Artificial intelligence-based product data analysis methods, devices, equipment, and media

Abstract

This invention relates to the field of data processing technology and discloses a product data analysis method, apparatus, equipment, and medium based on artificial intelligence. The invention constructs shared inference information and local inference information for each preset analysis target based on product demand information. Based on the reference to the shared inference information and the local inference information, it constructs inference context information for each preset analysis target. It calls the target analysis template corresponding to each preset analysis target from a preset analysis template library to analyze the inference context information of each preset analysis target, obtaining an analysis result summary. Based on the analysis result summary and evaluation information, it evaluates each candidate category in the candidate category set to obtain a recommended category set for the target product. This allows for the analysis of the candidate category set of the target product and related data of product demand information to obtain a recommended category set, achieving stability and reliability in product category decision-making while significantly reducing storage and computing resource overhead.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to product data analysis methods, apparatus, equipment, and media based on artificial intelligence. Background Technology

[0002] To improve the efficiency of product category selection, a common approach is to combine intelligent agent technology with large models. This involves constructing a multi-agent system framework, where a master scheduling agent uses a large model to perform reasoning, task breakdown, and dynamic distribution to sub-agents for parallel execution, thereby shortening the overall analysis cycle. However, this method heavily relies on the autonomous reasoning of the large model, leading to uncertainties in the analysis results. Furthermore, the reasoning process cannot consistently cover all analytical dimensions, resulting in unreliable and unstable category selection decisions. Additionally, the complete isolation of context between sub-agents causes redundant use of basic information, resulting in high overhead in storage and computational resources. Summary of the Invention

[0003] This invention provides a product data analysis method, apparatus, device, and medium based on artificial intelligence, to solve the problems that current product category selection methods cannot provide stable and reliable category selection decisions and have high overhead in storage and computing resources.

[0004] In a first aspect, the present invention provides a product data analysis method based on artificial intelligence, the method comprising: Obtain a set of candidate product categories and product demand information for the target product, and based on the product demand information, construct shared inference information and local inference information for each preset analysis target; Based on the reference to shared reasoning information and local reasoning information, reasoning context information for each preset analysis target is constructed; The target analysis templates corresponding to each preset analysis target are called from the preset analysis template library. The reasoning context information of each preset analysis target is analyzed to obtain a summary of the analysis results of each preset analysis target. Based on the analysis results summary and evaluation information of each preset analysis objective, each candidate category in the candidate category set is evaluated to obtain the recommended category set of the target product.

[0005] In one optional implementation, based on product demand information, shared inference information and local inference information for each preset analysis target are constructed, including: Based on the shared features corresponding to each preset analysis target, product demand information is extracted and the extracted data is stored in a shared memory pool to obtain shared inference information. Based on the dimensional features corresponding to each preset analysis target, product demand information is extracted to obtain local inference information for each preset analysis target.

[0006] In one optional implementation, inference context information for each preset analysis target is constructed based on references to shared inference information and local inference information, including: For each preset analysis target, the reference handle of the shared inference information is copied to the independent context corresponding to the preset analysis target, and the local inference information is deeply copied locally. By assembling the reference handles of shared inference information and local inference information, we obtain the inference context information of the preset analysis target.

[0007] In one optional implementation, a target analysis template corresponding to each preset analysis target is called from a preset analysis template library, and the inference context information of each preset analysis target is analyzed to obtain a summary of the analysis results for each preset analysis target, including: Based on preset matching rules, each preset analysis target is matched with a preset analysis template in the preset analysis template library to determine the target analysis template corresponding to each preset analysis target. Each preset analysis target is called with its corresponding target analysis template to analyze the reasoning context information of each preset analysis target and obtain a summary of the analysis results for each preset analysis target.

[0008] In one optional implementation, based on preset matching rules, each preset analysis target is matched with a preset analysis template in a preset analysis template library to determine the analysis template corresponding to each preset analysis target, including: The first priority is the preset analysis target, the second priority is the coverage of the product label of the target product by the preset analysis template, and the third priority is the historical usage frequency of the preset analysis template. The first priority is higher than the second priority, and the second priority is higher than the third priority. For each preset analysis target, the preset analysis templates are matched based on the first priority, second priority, and third priority to determine the target analysis template corresponding to the preset analysis template.

[0009] In one optional implementation, the target analysis template corresponding to each preset analysis target is invoked respectively to analyze the inference context information of each preset analysis target, thereby obtaining a summary of the analysis results for each preset analysis target, including: Based on the interface type of the data processing interface corresponding to each preset analysis target, a semaphore upper limit is allocated to the data processing interface corresponding to each preset analysis target; The system calls the data processing interfaces corresponding to each preset analysis target in parallel, loads the target analysis template, and performs inference on the inference context information to obtain the analysis result summary of each preset analysis target.

[0010] In one optional implementation, based on the analysis result summary and evaluation information of each preset analysis target, each candidate category in the candidate category set is evaluated to obtain a recommended category set for the target product, including: Based on each candidate product category, data is extracted from the summary of analysis results for each preset analysis objective to obtain the evaluation indicators for each candidate product category corresponding to each preset analysis objective. Based on the evaluation indicators of each candidate category corresponding to each preset analysis objective and the evaluation information of each preset analysis objective, the recommended score for each candidate category is determined. The recommended product categories are sorted based on their recommendation scores to obtain a set of recommended product categories.

[0011] In one alternative implementation, the method further includes: The analysis results summaries for each preset analysis target are verified separately; If the analysis result summary of the first analysis target fails the verification, the analysis result summary of the first analysis target will be regenerated based on the preset downgrade processing method.

[0012] In one optional implementation, based on a preset degradation processing method, a summary of the analysis results for the first analysis target is regenerated, including: If the number of analyses corresponding to the first analysis target is less than the preset number of analyses, the reasoning context information of the first analysis target is re-analyzed to obtain a summary of the analysis results of the first analysis target; If the number of analyses corresponding to the first analysis target reaches the preset number of analyses, or if the analysis of the reasoning context information of the first analysis target does not yield an analysis result summary of the first analysis target, the valid historical results of the first analysis target are queried based on the fingerprint information of the shared reasoning information, and the valid historical results are used as the analysis structure summary of the first analysis target, and the first analysis target is marked as the first degraded state; If no valid historical results for the first analysis target are found based on the fingerprint information of the shared inference information, the category baseline template corresponding to the first analysis target is obtained from the preset analysis template library based on the category tag in the shared inference information, and the analysis result summary of the first analysis target is filled in based on the category baseline template, and the first analysis target is marked as the second downgraded state. If the category baseline template corresponding to the first analysis target does not exist in the preset analysis template library, the data field of the analysis result summary of the first analysis target will be empty, and the first analysis target will be marked as the third downgraded state.

[0013] In one optional implementation, the evaluation information for the preset analysis target includes confidence information and weight information, wherein the confidence information is determined in the following manner: If the preset analysis target is not marked as downgraded, the confidence information of the preset analysis target is determined based on the data time characteristics, data source characteristics, and category coverage characteristics of the analysis result summary of the preset analysis target; If the preset analysis target is marked as the first degraded state, the confidence information of the preset analysis target is determined based on the caching duration of valid historical results; If the preset analysis target is marked as the second degraded state, the first preset confidence level will be used as the confidence level information of the preset analysis target; If the preset analysis target is marked as the third downgraded state, the second preset confidence level is used as the confidence level information of the preset analysis target, wherein the second preset confidence level is less than the first preset confidence level.

[0014] In one alternative implementation, the method further includes: If the analysis result summary of the first analysis target fails the verification, it is determined that the target analysis template corresponding to the first analysis target is not matched. If the analysis result summary of the second analysis target passes the verification, it is determined that the target analysis template corresponding to the second analysis target has been matched, and the historical usage count of the target analysis template corresponding to the second analysis target is updated.

[0015] In one alternative implementation, the method further includes: If the number of historical uses of the first analysis template in the preset analysis template library is greater than the first preset number of uses, the first analysis template will be used as the recommended template. If a second analysis template exists in the preset analysis template library but is not matched for the second preset number of consecutive times, the second analysis template will be used as a backup template.

[0016] In one alternative implementation, the method further includes: Obtain the consistency assessment rule base, which includes multiple consistency assessment rules, each corresponding to a different rule priority. Following the order of rule priority, the consistency between the analysis result summaries of each preset analysis objective is evaluated sequentially based on multiple consistency evaluation rules to obtain a consistency score.

[0017] Secondly, the present invention provides a product data analysis device based on artificial intelligence, the device comprising: The first information construction module is used to obtain the candidate category set and product demand information of the target product, and based on the product demand information, construct shared inference information and local inference information for each preset analysis target. The second information construction module is used to construct the reasoning context information of each preset analysis target based on the reference to shared reasoning information and local reasoning information; The information analysis module is used to call the target analysis templates corresponding to each preset analysis target from the preset analysis template library, analyze the reasoning context information of each preset analysis target, and obtain the analysis result summary of each preset analysis target. The category evaluation module is used to evaluate each candidate category in the candidate category set based on the analysis result summary and evaluation information of each preset analysis objective, and to obtain the recommended category set of the target product.

[0018] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the artificial intelligence-based product data analysis method described in the first aspect or any corresponding embodiment thereof.

[0019] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the artificial intelligence-based product data analysis method of the first aspect or any corresponding embodiment described above.

[0020] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the artificial intelligence-based product data analysis method described in the first aspect or any corresponding embodiment thereof.

[0021] The product data analysis method based on artificial intelligence provided in this invention constructs shared inference information and local inference information for each preset analysis target based on product demand information. This allows for the direct extraction of shared and local inference information from product demand information without relying on a large model, fundamentally eliminating the uncertainty of analysis results and the instability of analysis dimension coverage caused by the uncertainty of large model inference. Simultaneously, based on the reference to shared inference information and local inference information, inference context information for each preset analysis target is constructed. This allows for the construction of inference for each analysis target by using the reference to shared inference information instead of content copying. Contextual information significantly reduces storage and computing resource overhead. Furthermore, by calling target analysis templates corresponding to each preset analysis objective from a preset analysis template library, the inference context information of each preset analysis objective is analyzed to obtain a summary of the analysis results for each preset analysis objective. This summary format represents the analysis results, shortening the analysis cycle and improving analysis efficiency. Based on the summary of the analysis results and evaluation information of each preset analysis objective, each candidate category in the candidate category set is evaluated to obtain a recommended category set for the target product. This, combined with the evaluation information of each preset analysis objective, quantitatively evaluates each summary of analysis results, ensuring the consistency and stability of product data analysis decisions. Therefore, by analyzing the candidate category set of the target product and relevant data on product demand information, a recommended category set is obtained, achieving stability and reliability in product category decisions while significantly reducing storage and computing resource overhead. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the first process of a product data analysis method based on artificial intelligence according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a second process for a product data analysis method based on artificial intelligence according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the third process of the product data analysis method based on artificial intelligence according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the fourth process of the product data analysis method based on artificial intelligence according to an embodiment of the present invention; Figure 5This is a structural block diagram of an artificial intelligence-based product data analysis device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0026] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0027] To improve the efficiency of product category selection, a common approach is to combine intelligent agent technology with large models. This involves constructing a multi-agent system framework, where a master scheduling agent uses a large model to perform reasoning, task breakdown, and dynamic distribution to sub-agents for parallel execution, thereby shortening the overall analysis cycle. However, this method heavily relies on the autonomous reasoning of the large model, leading to uncertainties in the analysis results. Furthermore, the reasoning process cannot consistently cover all analytical dimensions, resulting in unreliable and unstable category selection decisions. Additionally, the complete isolation of context between sub-agents causes redundant use of basic information, resulting in high overhead in storage and computational resources.

[0028] Furthermore, in current multi-agent system frameworks, after a sub-agent completes its execution, the main scheduling agent simply concatenates or summarizes the natural language returned by the sub-agent, lacking a cross-dimensional, structured, and executable cross-validation mechanism. This leads to potential internal contradictions in the final category selection decision and a lack of traceable decision-making basis. Moreover, when a sub-agent fails due to external tool timeouts, limited data source access, or unstable model services, the main scheduling agent often directly interrupts the entire task or marks it as failed, lacking the ability to output decisions when dimensions are missing.

[0029] To address the aforementioned problems, this invention provides an artificial intelligence-based product data analysis method. Based on product demand information, it constructs shared inference information and local inference information for each preset analysis target. This allows for the direct extraction of shared and local inference information from product demand information without relying on a large model, fundamentally eliminating the uncertainty of analysis results and the instability of analysis dimension coverage caused by the uncertainty of large model inference. Simultaneously, based on the reference to shared inference information and local inference information, it constructs inference context information for each preset analysis target, thereby using the reference to shared inference information instead of content copying to construct each analysis target. The reasoning context information is significantly reduced, thus lowering storage and computing resource overhead. Furthermore, target analysis templates corresponding to each preset analysis objective are retrieved from a preset analysis template library to analyze the reasoning context information of each preset analysis objective, obtaining a summary of the analysis results for each preset analysis objective. This summary format represents the analysis results, shortening the analysis cycle and improving analysis efficiency. Based on the summary of the analysis results and evaluation information of each preset analysis objective, each candidate category in the candidate category set is evaluated to obtain a recommended category set for the target product. This, combined with the evaluation information of each preset analysis objective, quantitatively evaluates each summary of analysis results, ensuring the consistency and stability of product data analysis decisions. Therefore, by analyzing the candidate category set of the target product and relevant data on product demand information, a recommended category set is obtained, achieving stability and reliability in product category decisions while significantly reducing storage and computing resource overhead.

[0030] The product data analysis method based on artificial intelligence provided by this invention can be used in any scenario that requires analysis of product data and decision-making on product categories, such as product production planning, new product project initiation decision-making, market research and analysis, and product launch. No specific limitations are imposed here.

[0031] According to an embodiment of the present invention, an embodiment of a product data analysis method based on artificial intelligence is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0032] This embodiment provides a product data analysis method based on artificial intelligence. Figure 1 This is a schematic diagram of the first type of product data analysis method based on artificial intelligence according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the candidate category set and product demand information of the target product, and based on the product demand information, construct shared inference information and local inference information for each preset analysis target.

[0033] In this embodiment of the invention, the target product is the product for which category selection is required. The candidate category set consists of multiple candidate categories of the target product that need to be evaluated, determined based on factors such as product production and deployment needs, market research, or business experience. For example, if the target product is a portable energy storage power supply, the candidate category set may include three candidate categories: 500W portable energy storage power supply, 1000W portable energy storage power supply, and 2000W portable energy storage power supply. Product demand information is information input by the user to describe the relevant business scenarios of the target product, including but not limited to target deployment area, budget range, target sales channel, time window, and brand label. The target deployment area represents the expected deployment area or region of the target product; the budget range represents the expected resource consumption during the production, manufacturing, and transportation of the target product; the target sales channel represents the sales method of the target product; the time window identifies the deployment time node of the target product; and the brand label identifies the brand positioning of the target product. Product demand information can be obtained by the user submitting it through a form, voice input, or text input on the front-end interface, or by reading it from the business system database.

[0034] In this embodiment of the invention, the preset analysis objectives are pre-defined analytical dimensions that need to be completed in the category selection decision. Each preset analysis objective corresponds to an analytical agent, meaning that an analytical agent is constructed for each preset analysis objective, and the analytical agent is used to execute the analytical tasks corresponding to the preset analysis objective. The preset analysis objectives may include product demand trend and keyword analysis objectives, similar product research and analysis objectives, target audience profile analysis objectives, and user push analysis objectives. The product demand trend and keyword analysis objective is used to analyze the search popularity changes, keyword opportunities, and seasonal characteristics of the target product in the target region; the similar product research and analysis objective is used to analyze the distribution pattern, price distribution, and market concentration of similar products and related products in the target region; the target audience profile analysis objective is used to analyze the characteristics and consumption intentions of the potential user group of the target product; and the user push analysis objective is used to analyze the push channels and push direction of the target product. The number and type of preset analysis objectives can be adjusted according to actual business needs and are not specifically limited here.

[0035] In this embodiment of the invention, the shared inference information is the basic information commonly required by all preset analysis targets. Since each preset analysis target needs to be analyzed based on the same target region, candidate category set, budget range, target sales channel, time window, and brand tags, storing a complete copy of this basic information in each preset analysis target's corresponding analysis agent would waste storage and computing resources. Therefore, this embodiment extracts common basic information from each preset analysis target based on product demand information to construct shared inference information, reusing the same basic information among the preset analysis targets. Local inference information is the configuration information for each preset analysis target's respective analysis dimensions. Different preset analysis targets require different exclusive dimension data as local inference information to complete their respective analysis tasks. For the configuration information requirements of different preset analysis targets, corresponding local inference information is extracted from the product demand information. For example, the product demand trend and keyword analysis target requires keywords and regional time-series windows as local inference information, while the similar product survey analysis target requires similar product anchor points as local inference information.

[0036] Step S102: Based on the reference to shared reasoning information and local reasoning information, construct the reasoning context information for each preset analysis target.

[0037] In this embodiment of the invention, the inference context information refers to the contextual analysis environment upon which each preset analysis target relies when performing its analysis task. Each preset analysis target's corresponding analysis agent has its own independent inference context information, which is isolated from each other to ensure that the preset analysis targets do not interfere with each other during execution. Simultaneously, to avoid wasting storage and computational resources, shared inference information is reused among the preset analysis targets. This is achieved by combining references to shared inference information, such as pointers or reference handles pointing to shared inference information, with the local inference information of the preset analysis target to construct its inference context information.

[0038] Step S103: Call the target analysis template corresponding to each preset analysis target from the preset analysis template library, analyze the reasoning context information of each preset analysis target, and obtain the analysis result summary of each preset analysis target.

[0039] In this embodiment of the invention, the preset analysis template library is a database or storage space storing multiple preset analysis templates. Each preset analysis template includes execution steps, tool call sequences, input constraints, and output format definitions for a specific analysis task type. The preset analysis template is an encapsulation of analysis paths that have been validated through historical execution. By directly reusing validated analysis logic, it directly performs analysis and reasoning on each preset analysis target, thereby eliminating the dependence on a large language model and ensuring the determinism and reproducibility of the analysis process.

[0040] In this embodiment of the invention, the analysis result summary is a structured result returned after each preset analysis target completes the analysis. The analysis result summary includes a data field, an execution status field, a confidence level field, a version field, and a failure reason field. Specifically, the data field is filled with a predefined JSON schema according to the corresponding preset analysis target, containing the core conclusion data of that dimension's analysis; the execution status field indicates whether the analysis was successful or failed; the confidence level field quantifies the credibility of the analysis results for that dimension; the schema version field identifies the schema version followed by the data field, supporting compatibility and evolution between different schema versions; and the failure reason field records the reason for failure when execution fails. Unlike related technologies that use natural language strings to return summaries, this embodiment of the invention uses a structured, typed summary, enabling the analysis results of each dimension to be quantitatively parsed, verified, and integrated by subsequent steps, providing a data foundation for cross-dimensional cross-validation and traceable decision-making.

[0041] Step S104: Based on the analysis result summary and evaluation information of each preset analysis target, evaluate each candidate category in the candidate category set to obtain the recommended category set of the target product.

[0042] In this embodiment of the invention, the evaluation information is used to quantify the credibility of the analysis result summaries of each preset analysis target and their importance in category decision-making. For each candidate category in the candidate category set, the analysis result summaries and evaluation information related to each preset analysis target and that candidate category are comprehensively evaluated and quantified to achieve quantitative scoring for the recommendation and selection of candidate categories. From these, the candidate categories whose quantitative scores meet the user's needs are selected to obtain the recommended category set for the target product.

[0043] The product data analysis method based on artificial intelligence provided in this invention constructs shared inference information and local inference information for each preset analysis target based on product demand information. This allows for the direct extraction of shared and local inference information from product demand information without relying on a large model, fundamentally eliminating the uncertainty of analysis results and the instability of analysis dimension coverage caused by the uncertainty of large model inference. Simultaneously, based on the reference to shared inference information and local inference information, inference context information for each preset analysis target is constructed. This allows for the construction of inference for each analysis target by using the reference to shared inference information instead of content copying. Contextual information significantly reduces storage and computing resource overhead. Furthermore, by calling target analysis templates corresponding to each preset analysis objective from a preset analysis template library, the inference context information of each preset analysis objective is analyzed to obtain a summary of the analysis results for each preset analysis objective. This summary format represents the analysis results, shortening the analysis cycle and improving analysis efficiency. Based on the summary of the analysis results and evaluation information of each preset analysis objective, each candidate category in the candidate category set is evaluated to obtain a recommended category set for the target product. This, combined with the evaluation information of each preset analysis objective, quantitatively evaluates each summary of analysis results, ensuring the consistency and stability of product data analysis decisions. Therefore, by analyzing the candidate category set of the target product and relevant data on product demand information, a recommended category set is obtained, achieving stability and reliability in product category decisions while significantly reducing storage and computing resource overhead.

[0044] This embodiment provides a product data analysis method based on artificial intelligence. Figure 2 This is a schematic diagram of a second process for a product data analysis method based on artificial intelligence according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the candidate category set and product demand information of the target product, and construct shared inference information and local inference information for each preset analysis target based on the product demand information.

[0045] Specifically, step S201 includes: Step S2011: Based on the shared features corresponding to each preset analysis target, product demand information is extracted and the extracted data is stored in a shared memory pool to obtain shared inference information.

[0046] In this embodiment of the invention, the shared feature is a pre-defined product requirement information field that is commonly needed by all pre-defined analysis targets. The product requirement information may contain data with various attributes, such as target delivery area, budget range, target sales channel, time window, brand tags, user historical behavior data, competitor identifiers, etc. Only a portion of this data is needed by all pre-defined analysis targets; this portion of data corresponds to the shared feature. To avoid indiscriminately storing all product requirement information in a shared memory pool, thus wasting storage, and to avoid injecting irrelevant information into the context, thus causing interference, shared features are pre-defined for each pre-defined analysis target. These shared features are used to identify and extract data fields commonly needed by all pre-defined analysis targets from the product requirements, serving as shared inference information.

[0047] In one optional implementation, the product demand information can be parsed according to a preset data structure to filter out the field values ​​corresponding to the shared features; alternatively, query conditions can be constructed based on the shared features to read the data of the corresponding fields from the product demand information. The extracted shared inference information is stored in a shared memory pool, which is a storage area in the system memory allocated for storing shared inference information. Only one physical copy of the shared inference information is retained in the shared memory pool.

[0048] Step S2012: Based on the dimensional features corresponding to each preset analysis target, product demand information is extracted to obtain local inference information for each preset analysis target.

[0049] In this embodiment of the invention, the dimensional features are preset product demand information fields associated with a specific preset analysis target. Different preset analysis targets require different dedicated data as input when performing analysis tasks; the dedicated data required by each preset analysis target is different. For each preset analysis target, data from the corresponding fields are extracted from the product demand information based on its corresponding dimensional features to obtain the local inference information for that preset analysis target. The local inference information for each preset analysis target is independent and stored in the independent context of the analysis agent corresponding to each preset analysis target.

[0050] Step S202: Based on the reference to shared reasoning information and local reasoning information, construct the reasoning context information for each preset analysis target.

[0051] Specifically, step S202 includes: Step S2021: For each preset analysis target, copy the reference handle of the shared inference information to the independent context corresponding to the preset analysis target, and perform a local deep copy of the local inference information.

[0052] In this embodiment of the invention, each preset analysis target corresponds to an analysis agent with its own independent context. These independent contexts are isolated from each other to ensure that the preset analysis targets do not interfere with each other during execution. A reference handle is a pointer or reference identifier pointing to the physical address of shared inference information in a shared memory pool. The reference handle of the shared inference information is copied to the independent context corresponding to the preset analysis target, thereby enabling the reuse of a physical copy across multiple analysis agents and avoiding duplicate storage of the same basic information. The specific form of the reference handle can be any information that can uniquely identify the storage location of the shared inference information, such as a memory address, index value, or key value; this invention does not impose specific limitations on this.

[0053] In this embodiment of the invention, local deep copy is an operation that completely copies the content of local inference information to an independent context. Since different preset analysis targets may need to modify or expand their local inference information, such as dynamically supplementing keyword lists or adjusting demographic hypotheses during analysis, if multiple analysis targets share data segments in the local inference information, a modification by one preset analysis target may affect the analysis results of other analysis targets. Therefore, through local deep copy, a single copy of the local inference information is used uniquely in the analysis agent corresponding to each preset analysis target, ensuring that modifications to the local inference information by each preset analysis target do not affect other analysis targets, thus achieving data write isolation.

[0054] In one optional implementation, a reference counter is set for the shared inference information. The reference counter is initially set to zero, and its maximum value is set to the number of preset analysis targets. When the reference handle of the shared inference information is copied to the independent context of a preset analysis target, the reference counter is incremented by one. When the preset analysis target completes its analysis task and releases the reference, the reference counter is decremented by one. When the reference counter reaches zero, the system performs memory reclamation on the shared inference information. This precisely manages the lifecycle of the shared inference information and avoids memory leaks.

[0055] Step S2022: Assemble the reference handle of the shared inference information and the local inference information to obtain the inference context information of the preset analysis target.

[0056] In this embodiment of the invention, assembly refers to the operation of organizing the reference handle of shared inference information and the local inference information together according to a preset data structure to form a data environment that the preset analysis target can fully access when performing the analysis task. The data structure form of the assembled inference context information is not specifically limited. For example, the reference handle and the local inference information can be stored as two fields of a context structure, or the reference handle can be used as a prefix and the local inference information as a suffix to concatenate them into a continuous data area. A nested structure can also be used, where the shared inference information encapsulated by the reference handle is used as a remote data source reference, and the local inference information is used as local cached data, etc.

[0057] In one optional implementation, when the preset analysis target needs to read basic information from the product requirement information during the execution of the analysis task, it reads the shared inference information from the shared memory pool through the reference handle; when it needs to read the configuration information specific to the analysis target, it directly reads the local inference information from the independent context. The two parts of information together constitute the complete inference context of the preset analysis target.

[0058] Step S203: The target analysis template corresponding to each preset analysis target is retrieved from the preset analysis template library. The inference context information of each preset analysis target is analyzed to obtain a summary of the analysis results for each preset analysis target. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0059] Step S204: Based on the analysis result summaries and evaluation information of each preset analysis objective, evaluate each candidate category in the candidate category set to obtain the recommended category set for the target product. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0060] The product data analysis method based on artificial intelligence provided in this invention extracts shared inference information common to all analysis targets from product demand information based on shared features and stores it in a shared memory pool. It also extracts local inference information specific to each analysis target based on dimensional features, thereby achieving separate storage of shared inference information and local inference information, providing a data foundation for reference-based context construction. On this basis, by copying the reference handle of shared inference information to each independent context and performing a local deep copy of the local inference information and assembling it to obtain inference context information, the shared inference information can be reused by all analysis targets with only one physical copy, significantly reducing the overhead of storage and computing resources, while ensuring write isolation of local inference information for each analysis target.

[0061] This embodiment provides a product data analysis method based on artificial intelligence. Figure 3This is a schematic diagram of the third process of the product data analysis method based on artificial intelligence according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Obtain the candidate category set and product demand information for the target product, and based on the product demand information, construct shared inference information and local inference information for each preset analysis target. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0062] Step S302: Based on references to shared inference information and local inference information, construct the inference context information for each preset analysis target. For details, please refer to [link to relevant documentation]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0063] Step S303: Call the target analysis template corresponding to each preset analysis target from the preset analysis template library, analyze the reasoning context information of each preset analysis target, and obtain the analysis result summary of each preset analysis target.

[0064] Specifically, step S303 includes: Step S3031: Based on the preset matching rules, match each preset analysis target with the preset analysis template in the preset analysis template library to determine the target analysis template corresponding to each preset analysis target.

[0065] In this embodiment of the invention, a preset analysis template library stores multiple preset analysis templates. Different preset analysis templates are suitable for different preset analysis targets, different product categories, and have different historical performance. To ensure that each preset analysis target obtains the most suitable analysis template, template matching is performed based on preset matching rules to determine the target analysis template corresponding to the preset analysis target.

[0066] In one optional implementation, the preset matching rules employ a three-tiered matching strategy: the preset analysis target is given the first priority, the product tag coverage of the target product by the preset analysis template is given the second priority, and the historical usage frequency of the preset analysis template is given the third priority. The first priority is higher than the second priority, and the second priority is higher than the third priority. For each preset analysis target, the preset analysis template is matched based on the first, second, and third priorities to determine the corresponding target analysis template.

[0067] In one optional implementation, the first priority requires that the applicable task type of the preset analysis template is consistent with the preset analysis objective. For example, product demand trend and keyword analysis objectives can only match templates applicable to trend analysis tasks, not templates applicable to similar product research tasks. After filtering by the first priority, a second priority is used for further matching, comparing the intersection between the applicable category tags of each candidate template and the product tags of the target product, i.e., coverage, and prioritizing the template with the highest coverage. If the coverage is the same, a third priority is used for further matching to select the template with the highest historical usage frequency, such as cumulative hit count or hit rate. Thus, through a three-stage matching, it is ensured that the template loaded for each preset analysis objective is both the most relevant to the category within the same task type and has the best historical performance.

[0068] Step S3032: Call the target analysis template corresponding to each preset analysis target respectively, analyze the reasoning context information of each preset analysis target, and obtain the analysis result summary of each preset analysis target.

[0069] In this embodiment of the invention, after determining the target analysis template corresponding to each preset analysis target, each target analysis template is invoked to analyze and reason about the corresponding reasoning context information. Since the preset analysis targets are independent of each other and there is no data dependency, parallel execution is adopted to improve analysis efficiency.

[0070] In one optional implementation, each preset analysis target needs to call corresponding data processing interfaces when performing analysis, such as search trend query interfaces, similar product data collection interfaces, comment data mining interfaces, and advertising and promotion data interfaces. Different types of data processing interfaces have different access costs and stability characteristics. For example, search / trend interfaces typically have faster response times and lower costs, and can support higher concurrent access, while data estimation interfaces typically have slower response times and higher costs, and frequent calls may overload external interfaces. Based on this, before parallel scheduling, the interface types of the data processing interfaces corresponding to each preset analysis target are identified and classified. Based on the interface types of the data processing interfaces corresponding to each preset analysis target, a semaphore upper limit is allocated to each preset analysis target's corresponding data processing interface. The semaphore upper limit is used to control the number of concurrent accesses to various types of data processing interfaces. High-cost or high-risk interfaces are configured with smaller semaphore upper limits, while low-cost or low-risk interfaces are configured with larger semaphore upper limits. After the semaphore allocation is completed, the data processing interfaces corresponding to each preset analysis target are called in parallel, and the target analysis template is loaded to infer the inference context information to obtain the analysis result summary of each preset analysis target.

[0071] In one optional implementation, the analysis result summaries of each preset analysis target are validated separately. Since the analysis result summaries returned by each preset analysis target are structured and typed summaries, and their data fields are filled according to the predefined JSON Schema of the corresponding preset analysis target, validation is performed based on the Schema definition corresponding to each preset analysis target. Specifically, for each preset analysis target, the corresponding version of the JSON Schema definition is loaded according to the Schema version field in its analysis result summary. Field type validation is performed on the data fields to check whether the data type of each field is consistent with the Schema definition; mandatory field validation is performed to check whether fields marked as mandatory in the Schema exist and are not empty; and value range validation is performed to check whether numeric fields are within the reasonable range defined in the Schema.

[0072] In one optional implementation, if the analysis result summary of the first analysis target fails the verification, i.e., the data field of the analysis target fails any of the above verifications, the analysis result summary of the first analysis target is regenerated based on a preset degradation processing method. The preset degradation processing method includes a four-level degradation path: the first level is automatic retry, the second level is cache hit degradation, the third level is category baseline filling degradation, and the fourth level is explicit annotation of missing data.

[0073] Specifically, if the number of analyses corresponding to the first analysis target is less than the preset number of analyses, such as 2 times, it enters the first level of degradation, and re-analyzes the reasoning context information of the first analysis target to obtain the analysis result summary of the first analysis target. That is, the above steps of analyzing the reasoning context information are executed again to try to re-infer and obtain the analysis result summary. In this way, the automatic retry mechanism is used to deal with analysis failures caused by temporary failures such as instantaneous timeouts of external tools or network jitter.

[0074] If the number of analyses corresponding to the first analysis target reaches the preset number of analyses, or if the analysis of the reasoning context information of the first analysis target fails to yield an analysis result summary, the process enters the second level of degradation. Based on the fingerprint information of the shared reasoning information, valid historical results of the first analysis target are queried, and these valid historical results are used as the analysis structure summary of the first analysis target, marking the first analysis target as the first degradation state. The fingerprint information of the shared reasoning information is a unique identifier obtained by hashing the shared reasoning information; identical or similar product requirement information will generate the same or similar fingerprints. If a valid historical successful result matching this fingerprint exists in the cache, this valid historical result is used as the analysis result summary of the first analysis target. This reuse of recently verified analysis results ensures the availability of conclusions while avoiding redundant calculations.

[0075] If the fingerprint information based on shared inference information fails to retrieve valid historical results for the first analysis target, the process enters the third level of degradation. Based on the category tags in the shared inference information, the category baseline template corresponding to the first analysis target is retrieved from the preset analysis template library, and the analysis result summary of the first analysis target is populated based on the category baseline template. This population operation merges the category baseline template with the currently analyzed data, marking the first analysis target as a second degradation state. The category baseline template is a default analysis conclusion template preset for a specific category. It contains general conclusion data for that category in that analysis dimension. Although it lacks personalized analysis for specific candidate categories, it ensures the overall reasonableness of the conclusions. Furthermore, merging it with the currently analyzed data maximizes the reuse of real data.

[0076] If the baseline template for the category corresponding to the first analysis target does not exist in the preset analysis template library, the system enters the fourth level of degradation. In this degradation state, the data fields of the summary of the analysis results for the first analysis target are set to empty, and the first analysis target is marked as the third degradation state. In this degradation state, the confidence level of this analysis target will be set to zero during subsequent fusion evaluation, and its contribution to the final recommendation score will be completely masked. However, the overall task will not be interrupted due to the lack of data in this dimension. Thus, through this four-level degradation mechanism, the system can still output usable recommendation conclusions even when any single-dimensional analysis fails, significantly improving the robustness and usability of the product data analysis method.

[0077] In one optional implementation, if the analysis result summary verification of the first analysis target fails, it is determined that the target analysis template corresponding to the first analysis target is not matched, meaning the execution result of the template in this matching is a failure. If the analysis result summary verification of the second analysis target passes, it is determined that the target analysis template corresponding to the second analysis target is matched, meaning the execution result of the template in this matching is a success. The historical usage count of the target analysis template corresponding to the second analysis target is then updated, for example, by incrementing the cumulative hit count and recalculating the hit rate. Through this statistical mechanism, the hit status of each template in the preset analysis template library is continuously recorded, providing a data foundation for the automatic promotion and demotion of templates.

[0078] In one optional implementation, templates in the preset analysis template library are dynamically managed based on historical usage frequency. If a first analysis template in the preset analysis template library has a historical usage frequency greater than a first preset frequency (e.g., the cumulative hit count exceeds the promotion threshold, or the hit rate exceeds the promotion threshold), the first analysis template is recommended as a template so that subsequent similar tasks can prioritize calling this template during inference analysis. If a second analysis template in the preset analysis template library fails to hit for a second consecutive preset number of times, the second analysis template is considered as a candidate template, placing it at a lower priority during template matching or removing it from the preferred recommendation range. Thus, through the above-mentioned automatic promotion and demotion mechanism, templates with excellent historical performance in the preset analysis template library are prioritized, while poorly performing templates are gradually eliminated or weakened, thereby achieving continuous optimization of template content.

[0079] In an optional implementation, cross-validation can be performed on the consistency of the analysis result summaries of each preset analysis target to avoid conflicts or contradictions between the analysis result summaries of different preset analysis targets. This involves obtaining a consistency evaluation rule base and calling consistency evaluation rules from the base for cross-validation. The consistency evaluation rule base includes multiple consistency evaluation rules, each corresponding to a different rule priority. Each consistency evaluation rule is used to verify the logical consistency between the analysis result summaries of two preset analysis targets.

[0080] In one optional implementation, each consistency evaluation rule is defined by a six-tuple, including two preset analysis targets involved in the verification, a consistency determination expression, a consistency threshold, a conflict handling action, and a rule priority. The consistency determination expression is a logical expression that takes data fields from the analysis result summaries of the two preset analysis targets as input, and its return value characterizes the degree of consistency between the two preset analysis targets. The consistency threshold is used to determine whether the consistency level meets the requirements. The conflict handling action is used to perform corresponding actions when the consistency verification fails. The rule priority is used to determine the execution order among multiple consistency evaluation rules.

[0081] In one optional implementation, the consistency between the analysis result summaries of each preset analysis target is evaluated sequentially based on multiple consistency evaluation rules according to rule priority, resulting in a consistency score. Specifically, consistency evaluation rules are extracted one by one in descending order of rule priority. Data fields requiring verification are extracted from the analysis result summaries of the two preset analysis targets involved in the rule, substituted into the consistency judgment expression for evaluation, and a consistency score is obtained. This consistency score is compared with a consistency threshold; if the consistency score is lower than the consistency threshold, a conflict is identified. For consistency evaluation rules with conflicts, the corresponding conflict handling action is taken.

[0082] In one optional implementation, conflict handling actions may include, but are not limited to, the following: The first type is warning, where only conflict information is recorded upon detection, without interrupting the subsequent evaluation process, and the final recommendation includes a warning message to alert the business party to the cross-dimensional contradiction; the second type is downgrading, where the confidence level of the preset analysis target involved in the conflict is lowered to weaken its influence in subsequent weighted fusion; the third type is blocking, where an anomaly is thrown and the current evaluation process is interrupted, requiring intervention from upper-level scheduling or the business party; the fourth type is requesting manual review, where the conflict information is pushed to a manual review queue for business personnel to determine how to handle the conflict, without interrupting the current evaluation process. Thus, through the above-described categorized handling actions, a flexible and controllable conflict handling mechanism is provided while ensuring the reliability of the recommendation conclusion.

[0083] In one optional implementation, when any of the preset analysis targets participating in the verification are in a degraded state—that is, the preset analysis target has been marked as a first degraded state, a second degraded state, or a third degraded state—the reliability of the analysis result summary of that analysis target is low, and its significance in participating in consistency verification is limited. In this case, the rule can be marked as a degraded verification and skipped, and the consistency score of the consistency evaluation rule can be recorded as inapplicable, while not interrupting the execution of other rules. This avoids false alarm conflicts caused by abnormal data in the degraded dimension, which could interfere with the cross-validation of the normal dimension.

[0084] In one alternative implementation, after cross-validation is completed, the consistency score and conflict resolution records can be written into the audit log as part of the basis for subsequent traceable decision-making.

[0085] In one optional implementation, when a consistency assessment rule triggers a blocking action, the overall assessment process is interrupted, but the summary of analysis results for each pre-defined analysis objective that has already been completed is not discarded. After capturing the blocking anomaly, the pre-defined analysis objective involved in triggering the blocking action is marked as partially downgraded, and the confidence level of this dimension is lowered to a set confidence level value. Then, the assessment process is re-entered, thereby maximizing the use of existing analysis results while ensuring the internal consistency of the recommended conclusions.

[0086] Therefore, by performing cross-dimensional cross-validation on the analysis results summaries of each preset analysis target, logical conflicts between different analysis dimensions can be effectively identified, avoiding internal inconsistencies in recommendation decisions caused by cross-dimensional contradictions, thereby improving the reliability of the final recommended category set and the credibility of the decision basis. At the same time, through rule priority ranking, conflict classification and handling, and a rule skipping mechanism with degradation awareness, both processing efficiency and system robustness are taken into account while ensuring the verification effect.

[0087] Step S304: Based on the analysis result summaries and evaluation information of each preset analysis objective, evaluate each candidate category in the candidate category set to obtain the recommended category set for the target product. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0088] The product data analysis method based on artificial intelligence provided in this invention employs a three-stage matching rule to match the most suitable target analysis template from a preset analysis template library for each preset analysis target, ensuring the determinism and reproducibility of the matching results. Furthermore, it achieves automatic promotion and demotion of templates through the statistics of hit and miss results, continuously optimizing the quality of the preset analysis template library. Simultaneously, by allocating a semaphore upper limit based on the interface type and then calling each target analysis template in parallel to analyze the inference context information, it maximizes the parallel execution efficiency while ensuring that the external interface is not compromised, effectively shortening the overall analysis cycle.

[0089] This embodiment provides a product data analysis method based on artificial intelligence. Figure 4 This is a schematic diagram of the fourth process of the product data analysis method based on artificial intelligence according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps: Step S401: Obtain the candidate category set and product demand information for the target product, and based on the product demand information, construct shared inference information and local inference information for each preset analysis target. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0090] Step S402: Based on references to shared inference information and local inference information, construct the inference context information for each preset analysis target. For details, please refer to [link to relevant documentation]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0091] Step S403: The target analysis template corresponding to each preset analysis target is retrieved from the preset analysis template library. The inference context information of each preset analysis target is analyzed to obtain a summary of the analysis results for each preset analysis target. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0092] Step S404: Based on the analysis result summary and evaluation information of each preset analysis target, evaluate each candidate category in the candidate category set to obtain the recommended category set of the target product.

[0093] Specifically, step S404 includes: Step S4041: Based on each candidate category, extract data from the analysis result summary of each preset analysis target to obtain the evaluation index of each candidate category corresponding to each preset analysis target.

[0094] In this embodiment of the invention, for each candidate category in the candidate category set, indicator data related to that candidate category is extracted from the analysis result summary of each preset analysis objective. For example, indicators such as search volume growth rate and trend slope are extracted from the analysis result summary of the product demand trend and keyword analysis objective; indicators such as market concentration and competition index are extracted from the analysis result summary of the similar product survey analysis objective; indicators such as audience matching degree and consumption intention score are extracted from the analysis result summary of the target audience profile analysis objective; and indicators such as push reach rate and conversion prediction are extracted from the analysis result summary of the user push analysis objective. The specific type and number of each evaluation indicator can be predetermined by the schema definition of each preset analysis objective.

[0095] In one optional implementation, since the evaluation indicators under different analytical dimensions have different dimensions and numerical ranges, after extracting the evaluation indicators, the evaluation indicators of each candidate category under each preset analytical objective are normalized. Specifically, within the preset analytical objective, the evaluation indicators of all candidate categories are normalized and mapped to the [0, 1] interval to eliminate the interference of the difference in dimensions between different dimensions on the subsequent weighted fusion.

[0096] Step S4042: Based on the evaluation indicators of each candidate category corresponding to each preset analysis target and the evaluation information of each preset analysis target, determine the recommended score for each candidate category.

[0097] In this embodiment of the invention, the evaluation information includes confidence information and weight information for each preset analysis target. The weight information represents the preset importance of each analysis target in the final recommendation score. For example, the weights of the four preset analysis targets can be set to 0.25, 0.25, 0.30, and 0.20, respectively, with specific values ​​pre-set by the business side according to actual decision-making needs. The confidence information is used to quantify the credibility of the analysis results for each analysis target. Based on the confidence information and weight information of each preset analysis target, the evaluation indicators corresponding to each preset analysis target for each candidate category are weighted and fused to determine the recommendation score for each candidate category.

[0098] In one alternative implementation, the recommendation score can be calculated using the following formula:

[0099] in, Candidate categories Recommended rating, To pre-set analysis objectives Weight information, To pre-set analysis objectives Confidence information, Candidate categories Corresponding to the preset analysis target Evaluation indicators.

[0100] In this formula, the sum of the products of weight information, confidence information, and evaluation indicators of each candidate indicator under each preset analysis target is used as the numerator, and the sum of the products of weight information and confidence information under each preset analysis target is used as the denominator. Thus, an adaptive normalization factor is constructed based on weight and confidence. Even if the confidence information of a certain preset analysis target is zero, the final recommendation score still remains in the range of [0, 1], which has cross-decision comparability. At the same time, the contribution of the data of the preset analysis target with low confidence information to the final score is automatically weakened, thereby effectively suppressing the influence of evaluation indicators extracted from unreliable analysis result summaries on the decision results.

[0101] In one alternative implementation, the confidence information is determined as follows: If the preset analysis target is not marked as downgraded, meaning its analysis result summary has passed verification, or a usable analysis result summary has been generated after re-analysis, the confidence level of the preset analysis target is determined based on its data time characteristics, data source characteristics, and category coverage characteristics. Specifically, the data time characteristic reflects the proximity of the data sampling time in the analysis result summary to the current time; the closer the sampling time is to the current time, the higher the confidence level. The data source characteristic reflects the number of data sources involved in the inference analysis of the preset analysis target; the richer the sources, the higher the confidence level. The category coverage characteristic reflects the proportion of candidate categories for which the preset analysis target can provide effective evaluation indicators to the candidate category set; the higher the coverage, the higher the confidence level. These three characteristics can be combined into a weighted summation to form the confidence level value.

[0102] If the preset analysis target is marked as the first degraded state, meaning that the summary of the analysis results for the preset analysis target comes from a cache hit, the confidence information of the preset analysis target is determined based on the cache duration of valid historical results. The shorter the cache duration, the higher the data freshness of the valid historical results, and the closer the confidence is to the normal value; conversely, the longer the cache duration, the lower the confidence.

[0103] If the preset analysis target is marked as the second downgraded state, meaning that the summary of the analysis results for the preset analysis target comes from category baseline template filling, the first preset confidence level is used as the confidence level information of the preset analysis target. The first preset confidence level is a fixed value lower than the normal confidence level, for example, it can be set to 0.5, to reflect that the summary of the analysis results is not a true analysis result for a specific candidate category, but still has some reference value.

[0104] If the preset analysis target is marked as the third downgraded state, meaning that the data for the preset analysis target is completely missing, the second preset reliability is used as the confidence level information for the preset analysis target. The second preset reliability is lower than the first preset reliability, and the second preset reliability can be set to zero to completely mask the impact of the preset analysis target on the final recommendation score, ensuring that the overall analysis task is not interrupted due to the lack of data for the preset analysis target.

[0105] Step S4043: Sort the candidate categories based on their recommendation scores to obtain a set of recommended categories.

[0106] In this embodiment of the invention, after calculating the recommendation score for each candidate category in the candidate category set, the candidate categories are sorted in descending order of recommendation score. Optionally, one or more candidate categories with the highest recommendation score can be selected as the recommended category set for the target product, or the complete list of sorted candidate categories can be output as the recommendation result, from which the business side can select according to actual needs.

[0107] In one optional implementation, when outputting the set of recommended product categories, a recommendation conclusion for each recommended product category is also output. The recommendation conclusion records the relevant reference fields of the recommended product category to supplement the explanation of the reasons for selecting the recommended product category. The relevant reference fields in the recommendation conclusion may include data fields from the analysis result summaries of each preset analysis target corresponding to the recommended product category, confidence information and consistency scores of the analysis result summaries of each preset analysis target, evaluation indicators of each preset analysis target, etc. The relevant reference fields are filled and integrated according to the preset conclusion template to transform into a structured recommendation conclusion, thereby forming a complete traceable link from input to conclusion, enabling business parties to verify and reproduce the recommendation basis.

[0108] The product data analysis method based on artificial intelligence provided in this invention extracts evaluation indicators for each candidate category across various dimensions from the structured analysis results summary of each analysis target. It then combines the confidence and weight information of each analysis target to perform a confidence-weighted fusion calculation of the recommendation score, thereby automatically weakening or shielding the impact of unreliable or missing data on the final decision. By generating a set of recommended categories sorted by recommendation score and attaching traceable evidence, the method ensures the quantification, consistency, and verifiability of the recommendation decision.

[0109] This embodiment also provides an artificial intelligence-based product data analysis device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0110] This embodiment provides a product data analysis device based on artificial intelligence, such as Figure 5 As shown, it includes: The first information construction module 501 is used to obtain the candidate category set and product demand information of the target product, and to construct shared inference information and local inference information of each preset analysis target based on the product demand information. The second information construction module 502 is used to construct reasoning context information for each preset analysis target based on references to shared reasoning information and local reasoning information. The information analysis module 503 is used to call the target analysis template corresponding to each preset analysis target from the preset analysis template library, analyze the reasoning context information of each preset analysis target, and obtain the analysis result summary of each preset analysis target. The category evaluation module 504 is used to evaluate each candidate category in the candidate category set based on the analysis result summary and evaluation information of each preset analysis objective, and to obtain the recommended category set of the target product.

[0111] In one optional implementation, the first information construction module 501 includes: The first information extraction unit is used to extract product demand information based on the shared features corresponding to each preset analysis target, and store the extracted data into a shared memory pool to obtain shared inference information. The second information extraction unit is used to extract product demand information based on the dimensional features corresponding to each preset analysis target, and obtain local inference information for each preset analysis target.

[0112] In one optional implementation, the second information construction module 502 includes: The first information processing unit is used to copy the reference handle of the shared reasoning information to the independent context corresponding to the preset analysis target for each preset analysis target, and to perform a local deep copy of the local reasoning information. The second information processing unit is used to assemble the reference handle of shared reasoning information and local reasoning information to obtain the reasoning context information of the preset analysis target.

[0113] In one optional implementation, the information analysis module 503 includes: The template matching unit is used to match each preset analysis target with a preset analysis template in the preset analysis template library based on preset matching rules, and to determine the target analysis template corresponding to each preset analysis target. The reasoning analysis unit is used to call the target analysis template corresponding to each preset analysis target, analyze the reasoning context information of each preset analysis target, and obtain the analysis result summary of each preset analysis target.

[0114] In one alternative implementation, the template matching unit includes: The priority determination subunit is used to prioritize the preset analysis target as the first priority, the product label coverage of the target product by the preset analysis template as the second priority, and the historical usage frequency of the preset analysis template as the third priority. The first priority is higher than the second priority, and the second priority is higher than the third priority. The template matching subunit is used to match the preset analysis templates based on the first priority, second priority, and third priority for each preset analysis target, and to determine the target analysis template corresponding to the preset analysis template.

[0115] In one optional implementation, the information analysis module 503 includes: The semaphore allocation unit is used to allocate a semaphore upper limit to the data processing interface corresponding to each preset analysis target based on the interface type of the data processing interface corresponding to each preset analysis target. The reasoning and analysis unit is used to call the data processing interfaces corresponding to each preset analysis target in parallel, load the target analysis template, reason about the reasoning context information, and obtain the analysis result summary of each preset analysis target.

[0116] In one alternative implementation, the category evaluation module 504 includes: The data extraction unit is used to extract data from the analysis result summaries of each preset analysis target based on each candidate product category, and to obtain the evaluation indicators of each candidate product category corresponding to each preset analysis target. The indicator evaluation unit is used to determine the recommended score for each candidate category based on the evaluation indicators corresponding to each preset analysis target and the evaluation information of each preset analysis target. The scoring and sorting unit is used to sort the candidate categories based on their recommendation scores to obtain a set of recommended categories.

[0117] In one alternative embodiment, the device further includes: The abstract verification unit is used to verify the analysis result abstracts of each preset analysis target. The degradation processing unit is used to regenerate the analysis result summary of the first analysis target based on a preset degradation processing method if the analysis result summary of the first analysis target fails the verification.

[0118] In one optional implementation, the degradation processing unit includes: The first downgrade processing subunit is used to re-analyze the reasoning context information of the first analysis target if the number of analysis times corresponding to the first analysis target is less than the preset number of analysis times, and obtain a summary of the analysis results of the first analysis target. The second downgrade processing subunit is used to query the valid historical results of the first analysis target based on the fingerprint information of the shared reasoning information if the number of analysis times corresponding to the first analysis target reaches the preset number of analysis times, or if the analysis of the reasoning context information of the first analysis target does not yield an analysis result summary of the first analysis target, and uses the valid historical results as the analysis structure summary of the first analysis target, and marks the first analysis target as the first downgrade state. The third downgrade processing subunit is used to obtain the category baseline template corresponding to the first analysis target from the preset analysis template library based on the category tag in the shared reasoning information if no valid historical results of the first analysis target are found based on the fingerprint information of the shared reasoning information, and fill the analysis result summary of the first analysis target based on the category baseline template, and mark the first analysis target as the second downgrade state. The fourth downgrade processing subunit is used to set the data field of the analysis result summary of the first analysis target to empty and mark the first analysis target as the third downgrade state if the category baseline template corresponding to the first analysis target does not exist in the preset analysis template library.

[0119] In one optional implementation, the evaluation information of the preset analysis target includes confidence information and weight information, wherein the device further includes: The first confidence level determination module is used to determine the confidence level information of the preset analysis target based on the data time characteristics, data source characteristics, and category coverage characteristics of the analysis result summary of the preset analysis target if the preset analysis target is not marked as downgraded. The second confidence determination module is used to determine the confidence information of the preset analysis target based on the caching duration of valid historical results if the preset analysis target is marked as the first degraded state. The third confidence level determination module is used to use the first preset confidence level as the confidence level information of the preset analysis target if the preset analysis target is marked as the second degraded state. The fourth confidence level determination module is used to use the second preset confidence level as the confidence level information of the preset analysis target if the preset analysis target is marked as the third degraded state, wherein the second preset confidence level is less than the first preset confidence level.

[0120] In one alternative embodiment, the device further includes: The template status determination module is used to determine that the target analysis template corresponding to the first analysis target has not been matched if the analysis result summary verification of the first analysis target fails. The usage update module is used to determine if the analysis result summary of the second analysis target passes the verification, and to update the historical usage count of the target analysis template corresponding to the second analysis target.

[0121] In one alternative embodiment, the device further includes: The recommended template determination module is used to select the first analysis template as the recommended template if the number of historical uses of the first analysis template in the preset analysis template library is greater than the first preset number of uses. The alternative status determination module is used to select the second analysis template as an alternative template if the second analysis template in the preset analysis template library has not been matched for a second preset number of consecutive times.

[0122] In one alternative embodiment, the device further includes: The evaluation rule acquisition module is used to acquire a consistency evaluation rule library, which includes multiple consistency evaluation rules, each corresponding to a different rule priority. The consistency assessment module is used to evaluate the consistency between the analysis result summaries of each preset analysis target according to the priority of the rules and based on multiple consistency assessment rules, so as to obtain a consistency score.

[0123] The AI-based product data analysis device provided in this embodiment of the invention can execute the AI-based product data analysis method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0124] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0125] The following is a detailed reference. Figure 6This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0126] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0127] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory 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 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the artificial intelligence-based product data analysis method of the embodiments of the present invention.

[0128] Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0129] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the artificial intelligence-based product data analysis method shown in the above embodiments is implemented.

[0130] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0131] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A product data analysis method based on artificial intelligence, characterized in that, The method includes: Obtain a set of candidate product categories and product demand information for the target product, and based on the product demand information, construct shared inference information and local inference information for each preset analysis target; Based on the reference to the shared reasoning information and the local reasoning information, reasoning context information for each of the preset analysis targets is constructed; The target analysis templates corresponding to each of the preset analysis targets are called from the preset analysis template library, and the reasoning context information of each of the preset analysis targets is analyzed to obtain a summary of the analysis results of each of the preset analysis targets. Based on the analysis results summary and evaluation information of each of the preset analysis objectives, each candidate category in the candidate category set is evaluated to obtain the recommended category set of the target product.

2. The method according to claim 1, characterized in that, The construction of shared inference information and local inference information for each preset analysis target based on the product demand information includes: Based on the shared features corresponding to each of the preset analysis targets, the product demand information is extracted, and the extracted data is stored in a shared memory pool to obtain the shared inference information; Based on the dimensional features corresponding to each of the preset analysis targets, the product demand information is extracted to obtain the local inference information for each of the preset analysis targets.

3. The method according to claim 1, characterized in that, The construction of reasoning context information for each of the preset analysis targets based on references to the shared reasoning information and the local reasoning information includes: For each of the preset analysis targets, the reference handle of the shared inference information is copied to the independent context corresponding to the preset analysis target, and the local inference information is performed as a local deep copy. The reference handle of the shared reasoning information and the local reasoning information are assembled to obtain the reasoning context information of the preset analysis target.

4. The method according to claim 1, characterized in that, The step involves calling the target analysis template corresponding to each preset analysis target from the preset analysis template library, analyzing the inference context information of each preset analysis target, and obtaining a summary of the analysis results for each preset analysis target, including: Based on preset matching rules, each preset analysis target is matched with a preset analysis template in the preset analysis template library to determine the target analysis template corresponding to each preset analysis target. Each preset analysis target is called with its corresponding target analysis template to analyze the reasoning context information of each preset analysis target and obtain an analysis result summary of each preset analysis target.

5. The method according to claim 4, characterized in that, The step of matching each preset analysis target with a preset analysis template in the preset analysis template library based on preset matching rules, and determining the analysis template corresponding to each preset analysis target, includes: The preset analysis target is given as the first priority, the product label coverage of the preset analysis template for the target product is given as the second priority, and the historical usage count of the preset analysis template is given as the third priority, wherein the first priority is higher than the second priority, and the second priority is higher than the third priority. For each of the preset analysis targets, the preset analysis templates are matched based on the first priority, the second priority, and the third priority to determine the target analysis template corresponding to the preset analysis template.

6. The method according to claim 4, characterized in that, The step involves calling the target analysis template corresponding to each of the preset analysis targets respectively, analyzing the reasoning context information of each preset analysis target, and obtaining a summary of the analysis results for each preset analysis target, including: Based on the interface type of the data processing interface corresponding to each of the preset analysis targets, a semaphore upper limit is allocated to the data processing interface corresponding to each of the preset analysis targets; The data processing interfaces corresponding to each of the preset analysis targets are called in parallel, and the target analysis template is loaded to reason about the reasoning context information to obtain the analysis result summary of each of the preset analysis targets.

7. The method according to claim 1, characterized in that, The evaluation of each candidate category in the candidate category set based on the analysis result summary and evaluation information of each of the preset analysis objectives yields a recommended category set for the target product, including: Based on each of the candidate categories, data is extracted from the analysis result summaries of each of the preset analysis targets to obtain the evaluation indicators corresponding to each of the candidate categories and each of the preset analysis targets. Based on the evaluation indicators of each candidate category corresponding to each preset analysis target and the evaluation information of each preset analysis target, the recommended score of each candidate category is determined. The recommended category set is obtained by sorting the recommended scores of each candidate category.

8. The method according to claim 1, characterized in that, The method further includes: The analysis result summaries for each of the preset analysis targets are verified respectively; If the analysis result summary of the first analysis target fails the verification, the analysis result summary of the first analysis target is regenerated based on the preset downgrade processing method.

9. The method according to claim 8, characterized in that, The step of regenerating the analysis result summary of the first analysis target based on the preset degradation processing method includes: If the number of analyses corresponding to the first analysis target is less than the preset number of analyses, the reasoning context information of the first analysis target is re-analyzed to obtain a summary of the analysis results of the first analysis target; If the number of analyses corresponding to the first analysis target reaches the preset number of analyses, or if the analysis of the reasoning context information of the first analysis target does not yield an analysis result summary of the first analysis target, the valid historical results of the first analysis target are queried based on the fingerprint information of the shared reasoning information, and the valid historical results are used as the analysis structure summary of the first analysis target, and the first analysis target is marked as the first degraded state. If no valid historical results for the first analysis target are found based on the fingerprint information of the shared inference information, the category baseline template corresponding to the first analysis target is obtained from the preset analysis template library based on the category tag in the shared inference information, and the analysis result summary of the first analysis target is filled in based on the category baseline template, and the first analysis target is marked as the second downgraded state. If the category baseline template corresponding to the first analysis target does not exist in the preset analysis template library, the data field of the analysis result summary of the first analysis target is set to empty, and the first analysis target is marked as the third downgraded state.

10. The method according to claim 9, characterized in that, The evaluation information for the preset analysis target includes confidence information and weight information, wherein the confidence information is determined in the following manner: If the preset analysis target is not marked as downgraded, the confidence information of the preset analysis target is determined based on the data time characteristics, data source characteristics, and category coverage characteristics of the analysis result summary of the preset analysis target. If the preset analysis target is marked as the first degraded state, the confidence information of the preset analysis target is determined based on the caching duration of the valid historical results; If the preset analysis target is marked as the second degraded state, the first preset confidence level is used as the confidence level information of the preset analysis target; If the preset analysis target is marked as the third degraded state, the second preset confidence level is used as the confidence level information of the preset analysis target, wherein the second preset confidence level is less than the first preset confidence level.

11. The method according to claim 8, characterized in that, The method further includes: If the analysis result summary of the first analysis target fails the verification, it is determined that the target analysis template corresponding to the first analysis target is not matched; If the analysis result summary of the second analysis target passes the verification, it is determined that the target analysis template corresponding to the second analysis target has been hit, and the historical usage count of the target analysis template corresponding to the second analysis target is updated.

12. The method according to claim 11, characterized in that, The method further includes: If the number of historical uses of a first analysis template in the preset analysis template library is greater than a first preset number, the first analysis template will be used as a recommended template. If a second analysis template in the preset analysis template library fails to be matched for a second preset number of consecutive times, the second analysis template will be used as a candidate template.

13. The method according to claim 1, characterized in that, The method further includes: Obtain a consistency assessment rule base, wherein the consistency assessment rule base includes multiple consistency assessment rules, and the multiple consistency assessment rules correspond to different rule priorities; According to the priority of the rules, the consistency between the analysis result summaries of each preset analysis target is evaluated in turn based on multiple consistency evaluation rules to obtain a consistency score.

14. A product data analysis device based on artificial intelligence, characterized in that, The device includes: The first information construction module is used to obtain a set of candidate categories and product demand information for the target product, and to construct shared inference information and local inference information for each preset analysis target based on the product demand information. The second information construction module is used to construct reasoning context information for each of the preset analysis targets based on references to the shared reasoning information and the local reasoning information; The information analysis module is used to call the target analysis template corresponding to each preset analysis target from the preset analysis template library, analyze the reasoning context information of each preset analysis target, and obtain the analysis result summary of each preset analysis target. The category evaluation module is used to evaluate each candidate category in the candidate category set based on the analysis result summary and evaluation information of each of the preset analysis targets, so as to obtain the recommended category set of the target product.

15. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the product data analysis method based on artificial intelligence as described in any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the artificial intelligence-based product data analysis method according to any one of claims 1 to 13.

Images