Guidance information generation method and device

Abstract

The embodiment of the invention provides a guidance information generation method and device, and the method comprises the steps: obtaining multi-modal application data in a target application scene, splitting the multi-modal application data into a plurality of data blocks according to preset splitting rule information, and iteratively instructing the multi-modal large model to perform data analysis on the data block to be processed at this time based on the processing rule information obtained at each time to obtain a corresponding analysis result, and merging based on the analysis result to obtain the data block to be processed at the next time until only one data block to be processed exists and a corresponding target analysis result is obtained. And generating visual guidance information suitable for the target application scene based on the target analysis result. Wherein the multi-modal application data comprises user behavior data or supply resource data, and the processing rule information indicates that the multi-modal large model extracts user behavior characteristics from the to-be-processed data block and analyzes user behavior preferences, or extracts supply resource attributes and analyzes the adaptation degree of the corresponding supply resource data and the target application scene.

Description

TECHNICAL FIELD

[0001] One or more embodiments of the present specification relate to the technical field of artificial intelligence, and in particular to a guidance information generation method and device. BACKGROUND

[0002] With the development of artificial intelligence (AI) technology, large models (LLM) have been integrated into various fields. Based on the deep understanding ability of large models and the massive world knowledge and professional domain knowledge integrated therein, the large models have obvious advantages in data analysis and information mining.

[0003] At present, most large models are trained based on the Transformer framework. However, due to the context window constraint of the Transformer framework, the data analysis and information mining effect of the large models is not ideal for long sequence input, and it is difficult to meet many large-scale data application scenarios. SUMMARY

[0004] To improve the analysis ability of large models on large-scale data, the embodiments of the present specification provide a guidance information generation method and device.

[0005] In a first aspect, one or more embodiments of the present specification provide a guidance information generation method, comprising: obtaining multi-modal application data under a target application scenario, the multi-modal application data comprising user behavior data or supply resource data; splitting the multi-modal application data into a plurality of first data blocks according to split rule information preset for the target application scenario; iteratively performing a target processing operation based on the plurality of first data blocks until only one data block to be processed remains, and obtaining a target analysis result of the data block; generating visual guidance information applicable to the target application scenario based on the target analysis result, to provide guidance services for the management and operation of user behavior or supply resources; wherein the target processing operation comprises instructing a multi-modal large model to perform data analysis on the data block to be processed this time based on processing rule information obtained this time to obtain a corresponding analysis result, and merging the analysis result to obtain a data block to be processed next time, the first data block to be processed being the first data block; the processing rule information is used to instruct the multi-modal large model to extract user behavior features from each data block to be processed and analyze user behavior preferences based on the user behavior features, or extract supply resource attributes from each data block to be processed and analyze the adaptation degree of the corresponding supply resource data to the target application scenario based on the supply resource attributes.

[0006] In an optional embodiment, the multi-modal application data in the target application scenario is acquired, including: acquiring offline application data and online application data in the target application scenario, the offline application data and the online application data being multi-modal big data; performing a preprocessing operation on the offline application data and the online application data respectively to obtain multi-modal application data to be analyzed.

[0007] In an optional embodiment, the target processing operation is performed at any time, including: acquiring first processing rule information adapted to the target application scenario; inputting the plurality of data blocks to be processed this time and the first processing rule information into the multi-modal big model, instructing the multi-modal big model to perform data analysis on the plurality of data blocks to be processed this time in parallel to obtain a plurality of first analysis results; and performing partial merging on the plurality of first analysis results to obtain data blocks to be processed next time.

[0008] In an optional embodiment, the first processing rule information adapted to the target application scenario is acquired, including: based on a plurality of prompt word templates and a plurality of indication information templates preset for the target application scenario, acquiring a target prompt word template and a target indication information template as the first processing rule information adapted to the target application scenario; wherein different prompt word templates are used to instruct the multi-modal big model to perform data analysis on each data block to be processed in different analysis stages, and different indication information templates are used to instruct the multi-modal big model to perform data analysis on each data block to be processed in different analysis directions.

[0009] In an optional embodiment, the plurality of data blocks to be processed this time and the first processing rule information are input into the multi-modal big model to instruct the multi-modal big model to perform data analysis on the plurality of data blocks to be processed this time in parallel, including: inputting the plurality of data blocks to be processed this time and the first processing rule information as input data into a plurality of processing units respectively; and instructing the plurality of processing units to schedule the multi-modal big model to perform data analysis on the acquired data blocks to be processed according to the acquired first processing rule information in parallel.

[0010] In an optional embodiment, the plurality of first analysis results are partially merged to obtain data blocks to be processed next time, including: acquiring a plurality of verification rule information preset for the target application scenario; verifying the credibility of each first analysis result according to the plurality of verification rule information; and partially merging a plurality of first analysis results whose credibility meets a preset requirement to obtain data blocks to be processed next time.

[0011] In an optional embodiment, further comprising: determining at least one second data block corresponding to at least one first analysis result that does not meet the preset requirement; obtaining second processing rule information adapted to the target application scenario; and re-instructing the multi-modal large model to perform data analysis on the at least one second data block based on the at least one second data block and the second processing rule information until the credibility of the analysis result meets the preset requirement.

[0012] In an optional embodiment, re-instructing the multi-modal large model to perform data analysis on the at least one second data block based on the at least one second data block and the second processing rule information comprises: re-splitting the at least one second data block to obtain a plurality of third data blocks; and re-instructing the multi-modal large model to perform data analysis on the plurality of third data blocks in parallel based on the plurality of third data blocks and the second processing rule information.

[0013] In an optional embodiment, further comprising: storing a plurality of analysis results obtained by the multi-modal large model each time in a preset storage space.

[0014] In an optional embodiment, generating visual guidance information applicable to the target application scenario according to the target analysis result comprises: obtaining a target analysis result from the preset storage space; and generating visual guidance information applicable to the target application scenario according to the target analysis result.

[0015] In an optional embodiment, performing a target processing operation at any time comprises: obtaining third processing rule information adapted to the target application scenario; inputting a data block to be processed this time and the third processing rule information into the multi-modal large model to instruct the multi-modal large model to perform data analysis on the received data block to obtain a corresponding second analysis result; and merging the second analysis result with one first data block that has not been processed as a data block to be processed next time; wherein the first data block to be processed is one of the plurality of first data blocks.

[0016] In an optional embodiment, further comprising: displaying the visual guidance information in at least one visual manner; or sending the visual guidance information to a provider that provides the supply resource data.

[0017] In a second aspect, the specification one or more embodiments provide a guidance information generation apparatus, comprising: an acquisition module configured to acquire multi-modal application data under a target application scenario, the multi-modal application data comprising user behavior data or supply resource data; a splitting module configured to split the multi-modal application data into a plurality of first data blocks according to splitting rule information preset for the target application scenario; an execution module configured to iteratively execute a target processing operation based on the plurality of first data blocks until only one data block remains to be processed, and obtain a target analysis result of the data block; and a generation module configured to generate visual guidance information applicable to the target application scenario based on the target analysis result, to provide guidance services for management and operation of user behavior or supply resources; wherein the target processing operation comprises instructing a multi-modal large model to perform data analysis on the data block to be processed this time based on processing rule information acquired this time to obtain a corresponding analysis result, and merging the analysis result to obtain a data block to be processed next time, the first data block to be processed being the first data block; and the processing rule information is used to instruct the multi-modal large model to extract user behavior features from each data block to be processed and analyze user behavior preferences based on the user behavior features, or extract supply resource attributes from each data block to be processed and analyze the degree of adaptation of corresponding supply resource data to the target application scenario based on the supply resource attributes.

[0018] In an optional embodiment, the acquisition module acquires multi-modal application data under a target application scenario, and is configured to: acquire offline application data and online application data under the target application scenario, the offline application data and the online application data being multi-modal big data; and perform a preprocessing operation on the offline application data and the online application data respectively to obtain multi-modal application data to be analyzed.

[0019] In an optional embodiment, the acquisition module is further configured to acquire first processing rule information adapted to the target application scenario; and the execution module, when executing the target processing operation at any time, is configured to: input the plurality of data blocks to be processed this time and the first processing rule information into the multi-modal large model, instruct the multi-modal large model to perform data analysis on the plurality of data blocks to be processed this time in parallel, and obtain a plurality of first analysis results; and perform partial merging on the plurality of first analysis results to obtain a data block to be processed next time.

[0020] In an optional embodiment, the obtaining module obtains first processing rule information adapted to the target application scenario, for: based on a plurality of prompt word templates and a plurality of indication information templates preset for the target application scenario, obtaining a target prompt word template and a target indication information template as the first processing rule information adapted to the target application scenario; wherein different prompt word templates are used to instruct the multi-modal large model to perform data analysis on each data block to be processed in different analysis stages, and different indication information templates are used to instruct the multi-modal large model to perform data analysis on each data block to be processed in different analysis directions.

[0021] In an optional embodiment, the execution module inputs the plurality of data blocks to be processed this time and the first processing rule information into the multi-modal large model, and instructs the multi-modal large model to perform data analysis on the plurality of data blocks to be processed this time in parallel, for: sending the plurality of data blocks to be processed this time and the first processing rule information as input data to a plurality of processing units respectively; and instructing the plurality of processing units to schedule the multi-modal large model to perform data analysis on the obtained data blocks to be processed according to the obtained first processing rule information in parallel.

[0022] In an optional embodiment, the execution module performs partial merging on the plurality of first analysis results to obtain data blocks to be processed next time, for: obtaining a plurality of verification rule information preset for the target application scenario; verifying the credibility of each first analysis result according to the plurality of verification rule information; and performing partial merging on a plurality of first analysis results whose credibility meets preset requirements to obtain data blocks to be processed next time.

[0023] In an optional embodiment, the execution module is further configured to: determine at least one second data block corresponding to at least one first analysis result whose credibility does not meet preset requirements; obtain second processing rule information adapted to the target application scenario; and based on the at least one second data block and the second processing rule information, re-instruct the multi-modal large model to perform data analysis on the at least one second data block until the credibility of the analysis result meets preset requirements.

[0024] In an optional embodiment, the execution module re-instructs the multi-modal large model to perform data analysis on the at least one second data block based on the at least one second data block and the second processing rule information, for: re-splitting the at least one second data block to obtain a plurality of third data blocks; and re-instructing the multi-modal large model to perform data analysis on the plurality of third data blocks in parallel based on the plurality of third data blocks and the second processing rule information.

[0025] In an optional embodiment, the execution module is further configured to store the multiple analysis results obtained by the multi-modal large model each time in a preset storage space.

[0026] In an optional embodiment, the generation module is configured to generate the visual guidance information applicable to the target application scenario according to the target analysis result, and the generation of the visual guidance information comprises: obtaining the target analysis result from the preset storage space; and generating the visual guidance information applicable to the target application scenario according to the target analysis result.

[0027] In an optional embodiment, the obtaining module is further configured to obtain third processing rule information adapted to the target application scenario, and the execution module is configured to execute the target processing operation any time, and the execution of the target processing operation comprises: inputting the data block to be processed this time and the third processing rule information into the multi-modal large model, instructing the multi-modal large model to perform data analysis on the received data block to obtain a corresponding second analysis result; and merging the second analysis result and one first data block that has not been processed as a data block to be processed next time; wherein the first data block to be processed is one of the multiple first data blocks.

[0028] In an optional embodiment, the execution module is further configured to display the visual guidance information in at least one visual manner, or send the visual guidance information to a provider providing the supply resource data.

[0029] In a third aspect, one or more embodiments of the present specification provide an electronic device, the electronic device comprising: a memory configured to store a computer program product; and a processor configured to execute the computer program product stored in the memory, and the computer program product, when executed, is configured to implement the guidance information generation method of the first aspect.

[0030] In a fourth aspect, one or more embodiments of the present specification provide a computer readable storage medium, and the computer readable storage medium stores computer program instructions, and the computer program instructions, when executed, are configured to implement the guidance information generation method of the first aspect.

[0031] In summary, the guidance information generation method provided by one or more embodiments of the present specification can perform data analysis on multiple data blocks in parallel based on the iteration of the multi-modal large model. Moreover, when the multi-modal large model performs data analysis on each data block each time, the key features meeting the requirements can be extracted from each data block. Since these key features are more explicit than the data features of the original data blocks, the data amount of the data block to be processed next time obtained by merging part of the analysis results is smaller.

[0032] Therefore, through continuous iteration processing, not only the length of the sequence input to the multi-modal large model each time can be continuously reduced, but also the analysis direction of the multi-modal large model can be continuously focused, the accuracy of data analysis is improved, and especially for large-scale data application scenarios, better data analysis effect can be achieved. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of one or more embodiments of the present specification, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present specification, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0034] Figure 1 A flowchart of a guidance information generation method provided by one or more embodiments of the present specification.

[0035] Figure 2 A schematic diagram of a digital service platform performing a target processing operation process provided by one or more embodiments of the present specification.

[0036] Figure 3 A structural block diagram of a digital service platform provided by one or more embodiments of the present specification.

[0037] Figure 4 Another schematic diagram of a digital service platform performing a target processing operation process provided by one or more embodiments of the present specification.

[0038] Figure 5 A structural block diagram of a guidance information generation device provided by one or more embodiments of the present specification.

[0039] Figure 6 A structural schematic diagram of an electronic device provided by one or more embodiments of the present specification. DETAILED DESCRIPTION

[0040] The present specification will be further described in detail through the drawings and embodiments. Through these descriptions, the characteristics and advantages of the embodiments of the present specification will become clearer and more apparent.

[0041] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations. Unless specifically indicated otherwise, the drawings shown in the Figures are not necessarily to scale.

[0042] In addition, the technical features involved in different embodiments of the embodiments of the present specification described below can be combined with each other as long as they do not conflict with each other.

[0043] With the development of artificial intelligence (AI) technology, large models (LLM) have been integrated into various fields. Based on the deep understanding ability of large models and the massive world knowledge and professional field knowledge integrated, large models have obvious advantages in data analysis and knowledge mining. Traditional large models perform well in processing text data, but for complex types of data, it is difficult to achieve the expected processing effect. Therefore, in order to meet more diversified data processing needs, multimodal large models (MLLMs) have emerged. In terms of data processing, multimodal large models not only can analyze and mine knowledge from single modal text data, but also can achieve the same processing function for image, image-text, audio, video and other types of data, even for complex data of multiple types, and can meet the diversified data processing needs.

[0044] When selecting a model framework, the Transformer has become the core framework for processing cross-modal data due to its self-attention mechanism and parallel computing capability. Therefore, in practical applications, most multimodal large models are trained based on the Transformer framework. However, due to the context window constraint of the Transformer framework, the input data supported by the multimodal large model is limited. For the case of inputting long sequences, the processing effect of the multimodal large model is not ideal, especially for large-scale data application scenarios, it is difficult to achieve the expected processing effect.

[0045] Therefore, one or more embodiments of the present specification provide a guidance information generation method applied to a digital service platform. The method can split large-scale multimodal application data in a target application scenario into multiple data blocks, take the multiple data blocks as data blocks to be processed, and iteratively perform data analysis on each data block based on a multimodal large model through multiple execution paths. Moreover, when inputting the data block to be processed to the multimodal large model each time, the analysis results obtained by the previous partial execution path are merged and taken as the data block to be processed this time. In this way, the sequence length input to the multimodal large model is continuously reduced, the data processing difficulty of the multimodal large model is reduced, and the accuracy of data analysis is improved.

[0046] Before the method of generating guidance information provided by one or more embodiments of the present specification is described, first, the digital service platform to which the method is applied, the target application scenario, and the multi-modal application data in the target application scenario are explained.

[0047] In one or more embodiments of the present specification, the digital service platform refers to an intermediate platform that serves as a medium between users and third parties and provides various digital services for users and third parties, including but not limited to payment settlement platforms, e-commerce transaction platforms, financial technology platforms, credit intermediary platforms, and other emerging platforms. Accordingly, the target application scenario can be understood as any application scenario in the digital service platform, for example, retail consumption scenarios, cross-border trade scenarios, bulk transaction scenarios, supply chain collaboration scenarios, industry risk control scenarios, small and micro enterprise financing scenarios, transaction security scenarios, virtual asset transaction scenarios, etc. Depending on the type of digital service platform, the application scenario will also be different, which will not be described in detail here.

[0048] In one or more embodiments of the present specification, the digital service platform is used to display various supply resources, wherein the supply resources can be provided directly by the digital service platform or by a provider cooperating with the digital service platform. Optionally, in the following embodiments, the supply resources provided by a third party are taken as an example for description.

[0049] In one or more embodiments of the present specification, the specific content of the supply resources is not limited, and optionally, the supply resources include but are not limited to goods, services, activities, and information in various application scenarios. Based on this, the data generated by a series of interactive operations during the process of users purchasing various goods, using various services, participating in various activities, and viewing various information in the digital service platform is referred to as user behavior data. Accordingly, the attributes related to the supply resources are referred to as supply resource data, for example, including but not limited to the type, name, and origin of goods, the type, content, and service object of services, the type, content, and activity time of activities, etc.

[0050] In one or more embodiments of the present specification, the specific content of the multi-modal application data in the target application scenario is not limited, and optionally, in the following embodiments, the multi-modal application data including the user behavior data or the supply resource data in the target application scenario is taken as an example for description.

[0051] For ease of illustration, in one or more embodiments of the present specification, a commodity is taken as an example of a supply resource, and accordingly, the target application scenario refers to a retail consumption scenario, and the user behavior data generated in the application scenario refers to a series of interactive operations of users in the digital service platform, such as viewing, adding to a shopping cart, and payment settlement, etc. for various commodities, for example, including but not limited to browsing records, consultation records, shopping cart lists, transaction records, order records, etc.

[0052] Based on the above, the guidance information generation method provided by one or more embodiments of the present specification is described below in combination with the accompanying drawings.

[0053] Figure 1 A flowchart of a guidance information generation method provided by one or more embodiments of the present specification is shown in FIG. 1. Figure 1 As shown in FIG. 1, the method comprises:

[0054] S102, acquiring multi-modal application data in a target application scenario, the multi-modal application data comprising user behavior data or supply resource data.

[0055] S104, according to the split rule information preset for the target application scenario, splitting the multi-modal application data into a plurality of first data blocks.

[0056] S106, based on the plurality of first data blocks, iteratively performing a target processing operation until there is only one data block to be processed, and obtaining a target analysis result of the data block.

[0057] S108, according to the target analysis result, generating visual guidance information applicable to the target application scenario to provide guidance services for the management and operation of user behavior or supply resources.

[0058] In one or more embodiments of the present specification, the approach of the digital service platform to acquire multi-modal application data in the target application scenario is not limited. Alternatively, the digital service platform can acquire offline application data in the target application scenario, or can acquire online application data in the target application scenario, wherein the offline application data and the online application data are both multi-modal big data. That is, the digital service platform can generate visual guidance information offline, or can generate visual guidance information online.

[0059] Based on this, in order to carry out subsequent processing, after obtaining the offline application data and the online application data under the target application scenario, the digital service platform can first perform preprocessing operations on the offline application data and the online application data respectively to obtain the to-be-processed multi-modal application data. The specific type of the preprocessing operation is not limited, and can optionally include, but is not limited to, format compliance detection, malicious code scanning, redundant data removal, noise filtering, etc. Through the above preprocessing operations, compliant, effective and standardized multi-modal application data can be obtained.

[0060] Further optionally, the specific manner in which the digital service platform obtains the offline application data and the online application data under the target application scenario is not limited in the one or more embodiments of the present specification. Optionally, the offline application data and the online application data under the target application scenario can be reported to the digital service platform by each provider, or can be actively obtained by the digital service platform. For example, the digital service platform can obtain the offline application data and the online application data under the target application scenario from each provider through Web Crawler technology. Of course, the above is only an exemplary description, and in actual application, the specific acquisition manner is not limited thereto, and according to different actual requirements, a suitable acquisition manner can be flexibly selected, which will not be described in detail herein.

[0061] Based on this, after obtaining the multi-modal application data under the target application scenario, the digital service platform can split the multi-modal application data into multiple data blocks according to the preset splitting rule information of the target application scenario, as the first to-be-processed multiple data blocks. In order to distinguish, in the one or more embodiments of the present specification, the first to-be-processed multiple data blocks are referred to as multiple first data blocks. The specific content of the splitting rule information is not limited, and according to the actual situation of the multi-modal application data, the splitting rule information can also be different.

[0062] Optionally, the content of the splitting rule information can be to indicate equal splitting of the same type of data, and equal splitting after grouping by type for different types of data. For example, the digital service platform obtains 10 million user browsing records of a commodity, and the content of the splitting rule information can be to indicate that the 10 million browsing records are split into 10 data blocks, and each data block includes 1 million browsing records. For another example, the digital service platform obtains 10 million multi-modal application data, including 7 million browsing records and 3 million order records, and the content of the splitting rule information can be to indicate that the 7 million browsing records are grouped as one group of data, and the 3 million order records are grouped as another group of data; further, the 7 million browsing records are split into 7 data blocks, and each data block includes 1 million browsing records; and the 3 million order records are split into 3 data blocks, and each data block includes 1 million order records.

[0063] It should be noted that the above is only an example of explanation, and each example is intended to illustrate the principle of data splitting, and is not a limiting explanation. In actual application, the specific content of the splitting rule information can be flexibly set according to the type of the target application scenario, the type of the multi-modal application data, the needs of the provider, and other factors, which will not be described in detail here.

[0064] In one or more embodiments of the present specification, the target processing operation performed by the digital service platform each time includes data analysis by the multi-modal large model on the data block to be processed this time based on the processing rule information obtained this time to obtain the corresponding analysis result, and the next data block to be processed is obtained by merging based on the analysis result. Wherein, the processing rule information is used to instruct the multi-modal large model to extract user behavior features from each data block to be processed and analyze user behavior preferences based on the user behavior features, or extract supply resource attributes from each data block to be processed and analyze the adaptation degree of the corresponding supply resource data to the target application scenario. That is, when the multi-modal large model performs data analysis on the received data block each time, the multi-modal large model extracts the required key features from the corresponding data block according to the processing rule information as the corresponding analysis result.

[0065] Based on this, when the digital service platform performs the target processing operation for the first time, the multi-modal large model can be instructed to perform data analysis on the plurality of first data blocks respectively based on the obtained processing rule information to obtain the corresponding analysis result, and then the plurality of data blocks to be processed next are obtained by partially merging the plurality of analysis results. In this way, the above target processing operation is repeatedly performed until there is only one data block to be processed, and the multi-modal large model performs data analysis on the data block to obtain the corresponding target analysis result, which is used to generate the visual guidance information applicable to the target application scenario.

[0066] In one or more embodiments of the present specification, when the digital service platform performs the target processing operation at any time, the digital service platform can first acquire the first processing rule information adapted to the target application scenario, and input the plurality of data blocks to be processed this time and the first processing rule information into the multi-modal large model, instruct the multi-modal large model to perform data analysis on the plurality of data blocks to be processed this time in parallel, obtain a plurality of first analysis results, and then perform partial merging on the plurality of first analysis results to obtain data blocks to be processed next time.

[0067] That is, the digital service platform performs data analysis on different data blocks in parallel based on the multi-modal large model in each execution of the target processing operation, obtains analysis results corresponding to each execution path, and then merges the analysis results obtained by part of the execution paths to obtain a plurality of new data blocks as data blocks to be processed next time.

[0068] In one or more embodiments of the present specification, the specific manner in which the digital service platform performs partial merging on the plurality of first analysis results is not limited. Alternatively, the digital service platform can perform partial merging on the plurality of first analysis results after acquiring the plurality of first analysis results corresponding to all data blocks to be processed this time, or can perform merging on at least two first analysis results after acquiring the first analysis results corresponding to at least two data blocks, and the specific manner can be determined according to actual needs.

[0069] Figure 2 A schematic diagram of a process in which a digital service platform performs a target processing operation is provided for one or more embodiments of the present specification.

[0070] As shown in Figure 2 In this example, the digital service platform performs two-by-two merging on the plurality of first analysis results in each execution of the target processing operation process. Alternatively, in the following embodiments, the processing manner shown in Figure 2 is taken as an example for illustration.

[0071] In one or more embodiments of the present specification, the specific content of the processing rule information is not limited. Alternatively, the processing rule information can include a plurality of prompt word templates and a plurality of indication information templates preset for the target application scenario. Based on this, when the digital service platform acquires the first processing rule information adapted to the target application scenario, the digital service platform can acquire a target prompt word template and a target indication information template based on the plurality of prompt word templates and the plurality of indication information templates preset for the target application scenario as the first processing rule information adapted to the target application scenario.

[0072] Optionally, different prompt word templates are used to instruct the multi-modal large model to perform data analysis on each data block to be processed at different analysis stages, and different instruction information templates are used to instruct the multi-modal large model to perform data analysis on each data block to be processed in different analysis directions. For example, the first prompt word template used by the digital service platform when performing the target processing operation for the first time is different from the second prompt word template used when performing the target processing operation for the second time, so as to instruct the multi-modal large model to analyze each data block from different dimensions. Correspondingly, different instruction information templates can also be used to instruct the multi-modal large model to analyze each data block in the specified analysis direction each time the multi-modal model is instructed to perform data analysis on each data block.

[0073] The process of instructing the multi-modal large model to perform data analysis on multiple data blocks according to the processing rule information by the digital service platform will be exemplarily described below through specific examples.

[0074] Suppose that the digital service platform obtains 10 million pieces of user behavior data in a retail consumption scenario, and splits them into 10 first data blocks according to the splitting rule information, wherein each first data block includes 1 million pieces of user behavior data. Further, the digital service platform obtains a kind of processing rule information when performing the target processing operation for the first time, so as to instruct the multi-modal large model to perform data analysis on multiple first data blocks. Suppose that the processing rule information includes a first prompt word template, and the content of the first prompt word template is "perform data analysis on each data block from a global dimension, and identify the active time period of user purchase of goods, the proportion of active users, and the high-sales commodity type from each data block". Based on this, the digital service platform can input the first prompt word template and each first data block as input data into the multi-modal large model respectively, instruct the multi-modal large model to perform data analysis on each first data block according to the first prompt word template, and obtain the corresponding analysis result.

[0075] Suppose that for one of the first data blocks, the analysis result output by the multi-modal large model is "20:00-22:00 is the active time period of user purchase of goods, 70% of users purchase goods in this period, and the high-sales commodity type is snacks"; and for another first data block, the analysis result output by the multi-modal large model is "21:00-23:00 is the active time period of user purchase of goods, 62% of users purchase goods in this period, and the high-sales commodity type is drinks". Based on this, after obtaining the two analysis results output by the multi-modal large model, the digital service platform can combine the two analysis results as a new data block to be processed next time, and further optionally obtain another processing rule information to instruct the multi-modal large model to perform data analysis on the new data block.

[0076] Assuming that the digital service platform obtains another processing rule information including a second prompt word template and a first instruction information template, wherein the content of the second prompt word template is "from each data block, analyze the proportion of users whose consumption amount is concentrated in 500-1000 yuan in the time period of 20:00-23:00", and the content of the first instruction information template is "please determine the proportion of users whose consumption amount is concentrated in 500-1000 yuan according to the number of users who purchase goods in both the time periods of 20:00-22:00 and 21:00-23:00, and analyze the purchase reasons of these users". Based on this, the digital service platform can input the second prompt word template, the first instruction information template and the new data block as input data into the multi-modal large model, instruct the multi-modal large model to perform data analysis on the new data block according to the second prompt word template and the first instruction information template, to obtain the corresponding analysis result.

[0077] Assuming that the analysis result output by the multi-modal large model for the new data block is "the proportion of users who both purchase snacks and drinks and whose consumption amount is concentrated in 500-1000 yuan is 47% in the time period of 20:00-23:00, indicating that most users have a high probability of purchasing drinks after purchasing snacks, and the possible reason is that the taste of snacks is heavy and drinks can relieve the taste". Based on this, after obtaining the analysis result output by the multi-modal large model, the digital service platform can further obtain the processing rule information according to the processing requirement, instruct the multi-modal large model to perform subsequent data analysis, and so on until the target analysis result meeting the requirement is finally obtained, and generate the visual guidance information suitable for the retail consumption scene according to the target analysis result, to provide guidance service for user behavior.

[0078] It should be noted that the above is only an example for illustration, which aims to illustrate the process of instructing the multi-modal large model to perform data analysis on multiple data blocks according to the processing rule information when the digital service platform performs the target processing operation, which is not limited thereto in actual application.

[0079] For example, after the digital service platform merges the multiple analysis results obtained by performing the target processing operation this time to obtain a new data block, it can also obtain the processing rule information suitable for the next time when performing the target processing operation, instruct the multi-modal large model to analyze the common features, difference features and association features between each other of the data respectively from different data blocks in the new data block, and so on.

[0080] For example, in addition to analyzing the user behavior data, the digital service platform can also analyze the supply resource data by performing the target processing operation. For example, the digital service platform can indicate, by performing the target processing operation, the sales of different brands of goods, the sales of different goods under the same brand, the sales difference of the same brand of goods provided by different providers, and the like. Further optionally, the digital service platform can analyze both the user behavior data and the supply resource data, and further analyze the common features, difference features, and correlation features between the two data, and the like. Based on this, the digital service platform can generate visual guidance information suitable for the retail consumption scene according to the analysis results, to provide guidance services for user behavior, and management and operation guidance services for different providers for various brands of goods.

[0081] The implementation process of other examples is similar to the foregoing examples, and can be understood from the foregoing examples.

[0082] In one or more embodiments of the present specification, in addition to the plurality of prompt word templates and the plurality of indication information templates, the processing rule information preset for the target application scenario can also include a plurality of verification rule information, for verifying the analysis results obtained by the digital service platform each time the target processing operation is performed. The specific content of the plurality of verification rule information is not limited, and the content of the verification rule information can be different according to actual needs.

[0083] For example, one verification rule information is used to indicate whether the proportion of the data covered by the analysis results in the corresponding data block meets a preset coverage threshold (such as the coverage data proportion ≥ 95%), to avoid missing important data; for example, one verification rule information is used to indicate whether the matching degree of the analysis results and the analysis direction indicated by the indication information template used this time meets a preset matching threshold (such as the semantic similarity ≥ 0.8), to ensure that the analysis results do not deviate from the core target; for example, one verification rule information is used to indicate whether the proportion of repeated information in the analysis results compared with the analysis results of the same batch of data blocks in different iteration rounds is lower than a preset ratio (such as the repeated information proportion ≤ 30%), to avoid redundant analysis results; for example, one verification rule information is used to indicate whether the analysis results themselves have logical contradictions, whether the numerical range is correct, and the like.

[0084] Of course, the foregoing is only an example, and the specific content of the verification rule information can be set according to actual needs, which will not be repeated here.

[0085] Based on this, the digital service platform can first obtain the multiple preset verification rule information for the target application scenario when performing partial merging on the multiple first analysis results each time to obtain the next data block to be processed. Then, according to the multiple verification rule information, the credibility of each first analysis result is verified, and the multiple first analysis results whose credibility meets the preset requirement are partially merged to obtain the next data block to be processed.

[0086] Further optionally, for at least one first analysis result whose credibility does not meet the preset requirement, the digital service platform can determine the corresponding at least one second data block, and then obtain the second processing rule information adapted to the target application scenario, and based on the at least one second data block and the second processing rule information, instruct the multi-modal large model to perform data analysis on the at least one second data block again until the credibility of the analysis result meets the preset requirement. Alternatively, the digital service platform can also perform data splitting on the at least one second data block again to obtain multiple third data blocks, and then based on the multiple third data blocks and the second processing rule information, instruct the multi-modal large model to perform data analysis on the multiple third data blocks in parallel.

[0087] It should be noted that the process of re-analyzing at least one second data block or splitting at least one second data block into multiple third data blocks and iteratively analyzing the data is similar to the implementation principle described in the foregoing embodiments, and can be referred to the description of the foregoing embodiments, which will not be repeated here.

[0088] In the foregoing embodiments, the process of performing the target processing operation by the digital service platform is exemplarily described, and the system structure of the digital service platform is not limited.

[0089] Next, the system structure of the digital service platform is described from the perspective of functional modularization, and the process of performing the target processing operation by the digital service platform through each functional module is described.

[0090] Figure 3 The structure block diagram of a digital service platform is shown, as Figure 3 The digital service platform 300 includes a preprocessing module 310, a management module 320, an execution module 330, a storage module 340, and a display module 350, and each functional module will be described below.

[0091] In one or more embodiments of the present specification, the preprocessing module 310 obtains offline application data and online application data under a target application scenario, and performs preprocessing operation on the offline application data and online application data to obtain multi-modal application data. Based on this, as Figure 3As shown, the preprocessing module 310 obtains the multi-modal application data, and inputs the multi-modal application data to the management module 320. The management module 320 is configured to perform data splitting on the received multi-modal application data according to preset splitting rule information, to obtain a plurality of data blocks.

[0092] Further, as shown, Figure 3 The management module 320 inputs the plurality of data blocks to the execution module 330 after obtaining the plurality of data blocks, so that the execution module 330 schedules the multi-modal large model to perform data analysis on the plurality of data blocks in parallel.

[0093] In one or more embodiments of the present specification, the execution module 330 includes a configuration unit 331 and a plurality of processing units 332. The configuration unit 331 is configured to configure processing rule information for a target application scenario. Each processing unit 331 is configured to receive one data block input by the management module 320, and obtain the pre-configured processing rule information. Based on this, as shown, Figure 3 Each processing unit 331 inputs the received data block and the obtained processing rule information into the multi-modal large model, so as to schedule the multi-modal large model to perform data analysis on each data block according to the processing rule information in parallel.

[0094] That is, in one or more embodiments of the present specification, when the digital service platform performs a target processing operation each time, the digital service platform inputs a plurality of data blocks to be processed this time and first processing rule information into the multi-modal large model, and instructs the multi-modal large model to perform data analysis on the plurality of data blocks to be processed this time in parallel. In fact, the digital service platform sends the plurality of data blocks to be processed this time and the first processing rule information as input data to a plurality of processing units 331 respectively, and instructs the plurality of processing units 331 to schedule the multi-modal large model to perform data analysis on the obtained data blocks to be processed according to the obtained first processing rule information in parallel.

[0095] In one or more embodiments of the present specification, the storage module 340 is configured to store a plurality of analysis results obtained by the multi-modal large model each time according to application scenario classification. As shown, Figure 3 During the process in which the plurality of processing units 331 schedule the multi-modal large model to perform data analysis on the plurality of data blocks in parallel, each execution path obtains a corresponding analysis result. After obtaining each analysis result, the multi-modal large model can input the corresponding analysis result to the storage module 340. Further, after receiving each analysis result, the storage module 340 can store each analysis result in a preset storage space corresponding to the target application scenario.

[0096] Correspondingly, as shown, Figure 3As shown, the management module 320 is further configured to obtain the analysis result obtained by the multi-modal large model each time from the storage module 340, and combine part of the analysis results to obtain the data block to be processed next. Based on this, after multiple rounds of iteration processing, the multi-modal large model finally obtains the target analysis result, and inputs the target analysis result into the storage module 340.

[0097] Further, as shown in Figure 3 After the storage module 340 stores the target analysis result into the preset storage space, the execution module 330 can obtain the target analysis result from the preset storage space, and generate visual guidance information applicable to the target application scenario according to the target analysis result. Based on this, the digital service platform can specify a guide strategy for user behavior, and a management and operation strategy for supply resources according to the visual guidance information, so as to guide the user to use the supply resources in subsequent services and improve the user conversion rate in the platform.

[0098] In one or more embodiments of the present specification, the execution subject for obtaining the target analysis result from the preset storage space and generating the visual guidance information is not limited, and can be any one of the plurality of processing units 331, or can be a specified processing unit, which is not limited herein.

[0099] Further, as shown in Figure 3 After the storage module 340 stores the target analysis result into the preset storage space, the execution module 330 can obtain the target analysis result from the preset storage space, and generate visual guidance information applicable to the target application scenario according to the target analysis result. Based on this, the digital service platform can specify a guide strategy for user behavior, and a management and operation strategy for supply resources according to the visual guidance information, so as to guide the user to use the supply resources in subsequent services and improve the user conversion rate in the platform.

[0100] Further, as shown in Figure 3 After the storage module 340 stores the target analysis result into the preset storage space, the execution module 330 can obtain the target analysis result from the preset storage space, and generate visual guidance information applicable to the target application scenario according to the target analysis result. Based on this, the digital service platform can specify a guide strategy for user behavior, and a management and operation strategy for supply resources according to the visual guidance information, so as to guide the user to use the supply resources in subsequent services and improve the user conversion rate in the platform.

[0101] In each of the above embodiments, the digital service platform splits the multi-modal application data in the target application scenario into multiple first data blocks, and schedules the multi-modal large model to perform data analysis on each first data block in parallel, thereby reducing the size of each data block received by the multi-modal large model. Moreover, in each processing process, the digital service platform merges the multiple analysis results analyzed by the multi-modal large model to obtain the data block to be processed next, thereby further reducing the sequence length input to the multi-modal large model each time. Through continuous iterative processing, the data processing difficulty of the multi-modal large model is gradually reduced, the data analysis accuracy is improved, and for large-scale data application scenarios, better processing effect can be achieved.

[0102] In addition, in another optional embodiment of the present specification, for an application scenario with small data scale, after the digital service platform splits the multi-modal application data in the target application scenario into multiple first data blocks, the multiple first data blocks are taken as the multiple data blocks to be processed first. Further, the digital service platform can obtain third processing rule information adapted to the target application scenario, and input the multiple data blocks to be processed this time and the third processing rule information into the multi-modal large model, instructing the multi-modal large model to perform data analysis on the received multiple data blocks to obtain corresponding second analysis results.

[0103] Figure 4 A schematic diagram of a digital service platform performing a target processing operation process according to one or more embodiments of the present specification.

[0104] As shown in Figure 4 In this example, in any execution of the target processing operation process, the digital service platform selects any unprocessed data block from the multiple data blocks to be processed this time for data analysis to obtain corresponding second analysis results. Further, the obtained second analysis results are merged with the unprocessed data block to be processed next as the data block to be processed next, and the iterative processing is continued until there is only one unprocessed data block, and the multi-modal large model performs data analysis on the data block to obtain corresponding target analysis results. Based on this, the digital service platform can generate visual guidance information applicable to the retail consumption scenario based on the target analysis results to provide guidance services for the management and operation of user behavior or supply resources.

[0105] Compared with Figure 2 the processing mode shown, the processing mode of the present embodiment is simpler, and for an application scenario with small data scale, the required analysis results can be quickly and accurately obtained.

[0106] It should be noted that in the present embodiment, for how the digital service platform obtains the third processing rule information and how the digital service platform instructs the multi-modal large model to perform the target processing operation based on the third processing rule information, reference can be made to the descriptions of the corresponding parts in the foregoing embodiments, which will not be repeated here.

[0107] It should be further noted that in actual applications, the two processing modes shown in Figure 2 and Figure 4 may also be combined.

[0108] For example, after the digital service platform splits the multi-modal application data in the target application scenario into a plurality of first data blocks, the digital service platform can input the plurality of data blocks to be processed to the multi-modal large model through a plurality of execution paths (for example, input the plurality of data blocks to be processed to each processing unit 332 in Figure 3 , and schedule the multi-modal large model to perform data analysis on the plurality of data blocks to be processed received by each processing unit 332 through each processing unit 332). Further, in each execution path, the processing mode shown in Figure 4 is used to serially perform data analysis on the plurality of data blocks to be processed, and the processing mode shown in Figure 2 is used to merge the analysis results obtained by part of the execution paths in parallel to obtain the next data block to be processed. Such iterative processing is performed until there is only one data block to be processed, and then the multi-modal large model performs data analysis on the data block to obtain the corresponding target analysis result.

[0109] Of course, the above is only an example and is not limited thereto in actual applications. The specific processing mode can be determined according to the data characteristics and processing requirements.

[0110] In summary, the guidance information generation method provided by one or more embodiments of the present specification can perform data analysis on a plurality of data blocks in parallel through iterative processing of the multi-modal large model. Moreover, when the multi-modal large model performs data analysis on each data block each time, the key features meeting the requirements can be extracted from each data block. Since these key features are more explicit than the data characteristics of the original data blocks, the data amount of the next data block to be processed obtained by merging part of the analysis results is smaller.

[0111] As can be seen, through continuous iterative processing, not only can the length of the sequence input to the multi-modal large model each time be continuously reduced, but also the analysis direction of the multi-modal large model can be continuously focused, thereby improving the accuracy of data analysis. Especially for large-scale data application scenarios, better data analysis effect can be achieved.

[0112] It can be understood that the execution subject of each step in the above method for generating guidance information can be the same device, or the method can also be executed by different devices as the execution subject.

[0113] In addition, in some of the processes described in the above embodiments and the accompanying drawings, a plurality of operations appearing in a specific order are included, but it should be clearly understood that these operations can be executed in the order appearing in the text or in parallel, and the serial numbers of the operations, such as S102, S104, etc., are only used to distinguish different operations, and the serial numbers themselves do not limit the execution order. In addition, these processes can include more or fewer operations, and these operations can be executed in sequence or in parallel.

[0114] It should be noted that the "first", "second", and the like in the text are used to distinguish different information, devices, modules, etc., and do not represent the order, nor do "first" and "second" represent different types. The above embodiments are only examples, and the above embodiments can be modified during actual implementation. Those skilled in the art can understand that the modified method of the above embodiments without creative labor falls within the protection scope of the embodiments of the present specification, and the embodiments will not be described again. All optional technical solutions described above can be used as reference or combined to form optional embodiments of the present specification, which will not be described one by one.

[0115] Based on the same inventive concept, one or more embodiments of the present specification also provide a guidance information generation device. Figure 5 The structure block diagram of a guidance information generation device provided by one or more embodiments of the present specification is shown in Figure 5 As shown in the figure, the guidance information generation device 500 includes an acquisition module 501, an analysis module 502, an execution module 503, and a generation module 504, wherein:

[0116] The acquisition module 501 is configured to acquire multi-modal application data in a target application scenario, wherein the multi-modal application data comprises user behavior data or supply resource data; the splitting module 502 is configured to split the multi-modal application data into a plurality of first data blocks according to preset splitting rule information of the target application scenario; the execution module 503 is configured to iteratively execute a target processing operation based on the plurality of first data blocks until only one data block to be processed is left, and obtain a target analysis result of the data block; the generation module 504 is configured to generate visual guidance information applicable to the target application scenario based on the target analysis result, to provide guidance services for management and operation of user behavior or supply resources; wherein the target processing operation comprises instructing the multi-modal large model to perform data analysis on the data block to be processed this time based on the processing rule information acquired this time to obtain a corresponding analysis result, and merging to obtain a data block to be processed next time based on the analysis result, and the first data block to be processed is the first data block; the processing rule information is used to instruct the multi-modal large model to extract user behavior features from each data block to be processed and analyze user behavior preferences based on the user behavior features, or extract supply resource attributes from each data block to be processed and analyze the adaptation degree of the corresponding supply resource data to the target application scenario.

[0117] In an optional embodiment, the acquisition module 501 acquires multi-modal application data in a target application scenario, configured to: acquire offline application data and online application data in the target application scenario, wherein the offline application data and the online application data are multi-modal big data; and perform a preprocessing operation on the offline application data and the online application data respectively to obtain multi-modal application data to be analyzed.

[0118] In an optional embodiment, the acquisition module 501 is further configured to acquire first processing rule information adapted to the target application scenario; and the execution module 503, when executing the target processing operation at any time, is configured to: input the plurality of data blocks to be processed this time and the first processing rule information into the multi-modal large model, instruct the multi-modal large model to perform data analysis on the plurality of data blocks to be processed this time in parallel, and obtain a plurality of first analysis results; and perform partial merging on the plurality of first analysis results to obtain a data block to be processed next time.

[0119] In an optional embodiment, the acquisition module 502 acquires first processing rule information adapted to the target application scenario, configured to: based on a plurality of prompt word templates and a plurality of indication information templates preset for the target application scenario, acquire a target prompt word template and a target indication information template as the first processing rule information adapted to the target application scenario; wherein different prompt word templates are used to instruct the multi-modal large model to perform data analysis on each data block to be processed in different analysis stages, and different indication information templates are used to instruct the multi-modal large model to perform data analysis on each data block to be processed in different analysis directions.

[0120] In an optional embodiment, the execution module 503 inputs the plurality of data blocks to be processed this time and the first processing rule information into the multi-modal large model, instructs the multi-modal large model to perform data analysis on the plurality of data blocks to be processed this time in parallel, for: sending the plurality of data blocks to be processed this time and the first processing rule information as input data to a plurality of processing units respectively; instructing the plurality of processing units to schedule the multi-modal large model to perform data analysis on the obtained data blocks to be processed according to the obtained first processing rule information in parallel.

[0121] In an optional embodiment, the execution module 503 performs partial merging on the plurality of first analysis results to obtain data blocks to be processed next, for: obtaining a plurality of verification rule information preset for the target application scenario; verifying the credibility of each first analysis result according to the plurality of verification rule information; and performing partial merging on a plurality of first analysis results whose credibility meets preset requirements to obtain data blocks to be processed next.

[0122] In an optional embodiment, the execution module 503 is further configured to: determine at least one second data block corresponding to at least one first analysis result whose credibility does not meet the preset requirements; obtain second processing rule information adapted to the target application scenario; and instruct the multi-modal large model to perform data analysis on the at least one second data block again based on the at least one second data block and the second processing rule information until the credibility of the analysis result meets the preset requirements.

[0123] In an optional embodiment, the execution module 503 instructs the multi-modal large model to perform data analysis on the at least one second data block again based on the at least one second data block and the second processing rule information, for: performing data splitting on the at least one second data block again to obtain a plurality of third data blocks; and instructing the multi-modal large model to perform data analysis on the plurality of third data blocks in parallel based on the plurality of third data blocks and the second processing rule information.

[0124] In an optional embodiment, the execution module 503 is further configured to: store a plurality of analysis results obtained by the multi-modal large model each time in a preset storage space.

[0125] In an optional embodiment, the generation module 504 generates visual guidance information applicable to the target application scenario according to the target analysis result, for: obtaining the target analysis result from the preset storage space; and generating the visual guidance information applicable to the target application scenario according to the target analysis result.

[0126] In an optional embodiment, the obtaining module 501 is further configured to obtain third processing rule information adapted to the target application scenario; and the performing module 503 performs the target processing operation any one time, configured to: input the data block to be processed this time and the third processing rule information into the multi-modal large model, instruct the multi-modal large model to perform data analysis on the received data block, and obtain a corresponding second analysis result; and combine the second analysis result with one first data block that has not been processed, as a data block to be processed next time; wherein the first data block to be processed is one of the first data blocks.

[0127] In an optional embodiment, the performing module 503 is further configured to display the visual guidance information in at least one visual manner; or send the visual guidance information to a provider that provides the supply resource data.

[0128] It should be noted that, since the principle of the problem solved by the device for generating guidance information is similar to the foregoing method for generating guidance information, the implementation of the model training device can refer to the implementation of the foregoing method for training a model, and the repeated parts will not be described herein.

[0129] Based on the same inventive concept, one or more embodiments of the present specification also provide an electronic device.

[0130] Figure 6 A structural block diagram of an electronic device according to one or more embodiments of the present specification is provided.

[0131] As shown in Figure 6 the electronic device 600 can include a processor 601, a memory 602, a program or instruction stored on the memory 602 and executable on the processor 601, which, when executed by the processor 601, implements various processes of the foregoing method embodiments and achieves the same technical effects. To avoid repetition, details are not described herein.

[0132] It should be noted that the electronic device in one or more embodiments of the present specification includes a mobile electronic device and a non-mobile electronic device.

[0133] One or more embodiments of the present specification also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements various processes of the foregoing method embodiments and achieves the same technical effects. To avoid repetition, details are not described herein.

[0134] The processor is the processor in the electronic device in the foregoing embodiments. The readable storage medium includes a computer-readable storage medium, such as a computer readable only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, etc.

[0135] The specification is described with reference to flow diagrams and / or block diagrams of methods, apparatus (systems) and computer program products according to one or more embodiments of the specification. It will be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow diagrams and / or block diagrams. Figure 1 apparatus that implements or facilitates the functions specified in the flow diagrams and / or block diagrams block or blocks.

[0136] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow diagrams and / or block diagrams. Figure 1 apparatus that implements or facilitates the functions specified in the flow diagrams and / or block diagrams block or blocks.

[0137] The computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow diagrams and / or block diagrams. Figure 1 Figure 1 apparatus that implements or facilitates the functions specified in the flow diagrams and / or block diagrams block or blocks.

[0138] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments. In this paper, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. The above terms have specific meanings in the embodiments of the specification, which can be understood by those skilled in the art according to specific circumstances.

[0139] It should be noted that the embodiments in the specification and the features in the embodiments can be combined with each other without conflict. The specification is not limited to any single aspect or any single embodiment, nor is it limited to any combination and / or permutation of aspects and / or embodiments. Moreover, each aspect and / or embodiment of the specification can be used alone or in combination with one or more other aspects and / or embodiments.

[0140] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the specification, and not to limit them; although the specification has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the specification, and they should be covered in the scope of the specification.

Claims

1. A method for generating guidance information, characterized in that, include: Acquire multimodal application data in the target application scenario, wherein the multimodal application data includes user behavior data or supply resource data; Based on the pre-defined splitting rules for the target application scenario, the multimodal application data is split into multiple first data blocks; Based on the multiple first data blocks, the target processing operation is executed iteratively until there is only one data block to be processed, and then the target analysis result of that data block is obtained. Based on the target analysis results, visualized guidance information applicable to the target application scenario is generated to provide guidance services for the management and operation of user behavior or supply resources; The target processing operation includes instructing the multimodal large model to perform data analysis on the data block to be processed in the current time based on the processing rule information obtained this time to obtain the corresponding analysis results, and merging the analysis results to obtain the next data block to be processed, wherein the first data block to be processed is the first data block; The processing rule information is used to instruct the multimodal large model to extract user behavior features from each data block to be processed and analyze user behavior preferences based on the user behavior features, or to extract supply resource attributes from each data block to be processed and analyze the adaptability of the corresponding supply resource data with the target application scenario based on the supply resource attributes.

2. The method according to claim 1, characterized in that, Acquire multimodal application data in the target application scenario, including: Acquire offline and online application data in the target application scenario, wherein the offline and online application data are multimodal big data; Preprocessing operations are performed on the offline application data and the online application data respectively to obtain the multimodal application data to be analyzed.

3. The method according to claim 2, characterized in that, Any execution of a target processing operation includes: Obtain first processing rule information that is adapted to the target application scenario; The multiple data blocks to be processed and the first processing rule information are input into the multimodal large model, which instructs the multimodal large model to perform data analysis on the multiple data blocks to be processed in parallel to obtain multiple first analysis results; The multiple first analysis results are partially merged to obtain the next data block to be processed.

4. The method according to claim 3, characterized in that, Obtaining first processing rule information adapted to the target application scenario includes: Based on multiple preset prompt word templates and multiple preset instruction information templates for the target application scenario, the target prompt word template and target instruction information template are obtained as the first processing rule information adapted to the target application scenario; Among them, different prompt word templates are used to instruct the multimodal large model to perform data analysis on each data block to be processed in different analysis stages, and different instruction information templates are used to instruct the multimodal large model to perform data analysis on each data block to be processed according to different analysis directions.

5. The method according to claim 4, characterized in that, The multiple data blocks to be processed and the first processing rule information are input into the multimodal large model, instructing the multimodal large model to perform data analysis on the multiple data blocks to be processed in parallel, including: The multiple data blocks to be processed and the first processing rule information are used as input data and sent to multiple processing units respectively; The multiple processing units are instructed to schedule the multimodal large model in parallel to perform data analysis on the acquired data blocks to be processed according to the acquired first processing rule information.

6. The method according to claim 5, characterized in that, The multiple first analysis results are partially merged to obtain the next data block to be processed, including: Obtain information on multiple preset verification rules for the target application scenario; Based on the aforementioned multiple verification rule information, verify the credibility of each first analysis result; Multiple first analysis results that meet the preset credibility requirements are partially merged to obtain the next data block to be processed.

7. The method according to claim 6, characterized in that, Also includes: For at least one first analysis result whose credibility does not meet the preset requirements, determine at least one corresponding second data block; Obtain second processing rule information that is adapted to the target application scenario; Based on the at least one second data block and the second processing rule information, the multimodal large model is re-instructed to perform data analysis on the at least one second data block until the credibility of the analysis results meets the preset requirements.

8. The method according to claim 7, characterized in that, Based on the at least one second data block and the second processing rule information, the multimodal large model is re-instructed to perform data analysis on the at least one second data block, including: The at least one second data block is re-splitned to obtain multiple third data blocks; Based on the plurality of third data blocks and the second processing rule information, the multimodal large model is re-instructed to perform data analysis on the plurality of third data blocks in parallel.

9. The method according to any one of claims 1-8, characterized in that, Also includes: The multiple analysis results obtained from each iteration of the multimodal large model are stored in a preset storage space.

10. The method according to claim 9, characterized in that, Based on the target analysis results, visualized guidance information applicable to the target application scenario is generated, including: Obtain the target analysis results from the preset storage space; Based on the target analysis results, generate visual guidance information applicable to the target application scenario.

11. The method according to claim 2, characterized in that, Any execution of a target processing operation includes: Obtain third processing rule information that is compatible with the target application scenario; The data block to be processed and the third processing rule information are input into the multimodal large model, instructing the multimodal large model to perform data analysis on the received data block and obtain the corresponding second analysis result; and The second analysis result is merged with an unprocessed first data block to be processed as the next data block to be processed. The first data block to be processed is one of the multiple first data blocks.

12. The method according to claim 10 or 11, characterized in that, Also includes: The visual guidance information is displayed in at least one visual manner; or, Send the visual guidance information to the provider of the supply resource data.

13. A guidance information generation device, characterized in that, include: The acquisition module is used to acquire multimodal application data in the target application scenario, wherein the multimodal application data includes user behavior data or supply resource data; The splitting module is used to split the multimodal application data into multiple first data blocks according to the splitting rule information preset for the target application scenario; The execution module is used to iteratively execute target processing operations based on the plurality of first data blocks until there is only one data block to be processed, and then obtain the target analysis result of that data block. The generation module is used to generate visual guidance information applicable to the target application scenario based on the target analysis results, so as to provide guidance services for the management and operation of user behavior or supply resources; The target processing operation includes instructing the multimodal large model to perform data analysis on the data block to be processed in the current time based on the processing rule information obtained this time to obtain the corresponding analysis results, and merging the analysis results to obtain the next data block to be processed, wherein the first data block to be processed is the first data block; The processing rule information is used to instruct the multimodal large model to extract user behavior features from each data block to be processed and analyze user behavior preferences based on the user behavior features, or to extract supply resource attributes from each data block to be processed and analyze the adaptability of the corresponding supply resource data with the target application scenario based on the supply resource attributes.

14. An electronic device, characterized in that, The electronic device includes: Memory, used to store computer program products; A processor is configured to execute a computer program product stored in the memory, wherein, when the computer program product is executed, it implements the method described in any one of claims 1-12.

15. A computer-readable storage medium storing a computer program, characterized in that, The computer program is configured to implement the method of any one of claims 1-12 when executed by a processor.

Images