Intelligent Marketing Agent Optimization Methods, Media, and Equipment Based on Intelligent Questioning

Abstract

A method, medium, and device for optimizing an intelligent marketing agent based on intelligent data querying, relating to the field of intelligent agent technology, is disclosed. The method includes: acquiring first task data in response to a first task request, the first task data being obtained based on end-to-end data from the first task request to the execution of a corresponding first strategy; generating a first index based on the first task data, the first index including multiple sub-indexes corresponding to multiple preset feature dimensions; storing the first task data, the first index, and a first mapping relationship between the first index and the first task data in a storage system, the first index being used to recall the first task data as historical task data when at least one sub-index matches a new task request; and the updated storage system being used to optimize the strategy generation logic of the intelligent marketing agent. This method enables continuous automatic optimization of the intelligent marketing agent and effectively improves recall accuracy and efficiency.

Description

Technical Field

[0001] This article relates to the field of intelligent agent technology, specifically to an intelligent marketing intelligent agent optimization method, medium, and device based on intelligent questioning. Background Technology

[0002] As marketing tasks evolve from single-out outreach to multi-stage, long-cycle decision-making processes, intelligent marketing agents in related technologies rely on single-session interactions or static rules for task strategy. When faced with continuous marketing tasks, this leads to insufficient strategy consistency, and even contradictory decisions or repeated trial and error. Furthermore, the optimization of intelligent marketing agents typically relies on manual processing, resulting in low optimization efficiency. Summary of the Invention

[0003] This content section is provided to briefly introduce the ideas, which will be described in detail in the examples section later. This content section is not intended to identify key or essential features of the claimed content, nor is it intended to limit the scope of the claimed content.

[0004] Firstly, a method for optimizing intelligent marketing agents based on intelligent questioning is provided, including: In response to a first task request from the intelligent marketing agent, first task data is obtained. The first task data is obtained based on end-to-end data from the first task request to the execution of the corresponding first strategy. The first strategy is generated by the intelligent marketing agent based on the first task request, first historical task data, and first question data. The first historical task data is retrieved from the storage system based on the first task request, and the first question data is obtained by calling the intelligent question data system based on the first task request. The first task data includes the first task request, the first historical task data, the first question data, and the first strategy. A first index is generated based on the first task data. The first index includes multiple sub-indexes corresponding to multiple preset feature dimensions, and each sub-index corresponds to one preset feature dimension. The first task data, the first index, and the first mapping relationship between the first index and the first task data are stored in the storage system. The first index is used to recall the first task data as historical task data when at least one of the sub-indexes matches a new task request. The updated storage system is used to optimize the strategy generation logic of the intelligent marketing agent.

[0005] In a second aspect, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processing device, implements the method described in the first aspect.

[0006] Thirdly, an electronic device is provided, comprising: A storage device on which computer programs are stored; A processing device for executing the computer program in the storage device to implement the method described in the first aspect.

[0007] Using the above method, in response to a first task request from the intelligent marketing agent, first task data is acquired, and then a first index is generated based on the first task data. The first task data, the first index, and the first mapping relationship between the first index and the first task data are stored in the storage system. Specifically, the first task data is obtained based on end-to-end data from the first task request to the execution of the corresponding first strategy. The first strategy is generated by the intelligent marketing agent based on the first task request, first historical task data, and first question count data. The first historical task data is retrieved from the storage system based on the first task request, and the first question count data is obtained by calling the intelligent question count system based on the first task request. The first index includes multiple sub-indexes corresponding to multiple preset feature dimensions, each sub-index corresponding to one preset feature dimension. The first index is used to retrieve the first task data as historical task data when at least one sub-index matches a new task request, and the updated storage system is used to optimize the strategy generation logic of the intelligent marketing agent.

[0008] This method constructs a queryable historical experience database for the intelligent marketing agent by storing end-to-end data from task requests to strategy execution. This database enables data-driven automatic optimization of the agent's strategy generation logic. The two work synergistically to form a closed-loop optimization process for the intelligent marketing agent, effectively ensuring the consistency of long-term strategies, reducing inconsistencies in decisions or repeated trial and error, and shortening the optimization cycle. This allows for continuous automatic optimization of the intelligent marketing agent to meet rapidly iterating business needs. Furthermore, by constructing a multi-dimensional feature index of historical experience, it not only enables refined matching of historical experience from multiple dimensions but also supports multi-dimensional combined queries, effectively improving recall accuracy and efficiency. The index structure is also easily expandable to accommodate dynamic additions of different dimensions.

[0009] Other features and advantages will be described in detail in the following examples section. Attached Figure Description

[0010] The above and other features, advantages, and aspects of this document will become more apparent when viewed in conjunction with the accompanying drawings and the following examples. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale. In the drawings: Figure 1This is a schematic diagram illustrating an implementation environment according to an example.

[0011] Figure 2 This is a flowchart illustrating an intelligent marketing agent optimization method based on intelligent questioning, as shown in an example.

[0012] Figure 3 This is a schematic diagram of the structure of an intelligent marketing agent system, as illustrated by an example.

[0013] Figure 4 This is a flowchart illustrating an intelligent marketing agent optimization method based on intelligent questioning, as shown in an example.

[0014] Figure 5 This is a schematic diagram of an asynchronous sorting process illustrated by an example.

[0015] Figure 6 This is a schematic diagram of the structure of an intelligent marketing agent optimization device based on intelligent questioning, as illustrated by an example.

[0016] Figure 7 This is a schematic diagram of the structure of an electronic device as illustrated by an example. Detailed Implementation

[0017] The following description will be given in more detail with reference to the accompanying drawings. While certain scenarios are shown in the drawings, it should be understood that this document can be implemented in various forms and should not be construed as limited to the scenarios described herein. Rather, these scenarios are provided to provide a more thorough and complete understanding of this document. It should be understood that the accompanying drawings and the scenarios depicted are for illustrative purposes only and are not intended to limit the scope of this document.

[0018] It should be understood that the steps described in the method may be performed in different orders and / or in parallel. Furthermore, the method may include additional steps and / or omit the steps shown. The scope of this document is not limited in this respect.

[0019] The term "comprising" and its variations can be open-ended, meaning "including but not limited to". The term "based on" can mean "at least partially based on". The term "one case" means "at least one case"; the term "another case" means "at least one additional case"; the term "some cases" means "at least some cases". Definitions of other terms will be given in the following description.

[0020] It should be noted that the concepts of "first" and "second" are only used to distinguish different devices, modules or units, and are not used to limit the order of the functions performed by these devices, modules or units or their interdependencies.

[0021] It should be noted that the modifiers “one” and “multiple” can be illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as “one or more”.

[0022] The names of messages or information exchanged between multiple devices are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0023] It is understandable that the data involved in the technical solution (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of relevant regulations.

[0024] As marketing tasks have evolved from single-out outreach to cross-channel, long-term, multi-stage, and highly feedback-driven decision-making processes, marketing agents have begun to be applied to scenarios such as lead nurturing, target audience segmentation, campaign recommendations, communication script generation, outreach planning, follow-up support, user feedback analysis, and marketing performance review. Marketing agents are typically implemented based on Large Language Models (LLMs), rule engines, retrieval-enhanced generation systems, or automated workflow systems. Their inputs can include marketing objectives, historical interaction summaries, channel constraints, marketing materials, and real-time feedback. Their outputs can include marketing strategies, execution plans, marketing content, communication scripts, follow-up pacing, and anomaly handling suggestions.

[0025] In related technologies, intelligent marketing agents typically employ the following implementation methods: 1. Constraining model behavior based on static prompts or fixed system prompts; 2. Generating marketing actions based on pre-configured rules, tags, strategy patterns, or decision trees; 3. Periodically optimizing the intelligent marketing agent based on human experience; 4. Saving some historical experience based on ordinary logs or database records for human review. These methods can be effective in short-cycle, rule-clear, and context-limited scenarios, but they still have significant shortcomings in complex marketing tasks.

[0026] First, marketing tasks inherently have a long cycle. For example, when targeting the same audience (such as users), an agent needs to go through multiple interaction stages, including initial contact, needs analysis, marketing content creation, problem handling, follow-up, and goal achievement. A single dialogue or task context cannot fully cover previous interactions, historical strategy selections, user feedback, and actual results. If the agent relies solely on the current context or fixed rules, it is prone to repeating ineffective strategies or ignoring previously proven effective experiences.

[0027] Secondly, even if task logs are saved in related technologies, they are usually stored in chronological order, lacking a long-term memory organization method for agent reasoning and policy generation. This model is suitable for auditing and troubleshooting, but cannot meet the business needs of agents to perform similar case retrieval, failure reason recall, policy applicability condition judgment, and positive and negative sample comparison. Consequently, agents struggle to quickly locate valuable experiences from a large amount of historical data, and when faced with continuous tasks, policy coherence is insufficient, even leading to contradictory decisions or repeated trial and error.

[0028] Furthermore, experience in marketing scenarios is not stable in the long term. User needs, product characteristics, marketing channels, campaign cycles, script effectiveness, compliance constraints, and business objectives all change over time. Over time, duplicate experiences, outdated experiences, conflicting experiences, and low-value experiences gradually emerge, which in turn reduces the quality of strategy generation and may even cause the agent to reference experiences that are no longer valid or applicable.

[0029] Furthermore, agent optimization typically relies on manual processing. Manual methods suffer from long lead times, low efficiency, and limited coverage. They also struggle to promptly translate successful experiences and lessons learned from individual tasks into reusable execution logic for subsequent tasks, making it difficult to meet rapidly iterating business demands.

[0030] In view of this, this paper provides a method, medium and device for optimizing intelligent marketing agents based on intelligent questioning, in order to solve the above-mentioned technical problems.

[0031] The intelligent marketing agent optimization method based on intelligent questioning presented in this paper can be executed by an electronic device, which can be provided as at least one of a terminal and a server. Figure 1 This is an exemplary schematic diagram illustrating an implementation environment; see [link / reference]. Figure 1 The implementation environment includes: intelligent marketing agent 101, storage system 102, and intelligent data query system 103.

[0032] For example, in response to a task request, the intelligent marketing agent 101 obtains question data from the intelligent question data system 103 and retrieves historical task data from the storage system 102. Based on the task request, question data, and historical task data, it generates a first strategy and executes it to obtain execution feedback results, thereby acquiring end-to-end data. A multi-dimensional feature index is generated for the end-to-end data, and a mapping relationship between the multi-dimensional feature index and the end-to-end data is constructed. The end-to-end data, the multi-dimensional feature index, and the mapping relationship are stored in the storage system 102. Furthermore, the updated storage system 102 can be used to optimize the strategy generation logic of the intelligent marketing agent 101.

[0033] The front-end interface of the intelligent marketing agent 101 and the intelligent data query system 103 is implemented by the terminal, the back-end processing logic of the intelligent marketing agent 101 and the intelligent data query system 103 is implemented by the server, and the storage system 102 can be implemented by a server with storage function.

[0034] For example, a terminal can be at least one of the following devices: smartphone, smartwatch, desktop computer, laptop, virtual reality terminal, augmented reality terminal, wireless terminal, and laptop computer. The terminal has communication capabilities and can access wired or wireless networks. "Terminal" can refer to one of multiple terminals; those skilled in the art will understand that the number of terminals can be more or less. A server can be a standalone physical server, a server cluster consisting of multiple physical servers, or a distributed file system. It can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and artificial intelligence platforms.

[0035] Figure 2 This is a flowchart illustrating an intelligent marketing agent optimization method based on intelligent questioning, as shown in an example. Figure 2 As shown, the method may include the following steps: In step 201, in response to the first task request to the intelligent marketing agent, first task data is obtained. The first task data is obtained based on the end-to-end data from the first task request to the execution of the corresponding first strategy. The first strategy is generated by the intelligent marketing agent based on the first task request, the first historical task data, and the first question data. The first historical task data is retrieved from the storage system based on the first task request, and the first question data is obtained by calling the intelligent question system based on the first task request. The first task data includes the first task request, the first historical task data, the first question data, and the first strategy.

[0036] Among them, the intelligent marketing agent is a large-scale model as its core brain, possessing the capabilities of perception, reasoning, autonomous planning, tool invocation, and continuous self-optimization, and is an AI-powered digital marketing system for the entire marketing chain. The intelligent data query system can be a self-service data query and analysis platform deployed with the intelligent data query agent, capable of responding to data query requests and returning data query results.

[0037] In addition, it can also be applied to customer service agents, sales assistance agents, operational automation agents, or other agent systems that require long-term task experience accumulation.

[0038] In step 202, a first index is generated based on the first task data. The first index includes multiple sub-indexes corresponding to multiple preset feature dimensions, and each sub-index corresponds to a preset feature dimension.

[0039] Each sub-index corresponds to a preset feature dimension, and one or more sub-indexes may be included under the same preset feature dimension, without any restriction.

[0040] In step 203, the first task data, the first index, and the first mapping relationship between the first index and the first task data are stored in the storage system. The first index is used to recall the first task data as historical task data when at least one sub-index matches a new task request, and the updated storage system is used to optimize the strategy generation logic of the intelligent marketing agent.

[0041] In one scenario, the storage system includes an index repository and a database. Storing first task data, a first index, and a first mapping relationship between the first index and the first task data into the storage system includes: storing the first task data into the database and storing the first index and the first mapping relationship into the index repository.

[0042] For example, storing the original task data, the corresponding index, and the mapping relationship separately effectively decouples the data and the index. This allows for dynamic updates and flexible expansion of the index while ensuring data integrity. At the same time, by prioritizing access to the index and querying task data on demand, retrieval efficiency is effectively improved and storage and maintenance difficulty is reduced.

[0043] The database can be implemented using relational databases, document databases, vector databases, search engines, graph databases, object storage, or a combination of these. The index can be implemented using vector indexes, inverted indexes, tag indexes, graph indexes, time-series indexes, hybrid retrieval indexes, or multi-stage retrieval indexes; there are no restrictions on the type.

[0044] Furthermore, sensitive data can be anonymized, summarized, or access-stratified before being entered into the database. The index records necessary information such as task characteristics, strategy characteristics, effect labels, and evidence pointers, instead of directly saving sensitive data. Different roles can use different permissions when accessing raw data, organizing summaries, and index records to reduce data security risks in the long-term memory management process.

[0045] By employing the above method, a queryable historical experience database is built for the intelligent marketing agent by storing end-to-end data from task requests to strategy execution. This database enables data-driven automatic optimization of the agent's strategy generation logic. The synergy of these two approaches forms a closed-loop optimization process for the intelligent marketing agent, effectively ensuring the consistency of long-term strategies, reducing inconsistencies in decisions or repeated trial and error, and shortening the optimization cycle. This allows for continuous automatic optimization of the intelligent marketing agent to meet rapidly iterating business needs. Furthermore, by constructing a multi-dimensional feature index of historical experience, not only can refined matching of historical experience be performed from multiple dimensions, but multi-dimensional combined queries can also be supported, effectively improving recall accuracy and efficiency. The index structure is also easily expandable to accommodate dynamic additions of different dimensions.

[0046] Figure 3 This paper presents a system architecture diagram for the self-evolving marketing agent system. Each module can be implemented by an intelligent marketing agent, or collaboratively by multiple agents including an intelligent marketing agent; there are no restrictions on this. Modules can communicate with each other through message queues, databases, service interfaces, or shared storage. Figure 4 The flowchart below illustrates the intelligent marketing agent optimization method based on intelligent questioning provided in this paper. Figure 3 The system structure shown is Figure 4 The illustrated method flow provides a detailed explanation of the intelligent marketing agent optimization method based on intelligent questioning provided in this paper.

[0047] In one scenario, acquiring first task data includes: parsing a first task request to obtain first task information; in a storage system, performing index matching based on the first task information to obtain a second index matching the first task information and a second mapping relationship corresponding to the second index, and acquiring first historical task data based on the second mapping relationship; generating a first strategy based on the first task information, the first historical task data, and the first number of questions, and acquiring reasoning process data of the intelligent marketing agent generating the first strategy and execution feedback data of the first strategy; wherein, the first task data also includes first task information, reasoning process data, and execution feedback data.

[0048] For example, such as Figure 3As shown, the task receiving and context parsing module receives and parses task requests. Task requests can be triggered by users, operations systems, sales systems, customer relationship management systems, automated marketing platforms, or scheduled tasks. Task requests may include marketing objectives, target users, marketing stages, expected results, execution channels, available materials, constraints, and task deadlines, which can be determined based on the business scenario and are not restricted. The strategy generation and execution module uses historical experience retrieval and intelligent data collection based on the current task feature set. It can also obtain other information such as business constraints (e.g., general constraints or hard-coded constraints) and generate corresponding strategies based on the acquired data. Strategies may include marketing strategies, user classification, outreach plans (e.g., outreach channel selection, outreach timing), script content, marketing content, activity selection, follow-up actions, exception fallback plans, and manual handling conditions, and then call the corresponding execution tools to complete the task. The execution event collection module collects key events during task execution, including task understanding results, model inference summaries, strategy candidates, final strategies, execution actions, user responses, goal achievement results, manual intervention, exception information, and task completion status.

[0049] like Figure 4 As shown, the intelligent marketing agent receives and parses task requests to obtain task information. This task information may include task type, task objective, target users, marketing stage, product or service category, constraints, historical context (such as historical interaction content or its summary), and the current unresolved issue. Constraint information may include channel constraints, content constraints, compliance constraints, budget or frequency constraints, etc., which can be determined based on the business scenario and are not limited thereto. The task information can be viewed as a set of task features. If incomplete information exists in the context, the intelligent marketing agent can generate clarification questions or call external systems to supplement necessary fields.

[0050] The intelligent marketing agent then performs multi-dimensional feature index matching in the index library based on the current task feature set, and recalls the memory corresponding to the matching index (i.e., historical task data) in the database. Based on the obtained data, it generates marketing strategies and execution plans, and executes specific marketing actions according to the strategies, or generates execution suggestions for manual confirmation. During strategy execution, it continuously collects information such as user responses, channel feedback, strategy execution metrics, manual modification records, anomaly information, and task completion status. If the task fails, it can also record the reason for failure and the stage of failure. Furthermore, for long-term marketing tasks, subsequent strategies can be dynamically adjusted based on the execution feedback results of previously executed strategies (such as user feedback and strategy execution results), thereby achieving dynamic strategy adjustment.

[0051] This process preserves key facts, reasoning, strategy selection, execution feedback, and result tags from the task execution process into a searchable raw memory. A searchable, sortable, and interpretable memory index is then built upon this raw memory. This allows the intelligent marketing agent to generate search criteria based on the current task when handling new tasks, retrieve relevant experience from the long-term memory, and then generate the current strategy in conjunction with business constraints. After the task is completed, the new execution process and feedback results are written back to the long-term memory, forming a closed loop. This enables efficient retrieval, filtering, and reuse of historical experience, improving the interpretability of strategy output.

[0052] In one scenario, obtaining first historical task data based on a second mapping relationship includes: obtaining first candidate historical task data corresponding to the second mapping relationship, wherein the second mapping relationship is a mapping relationship between the second index and the first candidate historical task data; cleaning the first candidate historical task data to obtain second candidate historical task data; for each second candidate historical task data, determining a first score based on at least one of the following: semantic similarity between the second candidate historical task data and the first task information, label matching degree between the second candidate historical task data and the first task information, strategy execution indicators, time freshness, confidence, conflict penalty score, and expiration penalty score; and sorting the first candidate historical task data from high to low based on the first scores, and using the first preset number of second candidate historical task data as the first historical task data.

[0053] For example, such as Figure 3 As shown, the memory recall and sorting module is used to generate retrieval queries based on the task feature set, and perform coarse recall, fine sorting, deduplication, conflict filtering and context compression, and output historical experience that can be used by the strategy generation and execution module.

[0054] For example, such as Figure 4 As shown, the intelligent marketing agent can generate a set of retrieval queries based on the task feature set. The queries can include not only natural language semantic queries, but also structured filtering query conditions. That is, it can use a combination of one or more of the following: semantic recall, tag filtering, time filtering, strategy effect recall, failure experience recall, and similar object state recall. Alternatively, it can perform structured filtering first, and then perform semantic recall and comprehensive ranking. There are no restrictions on this.

[0055] For example, for the task of "following up with users who placed orders in the past year," search criteria can simultaneously include user stage, time range, follow-up strategy, reasons for failure, and recent effective communication scripts. This information is then used for index matching, and the historical experience corresponding to the matched indexes is used as initial candidate historical experience. This candidate historical experience is then cleaned to obtain more candidate historical experience, and finally, it is sorted to obtain the top predetermined number of candidate historical experience as recall historical experience to participate in strategy generation. This improves recall efficiency and strategy relevance, effectively ensuring the accuracy of the basis for subsequent strategy generation, while also effectively reducing the computational resource consumption of subsequent data processing, thus improving strategy generation efficiency.

[0056] For example, memory recall ranking can be achieved by comprehensively considering semantic similarity, tag matching, strategy execution metrics (strategy effectiveness), time freshness, confidence, conflict penalty, and expiration penalty to obtain a comprehensive ranking score. The comprehensive ranking score can be a weighted sum of these factors, and the weights of each factor can be set according to needs. For instance, different business scenarios can use different weights. For example, for scenarios with high compliance requirements, the weights of confidence and conflict penalty can be increased; for short-term marketing campaigns, the weight of time freshness can be increased; and for lead nurturing scenarios, the weights of object state matching and failure reason recall can be increased. There are no restrictions on these factors. This effectively improves the accuracy of memory recall ranking.

[0057] In one scenario, the cleaning process includes at least one of the following: for multiple first candidate historical task data with semantic similarity higher than a preset similarity, merging multiple first candidate historical task data and retaining the evidence source features corresponding to each first candidate historical task data; deleting first candidate historical task data with a relevance to the task objective lower than a preset relevance, wherein the first task data includes the task objective; deleting first candidate historical task data that meets a first preset conflict condition; and generating first constraint information for first candidate historical task data that meets a first preset task failure condition, wherein the first constraint information is used to indicate the constraints of the intelligent marketing agent in the process of generating the first strategy.

[0058] For example, such as Figure 4 As shown, the recalled memories can be filtered out as duplicates, those with low relevance to the task objective, and those with severe conflicts. For multiple similar memories, they can be merged and organized into an experience summary, while retaining the source index pointing to the original evidence (i.e., the decision-making basis for each memory generation strategy). For failed experiences that need to be prompted to be avoided by the intelligent marketing agent, they can be compressed into negative constraints, and so on, without restrictions. This improves recall efficiency and strategy relevance, effectively ensures the accuracy of the basis for subsequent strategy generation, and effectively reduces the computational resource consumption of subsequent data processing, thereby improving strategy generation efficiency.

[0059] In one scenario, the first historical task data includes: historical task data whose corresponding historical task meets the preset task success conditions; and / or, historical task data whose corresponding historical task meets the second preset task failure conditions.

[0060] For example, the recall results can include both successful and unsuccessful historical experiences, and the unsuccessful historical experiences can also include strategies to be avoided. Therefore, when generating strategies, the intelligent marketing agent can simultaneously refer to both successful and unsuccessful experiences, avoiding the repetition of known inefficient actions. The preset task success conditions and the second preset task failure conditions can be set according to needs. For example, the preset task success condition is the percentage of target users who perform the first preset action out of all target users; the first preset action can be set according to needs, such as a purchase action. The second preset task failure condition is that target users perform the second preset action; the second preset action can be set according to needs, such as providing negative feedback, and there are no restrictions on this.

[0061] For example, when an intelligent marketing agent receives a task request, it can recall past experiences from similar tasks where it performed well and generate corresponding strategies based on these experiences. It can also recall past failures from similar tasks and avoid adopting similar strategies from those failures. If the current task yields better feedback, this experience will be recorded as a correction to the original failed strategy.

[0062] In one scenario, storing the first task data into a database includes: storing the first task data into the database according to a preset structure, the preset structure including at least one of the following fields: a first field for recording data source information of the first task data; a second field for recording task-related information of the first task data; a third field for recording strategy-related information of the first strategy; a fourth field for recording execution result-related information of the first strategy; a fifth field for recording quality-related information of the first task data; and a sixth field for recording evidence source information of the first task data.

[0063] For example, such as Figure 3 As shown, the database is used to store raw evidence and structured metadata that is not currently stored. This can include structured fields, semi-structured task records, evidence summaries, dialogue fragments, policy summaries, feedback metrics, and manually annotated information. It can also store information such as source, generation time, version information, and task chain to support querying the evidence sources of historical experience. Structured fields are typically task-related information, while semi-structured tasks generally represent the model's reasoning and thinking process.

[0064] For example, such as Figure 4As shown, the entire data chain from task request to execution of the corresponding strategy can be stored in the database to obtain historical experience units for agent strategy generation. The first field (identity and source identifier) ​​can include memory identifier, task identifier, session identifier, source type, source time, version identifier, etc.; the second field (task feature set field) can include task type, marketing stage, target user, channel type, business objective, constraint set, etc.; the third field (strategy field) can include strategy type, strategy identifier, interaction script style, contact timing, follow-up interval, fallback plan, etc.; the fourth field (result field) can include execution result label (e.g., success or recognition), execution indicator status (whether the preset action was executed), feedback score, success indicator, failure reason, manual processing mark, etc.; the fifth field (quality field) can include confidence level, reuse count, time freshness score, value score, conflict mark, expiration mark, etc.; the sixth field (evidence field) can include evidence data body, evidence summary, feedback data body, tracking pointer, etc. These fields can be expanded according to different business scenarios and can be set according to requirements without limitation. The fifth field can record the initial score when it is stored in memory, and can be updated during the subsequent asynchronous processing stage.

[0065] This allows historical task processes and feedback results to be stored as searchable long-term memories, enhancing the ability to reuse long-term memories.

[0066] In one scenario, the preset feature dimensions may include at least two of the following: a first feature dimension, used to indicate the construction of a corresponding sub-index based on the semantic features in the first task data; a second feature dimension, used to indicate the construction of a corresponding sub-index based on the structured label features in the first task data; a third feature dimension, used to indicate the construction of a corresponding sub-index based on the time features in the first task data; a fourth feature dimension, used to indicate the construction of a corresponding sub-index based on the state features of a first object, where the first object is the task object in the first task data; a fifth feature dimension, used to indicate the construction of a corresponding sub-index based on the policy execution indicator features in the first task data; a sixth feature dimension, used to indicate the construction of a corresponding sub-index based on the task anomaly features in the first task data; and a seventh feature dimension, used to indicate the construction of a corresponding sub-index based on the evidence source features in the first task data.

[0067] For example, such as Figure 3As shown, the index generation module generates a multi-dimensional feature index based on the original memory and records the mapping relationship between the original memory and the multi-dimensional feature index. Specifically, the first feature dimension (semantic index) converts task objectives, strategy summaries, user feedback, reasons for failure, and success experiences into semantic vectors for recalling experiences from similar tasks and feedback. The second feature dimension (task type index) establishes label filtering conditions based on structured fields such as marketing scenario, user stage, channel type, strategy category, and result labels (success or failure), reducing irrelevant memories from entering the recall results. The third feature dimension (time index) generates a time-dimensional index based on memory generation time, recent reuse time, and expiration date, identifying recent valid and expired experiences. The fourth feature dimension (marketing object status index) establishes an index based on object lifecycle stage, demand intensity, historical interaction labels, and historical response types, reusing experiences in similar object states. The fifth feature dimension (strategy effect index) is based on the objectives of executing preset actions. The indexes are established based on metrics such as the proportion of objects to all objects, human scoring, reuse effectiveness, and failure rate, to prioritize the recall of high-performing strategies. The sixth feature index (failure reason index) indexes the failure types, timing errors, content mismatches, excessive frequency, abnormal feedback, and tool malfunctions in failed tasks, used to generate error avoidance strategies. The seventh feature dimension (evidence source index) establishes a mapping between "organized summaries, indexed records, recall results, and strategy outputs" and "original task records and original evidence fragments." This mapping can include original memory identifiers, task identifiers, source types, time, version, and original fragment locations, used to locate the original evidence corresponding to these memories. This supports reverse lookups from the final strategy to the original task records, enabling strategy explanation and auditing. For example, if a summary is derived from multiple original memories, this index can recall multiple corresponding original memories. Specific settings can be customized according to requirements and are not limited.

[0068] By employing a multi-dimensional memory indexing mechanism, the accuracy of locating historical experience is improved, and the probability of irrelevant or weakly related memories entering the policy context is reduced. Furthermore, a queryable memory management mechanism that preserves the mapping relationship between original evidence and the processed summary allows the agent's policy generation results to retrieve historical task evidence, enhancing interpretability and auditability.

[0069] In one scenario, the method may further include: in response to satisfying a preset triggering condition, asynchronously performing at least one of the following processing steps on historical task data in the storage system: merging second and third historical task data in the storage system to obtain fourth historical task data, and constructing an index and mapping relationship corresponding to the fourth historical task data, wherein the second and third historical task data satisfy a preset similarity condition, and the fourth historical task data is used to optimize the strategy generation logic; setting an expiration label and / or reducing the recall weight for fifth historical task data in the storage system, wherein the fifth historical task data satisfies a preset expiration condition; and setting an expiration label and / or reducing the recall weight for sixth and seventh historical task data in the storage system. A conflict label is assigned, and at least one of the following is recorded: conflict cause, difference in applicable conditions, and source of evidence. The sixth and seventh historical task data satisfy the second preset conflict condition. For the eighth historical task data in the storage system, a strategy pattern is generated, and the historical task corresponding to the eighth historical task data satisfies the preset task success condition. For the ninth historical task data in the storage system, the failure cause is summarized, and the historical task corresponding to the ninth historical task data satisfies the second preset task failure condition. For the tenth historical task data in the storage system, the quality score is reduced, and the tenth historical task data satisfies the preset score reduction condition. The eleventh historical task data in the storage system is marked, and the eleventh historical task data is used to optimize the strategy generation logic.

[0070] The preset trigger conditions may include reaching a fixed time period, the number of memories exceeding a threshold, the repetition rate of similar memories exceeding a threshold, a decline in recall quality, a strategy conflict, an increase in the failure rate of a certain type of task, manual annotation requiring organization, or the availability of idle system resources, etc., and there are no restrictions on these.

[0071] For example, such as Figure 3 As shown, the asynchronous organization module is used to organize historical memories outside the main task chain, including clustering and merging of repeated experiences, generating merged summaries, identifying expired experiences, downgrading low-value experiences, marking conflicting memories, extracting successful templates, and summarizing reasons for failures, to form organized experience summaries and quality scores.

[0072] For example, such as Figure 4 As shown, the decision to initiate a cleanup process is based on the triggering conditions. If asynchronous cleanup is not triggered, the current round of tasks ends. Newly written memories can still participate in subsequent recalls through incremental indexing. If an asynchronous cleanup task is triggered, it runs in an independent execution queue or background scheduler, without blocking the intelligent marketing agent from processing new task requests.

[0073] For example, such as Figure 5As shown, the database continuously receives historical experience generated from end-to-end data based on intelligent marketing agents, stores it as a memory record, and generates or updates corresponding incremental indexes, storing them in the index library. In response to triggered processing conditions, the asynchronous processing module retrieves the memory record to be processed from the database and performs asynchronous processing.

[0074] For example, similar memories that meet preset similarity criteria such as semantic similarity (similarity greater than a threshold), similar task type, similar strategy (similarity greater than a threshold), and consistent feedback results can be clustered to identify repeated experiences. Repeated experiences can be merged into an experience summary, and the corresponding index can be rebuilt. Simultaneously, the mapping between each original memory and its corresponding evidence can be preserved, facilitating evidence source retrieval. Whether a memory has expired can be determined based on whether preset expiration conditions are met, such as time, activity validity period, product status, channel rules, and business constraints. Expired memories can be deleted, or they can be marked as expired or have their recall weight reduced.

[0075] For example, the quality score of memories that meet preset de-score criteria (low-value memories) such as missing feedback, reuse count below a threshold, unclear results, weak relevance to current business objectives, or manual labeling as low-value can be reduced, and the aforementioned quality fields can be updated to minimize their impact on subsequent strategy generation. The quality score can be determined based on strategy execution metrics, reuse count, manual rating, time freshness, feedback completeness, task importance, or model evaluation results, without any restrictions.

[0076] For example, for memories that have opposite conclusions in the same or similar scenarios (where the degree of conflict of the second preset conflict condition is less than the degree of conflict of the first preset conflict condition), instead of directly deleting one of them, the cause of the conflict, the difference in applicable conditions, and the source of evidence can be recorded. During subsequent recall, the intelligent marketing agent can select a more suitable experience based on the current conditions.

[0077] For example, strategy patterns can be extracted from historical experience of successful tasks that meet preset criteria such as high metrics, high response rates, or effective manual confirmation. These patterns might include applicable target audience states, effective outreach timing, marketing content mix, interactive script style, and follow-up pace. Failure reasons can be summarized from failed tasks that meet preset failure conditions, such as incorrect user stage assessment, channel mismatch, excessive outreach frequency, lack of personalized content, poor timing, premature outreach, tool malfunction, or lack of human intervention. The severity of failure under the first preset failure condition is greater than that under the second preset failure condition. For example, if the first preset failure condition triggers irreversible losses, this is not a limitation.

[0078] For example, it can be done through, as Figure 3The index reconstruction module shown generates an index for the organized experience summary, applicable conditions, prohibited conditions, and quality scores, and records the corresponding mapping relationships. Experiences that meet preset conditions can also be marked and stored separately, such as being written into a high-value experience layer, and a corresponding index can be created. Organized high-value experiences can be saved in the form of summaries, rules, tags, strategy patterns, retrieval enhancement fragments, knowledge graph nodes, or external memory fragments of the model. Preset conditions can be conditions indicating high reference value, such as a quality score greater than a preset threshold or a number of similar memories greater than a preset threshold; there are no restrictions on these. Subsequently, strategies can be updated and rules generated based on the data in the high-value experience layer. The high-value experience layer can be stored separately or in a separate storage area in the database; there are no restrictions on this.

[0079] Through an asynchronous sorting mechanism, it is possible to continuously identify duplicate, expired, low-value, conflicting, successful, failed, and high-value memories, and merge, downweight, or mark them to reduce the interference of invalid memories on the reasoning of intelligent marketing agents and solve the problem of quality decline after the scale of long-term memory increases.

[0080] In addition, such as Figure 3 As shown, the index rebuilding module is used to rebuild or incrementally update the index based on the asynchronous reorganization results. Index rebuilding can be done through incremental updates or switching between old and new versions. Before the new version index is built, the main task chain continues to use the old version index; the switch to the new version is only made after the new version is verified to avoid service unavailability. For example, the index rebuilding module generates a new version index v2 based on the reorganization results. Before v2 is built and verified, the main task chain continues to use the old version index v1; after v2 is verified, the version pointer is switched so that subsequent tasks use v2. Thus, the reorganization and index rebuilding process will not block the ongoing marketing tasks, and subsequent similar tasks can prioritize the recall of high-value experiences and avoid negative experiences.

[0081] In one scenario, the strategy generation logic includes at least one of the following: recall ranking weights, used to adjust the recall priority of historical task data in the storage system; strategy candidate ranking rules, used to adjust the order of multiple candidate strategies generated by the intelligent marketing agent; prompt construction rules, used to construct the prompt context of the intelligent marketing agent; tool call priority, used to adjust the tool call order of the intelligent marketing agent; human intervention conditions, used to describe the conditions that trigger human intervention; constraint set, used to restrict the content that the intelligent marketing agent cannot adopt; fallback strategy, used as a backup solution when the strategy generated by the intelligent marketing agent is unavailable; and task planning template, used to plan the task processing procedure of the intelligent marketing agent.

[0082] For example, such as Figure 3As shown, the logic update module continuously updates the external execution logic of the intelligent marketing agent based on the high-value experience layer and strategy effectiveness statistics. It updates content related to the strategy generation logic of the intelligent marketing agent, such as recall weights, strategy candidate ranking, prompt construction rules, tool invocation priority, fallback strategies, constraint sets, and task planning templates. These can be configured according to requirements and are not restricted. In this way, even if the basic model of the intelligent marketing agent remains unchanged, the strategy selection and execution path in similar tasks will change with long-term experience accumulation.

[0083] For example, retrieval queries can be generated based on the current task feature set to recall similar successful cases, similar failed cases, historical interaction memories, strategy effect statistics, and unusable experiences in the current scenario. The recalled results can be sorted and compressed to construct a task-specific memory context, which can include reusable experiences, strategies to be avoided, applicable conditions, evidence source summaries, confidence levels, etc.

[0084] For example, the intelligent marketing agent generates multiple candidate strategies based on the memory context and performs self-checks on these strategies. The self-checks include whether the strategy meets the task objectives, complies with channel constraints, avoids repeating failed strategies, lacks human confirmation conditions, and does not rely on outdated experience. It can also select the strategy with the highest overall score as the execution strategy, or combine multiple strategies into a phased execution plan.

[0085] Furthermore, after the task is completed, the results are written back to the long-term memory, and the strategy performance statistics are updated. If a certain type of strategy continues to perform well, its ranking weight in similar tasks is increased; if a certain type of strategy continues to fail, negative experience is generated and its recall weight is reduced, etc.

[0086] Through the above process, the strategy generation logic of the intelligent marketing agent will change with the accumulation of historical experience. For example, after discovering that a certain type of object responds poorly to one type of interactive script content but well to another, the strategy priority of the latter can be increased under similar object states; after discovering that a certain channel provides poor feedback during a specific time period, the priority of that channel in similar tasks can be reduced; after discovering that a certain type of failure is often caused by errors in user stage judgment, stage verification can be prioritized in subsequent tasks, and so on, without any restrictions.

[0087] This allows for adjustments to strategy ranking, recall weights, and execution logic based on actual feedback from different scenarios, enabling the agent to gradually develop strategies better suited to the current business needs as tasks accumulate. Furthermore, by transforming experience accumulation, strategy correction, and logic optimization into an automated closed loop, the optimization efficiency of the intelligent marketing agent is improved. It can also support multi-channel, multi-stage, multi-target, and long-cycle marketing tasks, making it particularly suitable for agent applications that require continuous follow-up, continuous learning, and continuous optimization.

[0088] This paper provides an optimization method for intelligent marketing agents based on intelligent question counting, including: responding to a first task request to the intelligent marketing agent, parsing the first task request to obtain first task information; in a storage system, performing index matching based on the first task information to obtain a second index matching the first task information and a second mapping relationship corresponding to the second index, and obtaining first historical task data based on the second mapping relationship, wherein the second mapping relationship includes at least the mapping relationship between the first historical task data and the second index; generating a first strategy based on the first task information, the first historical task data, and the first question counting data, and obtaining the reasoning process data of the intelligent marketing agent in generating the first strategy and the execution feedback data of the first strategy; the first question counting data is based on... The system obtains the data from the first task request by calling the intelligent questioning system; it generates a first index based on the first task data, the first index including multiple sub-indexes corresponding to multiple preset feature dimensions, each sub-index corresponding to a preset feature dimension; the first task data includes first task information, inference process data, execution feedback data, first task request, first historical task data, first questioning data, and a first strategy; the first task data, the first index, and the first mapping relationship between the first index and the first task data are stored in a storage system, wherein the first index is used to recall the first task data as historical task data when at least one sub-index matches a new task request, and the updated storage system is used to optimize the strategy generation logic of the intelligent marketing agent.

[0089] By employing the above method, a queryable historical experience database is built for the intelligent marketing agent by storing end-to-end data from task requests to strategy execution. This database enables data-driven automatic optimization of the agent's strategy generation logic. The synergy of these two approaches forms a closed-loop optimization process for the intelligent marketing agent, effectively ensuring the consistency of long-term strategies, reducing inconsistencies in decisions or repeated trial and error, and shortening the optimization cycle. This allows for continuous automatic optimization of the intelligent marketing agent to meet rapidly iterating business needs. Furthermore, by constructing a multi-dimensional feature index of historical experience, not only can refined matching of historical experience be performed from multiple dimensions, but multi-dimensional combined queries can also be supported, effectively improving recall accuracy and efficiency. The index structure is also easily expandable to accommodate dynamic additions of different dimensions.

[0090] Thus, during the execution of tasks by the marketing agent, content such as task objectives, contextual understanding, strategy generation basis, execution process, user feedback, conversion results, human intervention, and anomaly handling records are stored as queryable long-term memory. Furthermore, multi-dimensional feature indexes, including semantic indexes, tag indexes, time indexes, object state indexes, strategy effect indexes, failure reason indexes, and evidence source indexes, are generated for this long-term memory to support the marketing agent in quickly locating and deeply recalling historical experience in subsequent tasks. Simultaneously, asynchronous processing is performed outside the main task chain, clustering and merging duplicate, expired, low-value, conflicting, or invalid memories, assessing their value, adjusting weights, extracting summaries, marking conflicts, and rebuilding indexes. This enables the marketing agent to generate better marketing strategies based on historical experience and continuously update the strategy generation logic. This effectively improves the stability, adaptability, interpretability, and long-term optimization capabilities of the intelligent marketing agent in complex business scenarios.

[0091] Figure 6 This is a schematic diagram of the structure of an intelligent marketing agent optimization device based on intelligent questioning, as illustrated by an example. Figure 6 As shown, the intelligent marketing agent optimization device 600 based on intelligent questioning includes: The acquisition module 601 is used to acquire first task data in response to a first task request from the intelligent marketing agent. The first task data is obtained based on the end-to-end data from the first task request to the execution of the corresponding first strategy. The first strategy is generated by the intelligent marketing agent based on the first task request, first historical task data, and first question data. The first historical task data is retrieved from the storage system based on the first task request, and the first question data is obtained by calling the intelligent question data system based on the first task request. The first task data includes the first task request, the first historical task data, the first question data, and the first strategy. The generation module 602 is used to generate a first index based on the first task data. The first index includes multiple sub-indexes corresponding to multiple preset feature dimensions, and each sub-index corresponds to one preset feature dimension. Storage module 603 is used to store the first task data, the first index, and the first mapping relationship between the first index and the first task data into the storage system, wherein the first index is used to recall the first task data as historical task data when at least one of the sub-indexes matches a new task request, and the updated storage system is used to optimize the strategy generation logic of the intelligent marketing agent.

[0092] Using the aforementioned device, a queryable historical experience database is built for the intelligent marketing agent by storing end-to-end data from task requests to strategy execution. This database enables data-driven automatic optimization of the agent's strategy generation logic. The two work synergistically to form a closed-loop optimization process for the intelligent marketing agent, effectively ensuring the consistency of long-term strategies, reducing inconsistencies in decisions or repeated trial and error, shortening the optimization cycle, and enabling continuous automatic optimization of the intelligent marketing agent to meet rapidly iterating business needs. Furthermore, by constructing a multi-dimensional feature index of historical experience, not only can refined matching of historical experience be performed from multiple dimensions, but multi-dimensional combined queries can also be supported, effectively improving recall accuracy and efficiency. The index structure is also easily expandable to accommodate dynamic additions of different dimensions.

[0093] Optionally, the preset feature dimension includes at least two of the following: The first feature dimension is used to indicate the construction of corresponding sub-indexes based on the semantic features in the first task data; The second feature dimension is used to indicate the construction of corresponding sub-indexes based on the structured label features in the first task data; The third feature dimension is used to indicate the construction of corresponding sub-indexes based on the time features in the first task data; The fourth feature dimension is used to indicate the construction of a corresponding sub-index based on the state features of the first object, where the first object is the task object in the first task data; The fifth feature dimension is used to indicate the construction of corresponding sub-indexes based on the strategy execution indicator features in the first task data; The sixth feature dimension is used to indicate the construction of corresponding sub-indexes based on the task anomaly features in the first task data; The seventh feature dimension is used to indicate the construction of corresponding sub-indexes based on the evidence source features in the first task data.

[0094] Optionally, the policy generation logic includes at least one of the following: Recall ranking weight is used to adjust the recall priority of historical task data in the storage system; A strategy candidate ranking rule is used to adjust the order of multiple candidate strategies, which are generated by the intelligent marketing agent. Prompt construction rules are used to construct the prompt context of the intelligent marketing agent; Tool call priority is used to adjust the order in which tools are called by the intelligent marketing agent. Human intervention conditions are used to describe the conditions that trigger manual intervention. A constraint set is used to restrict the content that the intelligent marketing agent cannot adopt; A fallback strategy is used as a backup solution in case the intelligent marketing agent generation strategy is unavailable. The task planning template is used to plan the task processing procedures of the intelligent marketing agent.

[0095] Optionally, the acquisition module 601 is used for: Parse the first task request to obtain the first task information; In the storage system, index matching is performed based on the first task information to obtain a second index that matches the first task information and a second mapping relationship corresponding to the second index, and the first historical task data is obtained based on the second mapping relationship; Based on the first task information, the first historical task data, and the first number of questions, the first strategy is generated, and the reasoning process data of the intelligent marketing agent in generating the first strategy and the execution feedback data of the first strategy are obtained. The first task data further includes the first task information, the inference process data, and the execution feedback data.

[0096] Optionally, the acquisition module 601 is used for: Based on the second mapping relationship, the first candidate historical task data corresponding to the second mapping relationship is obtained, wherein the second mapping relationship is the mapping relationship between the second index and the first candidate historical task data; The first candidate historical task data is cleaned to obtain the second candidate historical task data. For each second candidate historical task data, a first score is determined based on at least one of the following: semantic similarity between the second candidate historical task data and the first task information, label matching degree between the second candidate historical task data and the first task information, strategy execution index, time freshness, confidence, conflict penalty score, and expiration penalty score corresponding to the second candidate historical task data. Based on the first score corresponding to each of the second candidate historical task data, sort them from high to low, and take the first preset number of second candidate historical task data as the first historical task data.

[0097] Optionally, the cleaning process includes at least one of the following: For multiple first candidate historical task data with semantic similarity higher than a preset similarity, the multiple first candidate historical task data are merged, and the evidence source features corresponding to each first candidate historical task data are retained; Delete the first candidate historical task data whose relevance to the task objective is lower than a preset relevance, wherein the first task data includes the task objective; Delete the first candidate historical task data that meets the first preset conflict condition; For the first candidate historical task data that meets the first preset task failure condition, first constraint information is generated. The first constraint information is used to indicate the constraints of the intelligent marketing agent in the process of generating the first strategy.

[0098] Optionally, the first historical task data includes: Historical task data that corresponds to historical tasks that meet the preset task success conditions; and / or, Historical task data that corresponds to the historical task that meets the second preset task failure condition.

[0099] Optionally, the intelligent marketing agent optimization device 600 based on intelligent question data further includes a processing module for: In response to the fulfillment of a preset trigger condition, at least one of the following processes is asynchronously performed on the historical task data in the storage system: The second historical task data and the third historical task data in the storage system are merged to obtain the fourth historical task data, and an index and mapping relationship corresponding to the fourth historical task data are constructed, wherein the second historical task data and the third historical task data satisfy a preset similarity condition. For the fifth historical task data in the storage system, an expiration tag is set and / or the recall weight is reduced, and the fifth historical task data meets the preset expiration conditions; For the sixth historical task data and the seventh historical task data in the storage system, a conflict label is set, and at least one of the conflict cause, difference in applicable conditions, and source of evidence is recorded. The sixth historical task data and the seventh historical task data satisfy the second preset conflict condition. For the eighth historical task data in the storage system, a strategy pattern is generated, wherein the historical task corresponding to the eighth historical task data meets the preset task success conditions. For the ninth historical task data in the storage system, the reasons for failure are summarized, and the historical task corresponding to the ninth historical task data meets the second preset task failure condition. The quality score of the tenth historical task data in the storage system is reduced, and the tenth historical task data meets the preset score reduction conditions. The eleventh historical task data in the storage system is marked, and the eleventh historical task data is used to optimize the strategy generation logic.

[0100] Regarding the intelligent marketing agent optimization device 600 based on intelligent questioning mentioned above, the execution logic of each functional module has been explained in detail in the section on methods, and will not be repeated here.

[0101] Based on the same concept, a computer-readable medium is also provided, on which a computer program is stored, which, when executed by a processing device, implements the steps of any of the above-described intelligent marketing agent optimization methods based on intelligent question numbers.

[0102] Therefore, by storing end-to-end data from task requests to strategy execution, a queryable historical experience database is built for the intelligent marketing agent. This database allows for data-driven automatic optimization of the agent's strategy generation logic. The synergy of these two aspects forms a closed-loop optimization process for the intelligent marketing agent, effectively ensuring the consistency of long-term strategies, reducing inconsistencies in decisions or repeated trial and error, shortening the optimization cycle, and enabling continuous automatic optimization of the intelligent marketing agent to meet rapidly iterating business needs. Furthermore, by constructing a multi-dimensional feature index of historical experience, not only can refined matching of historical experience be performed from multiple dimensions, but multi-dimensional combined queries can also be supported, effectively improving recall accuracy and efficiency. The index structure is also easily expandable to accommodate dynamic additions of different dimensions.

[0103] Based on the same concept, an electronic device is also provided, which may include: A storage device on which computer programs are stored; A processing device for executing a computer program stored in a storage device to implement any of the above-described intelligent marketing agent optimization methods based on intelligent questions.

[0104] Therefore, by storing end-to-end data from task requests to strategy execution, a queryable historical experience database is built for the intelligent marketing agent. This database allows for data-driven automatic optimization of the agent's strategy generation logic. The synergy of these two aspects forms a closed-loop optimization process for the intelligent marketing agent, effectively ensuring the consistency of long-term strategies, reducing inconsistencies in decisions or repeated trial and error, shortening the optimization cycle, and enabling continuous automatic optimization of the intelligent marketing agent to meet rapidly iterating business needs. Furthermore, by constructing a multi-dimensional feature index of historical experience, not only can refined matching of historical experience be performed from multiple dimensions, but multi-dimensional combined queries can also be supported, effectively improving recall accuracy and efficiency. The index structure is also easily expandable to accommodate dynamic additions of different dimensions.

[0105] Based on the same concept, a computer program product is also provided, including a computer program that, when executed by a processor, implements the above-described intelligent marketing agent optimization method based on intelligent question data.

[0106] Therefore, by storing end-to-end data from task requests to strategy execution, a queryable historical experience database is built for the intelligent marketing agent. This database allows for data-driven automatic optimization of the agent's strategy generation logic. The synergy of these two aspects forms a closed-loop optimization process for the intelligent marketing agent, effectively ensuring the consistency of long-term strategies, reducing inconsistencies in decisions or repeated trial and error, shortening the optimization cycle, and enabling continuous automatic optimization of the intelligent marketing agent to meet rapidly iterating business needs. Furthermore, by constructing a multi-dimensional feature index of historical experience, not only can refined matching of historical experience be performed from multiple dimensions, but multi-dimensional combined queries can also be supported, effectively improving recall accuracy and efficiency. The index structure is also easily expandable to accommodate dynamic additions of different dimensions.

[0107] The following is for reference. Figure 7 The diagram illustrates a structural schematic of an electronic device 700 suitable for implementing the above-described method. The terminal device may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Personal Computers), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs (Televisions), desktop computers, etc. Figure 7 The electronic device shown is merely an example and should not be construed as limiting its functionality or scope of use.

[0108] like Figure 7 As shown, the electronic device 700 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device 700. The processing unit 701, the ROM 702, and the RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0109] Typically, the following devices can be connected to the input / output interface 705: input devices 706 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 707 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 708 including, for example, magnetic tape, hard disk, etc.; and communication devices 709. Communication device 709 allows electronic device 700 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 An electronic device 700 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0110] In particular, depending on certain circumstances, the processes described in the flowchart above can be implemented as computer software programs. For example, a computer program product is provided, 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 flowchart. This computer program can be downloaded and installed from a network via communication device 709, or installed from storage device 708, or installed from read-only memory 702. When the computer program is executed by processing device 701, it performs the functions defined in the above-described methods.

[0111] It should be noted that the aforementioned computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM, or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In one case, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In another case, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.

[0112] In some cases, communication can be conducted using any currently known or future-developed network protocol, such as HTTP (Hypertext Transfer Protocol), and can be interconnected with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include Local Area Networks (LANs), Wide Area Networks (WANs), Internets (e.g., the Internet), and peer-to-peer networks (e.g., ad-hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0113] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0114] The aforementioned computer-readable medium carries one or more programs. When the electronic device executes the aforementioned one or more programs, the electronic device causes the following to occur: In response to a first task request to an intelligent marketing agent, the electronic device acquires first task data, which is obtained based on end-to-end data from the first task request to the execution of a corresponding first strategy; wherein the first strategy is generated by the intelligent marketing agent based on the first task request, first historical task data, and first question data, the first historical task data being retrieved from a storage system based on the first task request, and the first question data being obtained by calling an intelligent question system based on the first task request; the first task data includes the first task request, first historical task data, first question data, and the first strategy; generates a first index based on the first task data, the first index including multiple sub-indexes corresponding to multiple preset feature dimensions, each sub-index corresponding to a preset feature dimension; and stores the first task data, the first index, and a first mapping relationship between the first index and the first task data into a storage system, wherein the first index is used to retrieve the first task data as historical task data when at least one sub-index matches a new task request, and the updated storage system is used to optimize the strategy generation logic of the intelligent marketing agent.

[0115] Computer program code for performing the above operations can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages, as well as conventional procedural programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0116] The flowcharts and block diagrams in the accompanying figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products under various scenarios. Each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that, in some alternative cases, the functions indicated in the blocks may occur in a different order than those indicated in the figures. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0117] The modules mentioned above can be implemented in software or hardware. In some cases, the name of a module does not necessarily limit the functionality of that module.

[0118] The functions described above can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field-Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application-Specific Standard Parts (ASSPs), Systems on Chips (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0119] In this context, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0120] The above description is merely illustrative and explains the technical principles employed. Those skilled in the art should understand that the scope of this document is not limited to the specific combinations of the above-described technical features, but should also cover any combination of the above-described technical features or their equivalents without departing from the above concept.

[0121] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. Multitasking and parallel processing may be advantageous in certain environments. Similarly, while some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of this paper. Certain features described in the context of a single example can also be implemented in combination in a single example. Conversely, various features described in the context of a single example can also be implemented individually or in any suitable sub-combination in multiple examples.

[0122] Although this document has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims. Regarding the aforementioned apparatus, the specific manner in which the various modules perform their operations has already been described in detail in the section concerning the method, and will not be elaborated upon here.

Claims

1. A method for optimizing an intelligent marketing agent based on intelligent questioning, comprising: In response to a first task request from the intelligent marketing agent, first task data is obtained. The first task data is obtained based on end-to-end data from the first task request to the execution of the corresponding first strategy. The first strategy is generated by the intelligent marketing agent based on the first task request, first historical task data, and first question data. The first historical task data is retrieved from the storage system based on the first task request, and the first question data is obtained by calling the intelligent question data system based on the first task request. The first task data includes the first task request, the first historical task data, the first question data, and the first strategy. A first index is generated based on the first task data. The first index includes multiple sub-indexes corresponding to multiple preset feature dimensions, and each sub-index corresponds to one preset feature dimension. The first task data, the first index, and the first mapping relationship between the first index and the first task data are stored in the storage system. The first index is used to recall the first task data as historical task data when at least one of the sub-indexes matches a new task request. The updated storage system is used to optimize the strategy generation logic of the intelligent marketing agent.

2. The method according to claim 1, wherein the preset feature dimension includes at least two of the following: The first feature dimension is used to indicate the construction of corresponding sub-indexes based on the semantic features in the first task data; The second feature dimension is used to indicate the construction of corresponding sub-indexes based on the structured label features in the first task data; The third feature dimension is used to indicate the construction of corresponding sub-indexes based on the time features in the first task data; The fourth feature dimension is used to indicate the construction of a corresponding sub-index based on the state features of the first object, where the first object is the task object in the first task data; The fifth feature dimension is used to indicate the construction of corresponding sub-indexes based on the strategy execution indicator features in the first task data; The sixth feature dimension is used to indicate the construction of corresponding sub-indexes based on the task anomaly features in the first task data; The seventh feature dimension is used to indicate the construction of corresponding sub-indexes based on the evidence source features in the first task data.

3. The method according to claim 1, wherein the strategy generation logic comprises at least one of the following: Recall ranking weight is used to adjust the recall priority of historical task data in the storage system; A strategy candidate ranking rule is used to adjust the order of multiple candidate strategies, which are generated by the intelligent marketing agent. Prompt construction rules are used to construct the prompt context of the intelligent marketing agent; Tool call priority is used to adjust the order in which tools are called by the intelligent marketing agent. Human intervention conditions are used to describe the conditions that trigger manual intervention. A constraint set is used to restrict the content that the intelligent marketing agent cannot adopt; A fallback strategy is used as a backup solution in case the intelligent marketing agent generation strategy is unavailable. The task planning template is used to plan the task processing procedures of the intelligent marketing agent.

4. The method according to any one of claims 1-3, wherein obtaining the first task data includes: Parse the first task request to obtain the first task information; In the storage system, index matching is performed based on the first task information to obtain a second index that matches the first task information and a second mapping relationship corresponding to the second index, and the first historical task data is obtained based on the second mapping relationship; Based on the first task information, the first historical task data, and the first number of questions, the first strategy is generated, and the reasoning process data of the intelligent marketing agent in generating the first strategy and the execution feedback data of the first strategy are obtained. The first task data further includes the first task information, the inference process data, and the execution feedback data.

5. The method according to claim 4, wherein obtaining the first historical task data based on the second mapping relationship includes: Based on the second mapping relationship, the first candidate historical task data corresponding to the second mapping relationship is obtained, wherein the second mapping relationship is the mapping relationship between the second index and the first candidate historical task data; The first candidate historical task data is cleaned to obtain the second candidate historical task data. For each second candidate historical task data, a first score corresponding to the second candidate historical task data is determined based on at least one of the semantic similarity between the second candidate historical task data and the first task information, the label matching degree between the second candidate historical task data and the first task information, and the evaluation parameters corresponding to the second candidate historical task data. The evaluation parameters include at least one of the strategy execution index, time freshness, confidence, conflict penalty score, and expiration penalty score. Based on the first score corresponding to each of the second candidate historical task data, sort them from high to low, and take the first preset number of second candidate historical task data as the first historical task data.

6. The method according to claim 5, wherein the cleaning process comprises at least one of the following: For multiple first candidate historical task data with semantic similarity higher than a preset similarity, the multiple first candidate historical task data are merged, and the evidence source features corresponding to each first candidate historical task data are retained; Delete the first candidate historical task data whose relevance to the task objective is lower than a preset relevance, wherein the first task data includes the task objective; Delete the first candidate historical task data that meets the first preset conflict condition; For the first candidate historical task data that meets the first preset task failure condition, first constraint information is generated. The first constraint information is used to indicate the constraints of the intelligent marketing agent in the process of generating the first strategy.

7. The method according to claim 4, wherein the first historical task data includes: Historical task data that corresponds to historical tasks and meets the preset task success conditions; And / or, Historical task data that corresponds to the historical task that meets the second preset task failure condition.

8. The method according to any one of claims 1-3, further comprising: In response to the fulfillment of a preset trigger condition, at least one of the following processes is asynchronously performed on the historical task data in the storage system: The second historical task data and the third historical task data in the storage system are merged to obtain the fourth historical task data, and an index and mapping relationship corresponding to the fourth historical task data are constructed, wherein the second historical task data and the third historical task data satisfy a preset similarity condition. For the fifth historical task data in the storage system, an expiration tag is set and / or the recall weight is reduced, and the fifth historical task data meets the preset expiration conditions; For the sixth historical task data and the seventh historical task data in the storage system, a conflict label is set, and at least one of the conflict cause, difference in applicable conditions, and source of evidence is recorded. The sixth historical task data and the seventh historical task data satisfy the second preset conflict condition. For the eighth historical task data in the storage system, a strategy pattern is generated, wherein the historical task corresponding to the eighth historical task data meets the preset task success conditions. For the ninth historical task data in the storage system, the reasons for failure are summarized, and the historical task corresponding to the ninth historical task data meets the second preset task failure condition. The quality score of the tenth historical task data in the storage system is reduced, and the tenth historical task data meets the preset score reduction conditions. The eleventh historical task data in the storage system is marked, and the eleventh historical task data is used to optimize the strategy generation logic.

9. A computer-readable medium having a computer program stored thereon, wherein, When executed by a processing device, the computer program implements the method described in any one of claims 1-8.

10. An electronic device, comprising: A storage device on which computer programs are stored; A processing apparatus for executing the computer program in the storage device to implement the method of any one of claims 1-8.

Images