing result, includes: Using the first intelligent agent, the target data is processed into data conforming to the cache structure form to obtain the initial processing result; The initial processing result is subjected to a predefined structure verification to obtain the verification result; In response to the verification result indicating that the initial processing result conforms to the predefined structure, the initial processing result is determined as the target processing result.

7. The method according to claim 6, wherein, The method further includes: In response to the verification result indicating that the initial processing result does not conform to the predefined structure, the first agent is instructed to re-preprocess the initial processing result to obtain a reprocessed result. In response to the verification result representation corresponding to the reprocessing result conforming to the predefined structure, the reprocessing result is determined as the target processing result.

8. The method according to claim 3, wherein, The second intelligent agent performs semantic reasoning on the target processing result based on the occluded reasoning content to obtain the reasoning content, including: Using the second intelligent agent, the following determination steps are performed: Based on the occluded reasoning content, semantic reasoning is performed on the target processing result to obtain the current reasoning content; Based on the target processing result and the current inference content, generate the prediction inference content corresponding to the next prediction processing result; The current inference content and the predicted inference content are determined as the inference content.

9. The method according to claim 3, wherein, The step of masking erroneous content in the historical reasoning content in response to detecting that the semantics of the data corresponding to the target processing result satisfies the semantic inversion condition includes: In response to the detection that the semantics of the data corresponding to the target processing result satisfy the semantic inversion condition, the following processing steps are performed: Extract contextual reasoning content related to the target processing result from the historical reasoning content, and use it as error content; Construct error mask information for the erroneous content; Based on the error masking information, the historical reasoning content is masked to remove errors.

10. The method according to claim 1, wherein, Before generating the display content corresponding to the inference content in response to determining that the preprocessing of the target data is complete, the method further includes: Determine whether the target processing result contains information related to the end of preprocessing; In response to the determination of existence, the inference confidence level corresponding to the inference content is determined; In response to the inference confidence meeting the confidence condition, it is determined that the preprocessing of the target data is complete.

11. The method according to claim 1, wherein, The first intelligent agent and the second intelligent agent each have their own corresponding short-term memory buffers; as well as The step of using a first intelligent agent to preprocess the target data to obtain the target processing result in real time includes: Retrieve first associated data related to the target data from the long-term memory database; Obtain the context data corresponding to the target data from the short-term memory buffer corresponding to the first agent; The first associated data and the context data are fused to obtain the first fused data; Based on the first fused data, the first intelligent agent is used to preprocess the target data to obtain the target processing result output in real time.

12. The method according to claim 11, wherein, The method further includes: The context data is stored in the short-term memory buffer corresponding to the second agent; and The step of using a second intelligent agent to perform semantic reasoning on the target processing result to obtain the reasoning content includes: Retrieve second related data relating to the target processing result from the long-term memory database; The context data in the short-term memory buffer corresponding to the second agent is fused with the second associated data to obtain the second fused data; Based on the second fused data, the second intelligent agent performs semantic reasoning on the target processing result to obtain the reasoned content.

13. The method according to claim 11, wherein, The step of fusing the first associated data and the context data to obtain the first fused data includes: Determine whether there is a data conflict between the first associated data and the context data; In response to determining that a data conflict exists, conflicting associated sub-data in the first associated data and conflicting context sub-data in the context data are determined; Filter the earliest stored sub-data from the associated sub-data and the context sub-data; The first fused data is obtained by fusing the non-conflicting associated sub-data in the first associated data, the non-conflicting context sub-data in the context data, and the earliest time sub-data.

14. The method according to claim 11, wherein, After determining that preprocessing of the target data is complete and generating the display content corresponding to the inference content, the method further includes: Based on the reasoning, generate reusable empirical data as long-term memory data. Based on the results of each target processing, empirical data supporting reuse is generated, and each long-term memory data is obtained. The obtained long-term memory dataset is stored in the long-term memory database.

15. A data processing apparatus, comprising: The first execution unit is configured to, in response to the current execution of data preprocessing for target data, use a first intelligent agent to preprocess the target data to obtain a target processing result in real time, and based on the target processing result, perform the following generation steps: use a second intelligent agent to perform semantic reasoning for the target processing result to obtain reasoned content; In response to determining that the preprocessing of the target data is complete, display content corresponding to the inference content is generated; The second execution unit is configured to, in response to obtaining the next processing result, take the next processing result as the target processing result and continue to execute the generation step, wherein the next processing result is the processing result of the first agent continuing to preprocess the target data and outputting it at the next time step.

16. An electronic device comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-14.

17. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-14.

18. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-14.

Images