A method for generating slots for natural language delivery intent
By compiling and executing fidelity calculations using reversible APIs, and combining consumption distribution divergence and in-account competition graphs, the generation of natural language targeting intent slots is optimized. This solves the problems of slot generation being disconnected from platform interface constraints and unreasonable follow-up strategies, achieving efficient and low-cost ad delivery.
Patent Information
- Application Number
- CN202611133553.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-29
- Publication Date
- 2026-11-10
AI Technical Summary
Existing natural language ad placement slot generation technology does not fully consider the interface constraints of the target advertising platform, resulting in slot generation incompatibility, mismatched enumeration values, missing required fields, and unreasonable follow-up question strategies, which increases user interaction costs, causes traffic overlap, and reduces ad placement efficiency.
By calculating the fidelity of reversible API compilation and execution, combined with the consumption distribution divergence and the intra-account competition graph, the slot generation process is optimized to ensure slot compatibility and execution consistency. Furthermore, interaction costs and traffic competition are reduced through minimum follow-up questions and orthogonalization.
It improves the consistency between the advertising intent and the platform's execution results, reduces unnecessary user interactions, minimizes traffic overlap, and enhances overall advertising efficiency.
Smart Images

Figure CN123071027A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of natural language processing and advertising technology, and more specifically, to a method for generating natural language advertising intent slots. Background Technology
[0002] With the rapid development of the digital advertising industry, the automation level of advertising delivery is constantly improving. Natural language interaction, as a convenient human-computer interaction method, is gradually being applied to the expression and processing of advertising delivery intentions. The structured transformation of natural language delivery intentions has become one of the core technologies for realizing intelligent advertising delivery.
[0003] Existing natural language ad placement slot generation technologies mostly extract basic slots only during the front-end semantic parsing stage, failing to incorporate the target advertising platform's interface constraints into the entire slot generation process beforehand. This leads to slots frequently exhibiting platform-unexecutable issues such as field incompatibility, mismatched enumeration values, and missing required fields, requiring manual secondary correction. Furthermore, the ad placement intent is prone to semantic loss, degradation, or shift during conversion into platform parameters, resulting in significant deviations between the actual execution results and the user's original ad placement intent. Existing follow-up triggering mechanisms typically rely solely on the confidence level of natural language parsing, neglecting the substantial impact of different candidate slots on the actual ad spend distribution. This can lead to blind automatic supplementation of slots with low semantic ambiguity but high spend impact, or unnecessary excessive follow-up on slots with high semantic ambiguity but low spend impact, increasing user interaction costs. Simultaneously, existing technologies do not adequately consider the traffic consumption of existing ad campaigns within the same ad account. Newly generated slots can easily overlap with existing campaigns in terms of region, audience, and time of day, triggering traffic competition within the account and consequently increasing traffic costs and decreasing overall ad delivery efficiency.
[0004] To address the above problems, this invention proposes a method for generating natural language delivery intent slots. Summary of the Invention
[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method for generating natural language delivery intent slots.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for generating natural language delivery intent slots includes the following steps: Step S1: Receive a natural language delivery request input by the user, perform sentence segmentation, phrase boundary recognition, action word recognition, and dominance relationship recognition on the natural language delivery request, and extract delivery semantic fragments and generate a candidate slot bundle containing positive slots and excluded slots when the request type belongs to plan creation, copy creation, or plan variant creation. Step S2: Read the interface constraints of the target advertising platform, compile the candidate slot bundle into a set of parameters that the platform can accept, and reverse the parameter set to obtain the reverse semantic slot bundle. Determine the execution fidelity based on the semantic difference between the candidate slot bundle and the reverse semantic slot bundle. Step S3: Predict the consumption distribution for candidate slot bundles that meet the execution fidelity conditions, and determine whether to generate a minimum follow-up based on the consumption distribution divergence between different candidate slot bundles. Step S4: Obtain the set of existing advertising campaigns in the account, convert the candidate slot bundles and existing advertising campaigns into traffic occupancy vectors respectively, and perform slot orthogonalization processing on the candidate slot bundles based on the competition overlap value; Step S5: Determine and output the target slot bundle from the candidate slot bundles based on execution fidelity, account contention overlap value, candidate uncertainty, and excluded slot violation degree; Step S6: Based on the target slot bundle, perform the following: split the delivery level, execute the regional slots and time slots, configure the audience slots and exclusion slots for delivery targeting, execute the budget slots and bid slots, bind the creative slots and conversion links, and perform consistency verification of interface parameters before submission. After user confirmation, call the target advertising platform's creation interface or launch interface to output the advertising delivery results.
[0007] Furthermore, the semantic fragments of the campaign include the target audience, campaign objective, budget, region, time, demographics, creative materials, bidding method, conversion components, landing page, and exclusion objects. When there are multiple feasible interpretations for the same semantic fragment, candidate slot items are generated separately, and they are combined into a candidate slot bundle according to slot type, scope of application, and semantic compatibility.
[0008] Furthermore, the exclusion slot includes exclusion dimension, exclusion object, scope of application, trigger text fragment, execution operator, and priority; for explicit negation expressions, the exclusion scope is determined according to the dependency relationship between the negation word and the dominated object; for indirect exclusion expressions, the object to be avoided is identified, and competitive exclusion slots, crowd exclusion slots, or time period exclusion slots are generated according to the type of the object to be avoided.
[0009] Furthermore, when affirmative slots and exclusion slots conflict on the same dimension, they are resolved according to source priority; among them, exclusion slots explicitly given by the user take precedence over automatically supplemented affirmative slots, platform mandatory interface constraints take precedence over ordinary delivery preferences, and affirmative slots explicitly given by the user take precedence over default supplemented slots; when conflicts cannot be resolved by priority, multiple candidate slot bundles are retained for consumption distribution divergence judgment.
[0010] Furthermore, the interface constraints are determined by the interface field description, account availability capabilities, field enumeration returned by the platform, field mandatory relationships, field mutual exclusion relationships, and interface pre-validation results; the forward compilation includes field naming conversion, field type conversion, enumeration value conversion, mandatory field completion, mutual exclusion field deletion, approximate replacement of fields not supported by the platform, and execution conversion of excluded slots.
[0011] Furthermore, the reverse restoration includes restoring the promotion target, goal, budget, region, audience, time, creative, conversion component and exclusion conditions based on the reverse correspondence between platform fields and delivery semantics; when a natural language slot is deleted, replaced by a default field or replaced by an approximate field in the forward compilation, it is recorded as missing, downgraded or semantically offset in the reverse semantic slot bundle, and the execution fidelity is determined based on the missing, downgraded or semantically offset.
[0012] Furthermore, when the execution fidelity does not meet the preset fidelity conditions, the candidate slot bundle is repaired; the repair includes splitting the non-executable slot into multiple executable sub-slots, converting semantic audiences not supported by the platform into executable audience combinations, converting geographical radii that cannot be directly expressed into geofences or administrative region combinations supported by the platform, or splitting incompatible conversion targets and material types into different candidate slot bundles; the repaired candidate slot bundles are then re-compiled and reconstructed.
[0013] Furthermore, when predicting the consumption distribution, the geographic unit, time period unit, audience unit, creative category and conversion goal are jointly determined as the consumption unit. Based on the estimated traffic returned by the target platform traffic estimation interface, the average consumption intensity of the account in the same or similar consumption units recently, the candidate bid level, the budget release method and the creative availability status, the consumption intensity of the candidate slot bundle in each consumption unit is determined and normalized to obtain the consumption distribution.
[0014] Furthermore, when the overall consumption distribution divergence of the candidate slot bundle set meets the preset low divergence condition, the candidate slot bundle with higher fidelity and lower account competition risk is automatically supplemented; when the overall consumption distribution divergence does not meet the preset low divergence condition, a set of follow-up questions is generated for the undetermined slots, and the follow-up questions are selected according to the divergence reduction benefit corresponding to the unit interaction cost, until the candidate slot bundle set meets the preset low divergence condition, or the candidate slot bundles are further distinguished by subsequent account competition processing.
[0015] Furthermore, step S6 includes: splitting the target slot bundle into advertising targets corresponding to different delivery levels based on the scope of each slot in the target slot bundle; performing platform execution processing on the split advertising targets to convert the regional slots, time slots, audience slots, exclusion slots, budget slots, bidding slots, creative slots, conversion component slots, and account competition correction records into delivery configurations supported by the target advertising platform; generating interface parameters according to the object creation order of the target advertising platform, and verifying the consistency between delivery configurations and the executable status of exclusion slots and account competition avoidance configurations before submission; if the verification result does not meet the submission conditions, the verification exception is located to the corresponding slot in the target slot bundle, and re-execute reversible API compilation, slot orthogonalization processing, or minimal follow-up processing according to the exception type; if the verification result meets the submission conditions, the creation interface or launch interface of the target advertising platform is called after user confirmation, and the advertising delivery result including platform object identifier, delivery status, review status, actual effective slots, account competition correction records, platform return information, and reasons for incompletely effective slots is output.
[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. To address the issues of disconnect between slot generation and platform interface constraints, and low intent execution fidelity in existing technologies, this invention employs a reversible API compilation and execution fidelity calculation technique. It introduces the target advertising platform's interface constraints into the slot generation process beforehand. Candidate slot bundles are forward-compiled into platform-acceptable parameters and then reverse-engineered back into semantic slots. The semantic difference between the two is calculated to obtain execution fidelity, and candidate slot bundles that do not meet the conditions are specifically repaired. This approach ensures that the generated slot bundles retain the original semantics of natural language advertising while also possessing platform executability, thus improving the consistency between the advertising intent and the actual platform execution results. 2. To address the issues of unreasonable follow-up strategies and lack of consideration for traffic competition within accounts in existing technologies, this invention employs minimum follow-up based on consumption distribution divergence and slot orthogonalization techniques based on the intra-account competition graph. By predicting the consumption distribution of different candidate slot bundles, calculating the overall divergence, and selecting follow-up questions based on the divergence per unit interaction cost to reduce unnecessary user interactions, this invention reduces unnecessary user interactions. Simultaneously, by converting candidate slot bundles and existing plans into traffic occupancy vectors, calculating the competition overlap value, and performing orthogonalization correction, the degree of traffic overlap between new plans and existing plans within the account can be reduced, improving the overall account delivery efficiency. Attached Figure Description
[0017] Figure 1 A flowchart of a method for generating natural language delivery intent slots; Figure 2 This is a flowchart of the reversible API compilation and execution fidelity calculation steps of the present invention; Figure 3 This is a flowchart of the minimum inquiry steps based on the consumption distribution divergence of the present invention. Detailed Implementation
[0018] Example, refer to Figure 1 The natural language delivery intent slot generation method of this embodiment specifically includes the following steps: Step S1: Natural language parsing and generation of polarized candidate slot bundles.
[0019] The natural language delivery request is converted into multiple candidate slot bundles, and positive intent and negative intent are identified in the same process. The output is a candidate set for subsequent platform compilation, consumption divergence judgment and account competition processing.
[0020] After receiving a user's natural language delivery request, the system performs sentence segmentation, phrase boundary recognition, action word recognition, and dominance relation recognition on the text. A dependency parsing-based natural language processing tool is used to perform the above processing, parsing the input text to extract the predicate-object structure from the sentence. The predicate corresponds to the action word, and the object corresponds to the object governed by the action word. Action words include words that can influence ad campaign generation, such as create, copy, rerun, scale up, scale down, pause, exclude, don't, avoid, and stagger. These can be predefined as a set containing the above words. Action words in the text are identified through string matching or semantic similarity calculation. The request type is determined based on the action word and its governed object. When the request type belongs to campaign creation, copy creation, or campaign variant creation, the system enters the candidate slot bundle generation process. Semantic fragments for ad placement are extracted from natural language requests. These extraction objects include the target audience, ad objective, budget, region, time, demographics, creatives, bidding method, conversion components, landing page, and excluded objects. For each semantic fragment, it is not directly determined as the final slot. Instead, candidate slot items are first generated. If multiple feasible interpretations exist for the same semantic fragment, candidate slot items are generated separately for each fragment. These candidate slot items are then combined into a bundle based on slot type, scope, and semantic compatibility. Semantic compatibility refers to the absence of logical conflicts between different slot items. It can be determined by predefined slot type constraint rules. Multiple slot items in the same dimension are compatible if their ranges overlap and their polarities are the same, and incompatible if their ranges are mutually exclusive. Slot items in different dimensions are compatible if there is no predefined mutual exclusion relationship. Candidate slot bundle Each slot in the system is denoted as ,in Indicates the candidate slot bundle number. This indicates the slot dimension number; slot dimensions can include budget, region, audience, time period, creative, conversion target, and bidding method, etc.; for slots from the user's original text, record the position of the text fragment; for slots obtained by supplementing with interface constraints or necessary fields, record the reason for supplementation; for slots generated by multiple candidate interpretations, record the candidate source and candidate relationship. Exclusion intents are identified synchronously at this stage. Exclusion intents are not saved as notes, but are converted into exclusion slots. Exclusion slots include at least the exclusion dimension, exclusion object, scope, trigger text fragment, execution operator, and priority. Exclusion dimensions can be audience, region, time period, creative, existing campaign, conversion target, or traffic set. The execution operator indicates what kind of constraint the exclusion slot should be converted into in subsequent execution, such as excluding the audience, prohibiting advertising during the time period, disabling creatives, avoiding traffic from existing campaigns, or isolating conversion targets. For explicit negative expressions such as "don't," "exclude," and "don't invest," the exclusion scope is determined according to the dependency relationship between the negative word and the object being excluded. For indirect exclusion expressions such as "don't compete with a certain plan," "avoid live stream reservations," "don't target existing customers," and "don't accept forms outside of business hours," the object to be excluded is first identified, and then corresponding exclusion slots are generated based on the type of the object to be excluded. When the object to be excluded points to an existing plan or plan type, a competition exclusion slot is generated. When the object to be excluded points to a user group, a user group exclusion slot is generated. When the object to be excluded points to a time range, a time period exclusion slot is generated. The identification of the object to be excluded can be achieved through semantic similarity calculation, matching the referent in the text with the existing plan name, plan type tag, user group tag, and time period description in the account, and selecting the object with the highest similarity as the object to be excluded. When affirmative slots and exclusion slots conflict on the same dimension, they are resolved according to source priority; exclusion slots explicitly given by the user take precedence over automatically supplemented affirmative slots; platform mandatory interface constraints take precedence over normal delivery preferences; affirmative slots explicitly given by the user take precedence over default supplemented slots. If the conflict cannot be resolved by priority, multiple candidate slot bundles are retained for subsequent consumption distribution divergence judgment.
[0021] Step S2: Calculate the fidelity of reversible API compilation and execution.
[0022] like Figure 2 As shown, the advertising platform interface constraints are introduced into the slot generation process in advance. The candidate slot bundle is first compiled into platform executable parameters, and then the platform parameters are reverse-engineered into semantic slots. This allows us to determine whether the semantics after platform execution still match the original placement intention. Figure 1 To.
[0023] For each candidate slot bundle The interface constraints of the target advertising platform are read. These constraints can be determined by the interface field descriptions, account availability capabilities, field enumerations returned by the platform, field mandatory relationships, field mutual exclusion relationships, and interface pre-validation results. Based on these interface constraints, candidate slot bundles are selected. Compile into a set of parameters that the target advertising platform can accept. The compilation process includes field naming conversion, field type conversion, enumeration value conversion, required field completion, mutual exclusion field deletion, approximate replacement of fields not supported by the platform, and execution conversion of excluded slots. Approximate replacement of fields not supported by the platform can be achieved through semantic mapping. A mapping relationship between platform-supported fields and common unsupported semantics is established, the supported field with the closest semantics is selected for replacement, and the replacement relationship is recorded. After completing the forward compilation, do not directly... As a final result, but rather Reverse reconstruction into reverse semantic slot bundle During reverse reconstruction, based on the reverse correspondence between platform fields and campaign semantics, semantic items such as promotion target, goal, budget, region, audience, time, creative, conversion components and exclusion conditions are recovered; if a natural language slot is deleted, replaced by a default field or replaced by a similar field during forward compilation, it is recorded as missing, downgraded or semantically offset in the reverse semantic slot bundle. The execution fidelity between the candidate slot bundle and the reverse semantic slot bundle is calculated according to the following formula: ; in, Indicates the first The execution fidelity of each candidate slot bundle is dimensionless. Indicates the number of slot dimensions involved in the comparison; Indicates candidate slot bundle Middle One slot; Indicates the set of platform parameters The corresponding slot obtained by reverse reconstruction; Indicates the first The weights for each slot dimension range from 0 to 1, and the sum of the weights for all dimensions is 1. The weights can be configured according to the impact of the slot on ad spending, compliance risks, and conversion goals. The weights for budget and conversion goals can be set to 0.15 to 0.25, the weights for region and audience can be set to 0.1 to 0.2, and the weights for time period, creative, and bidding method can be set to 0.05 to 0.15. The semantic difference function has a value range of 0 to 1 and can be determined based on the states of consistency, approximation, missing, and conflict. It takes the value of 0 when the slots are completely consistent, 0.2 to 0.5 when the semantics are approximation, 0.8 to 1 when the slots are missing, and 1 when the semantics are conflicting. when If the preset fidelity conditions are not met, the candidate slot bundles are repaired. The repair methods include splitting unexecutable slots into multiple executable sub-slots, converting semantic audiences not supported by the platform into executable audience combinations, converting geographical radii that cannot be directly expressed into geofences or administrative region combinations supported by the platform, and splitting incompatible conversion targets and material types into different candidate slot bundles. The repaired candidate slot bundles are then re-compiled and reconstructed. If the fidelity conditions are still not met after repair, the corresponding undetermined slots are marked as slots to be further investigated. Through the above processing, the advertising platform API constraints are no longer located after slot generation, but directly affect the candidate results during slot generation. The resulting slot bundle retains the semantics of natural language delivery and also has platform executability.
[0024] Step S3: Minimal pursuit based on consumption distribution divergence.
[0025] like Figure 3 As shown, determining whether ambiguity in candidate slots requires follow-up questions to the user is not solely based on language confidence, but rather on assessing whether different candidate slot bundles will cause a substantial shift in ad spend across regions, time periods, demographics, creatives, or objectives.
[0026] For a candidate slot bundle processed through reversible API compilation, predict its consumption distribution. The statistical space of the consumption distribution is denoted as . Each element This represents a campaign spend unit; a campaign spend unit can be jointly determined by geographic unit, time slot unit, audience unit, creative category, and conversion goal. The geographic unit can be set to province, city, district, or business area; the time slot unit can be set to 1 hour, 2 hours, or half a day; the audience unit can be set to a platform-predefined audience package or a custom audience tag; the creative category can be set to short video, image, live stream, etc. Each spend unit corresponds to a unique combination of the above dimensions; [This refers to] candidate slot bundles. In the consumption unit The consumption intensity is estimated and normalized to obtain the consumption distribution: ; in, Indicates candidate slot bundle In the consumption unit Normalized consumption probability; Indicates candidate slot bundle The corresponding unnormalized consumption intensity, whose dimensions are budget consumption intensity or traffic consumption intensity, is calculated by a linear weighted model. The estimated traffic of the consumption unit returned by the target platform's traffic estimation interface, the account's recent average consumption intensity in the same or similar consumption units, the ratio of the candidate bid level to the account's historical average bid, the consumption rate coefficient corresponding to the budget release method, and the coefficient corresponding to the material's availability status are weighted and summed. Indicates candidate slot bundle Total consumption intensity across all consumption units, and They have the same dimensions; Represents the set of all candidate consumption units; express Any consumption unit used for normalization calculation; the consumption rate coefficient corresponding to the budget release method, which is 1 when releasing at a constant speed and 1.2~1.8 when releasing at an accelerated speed; the coefficient corresponding to the material's availability status, which is 1 when the material has been approved and is available for release, 0.3~0.7 when the material is pending approval, and 0 when the material has not been approved. After obtaining the consumption distribution of multiple candidate slot bundles, the candidate set is calculated. Overall consumption distribution divergence: ; in, Represents the set of candidate slot bundles The overall consumption distribution divergence; and They represent the first The and the first Consumption distribution corresponding to each candidate slot bundle; Indicates by and The intermediate distribution obtained by applying the same weights, i.e. ; This represents the calculation of relative entropy for discrete distributions. and , ; This represents the average value of the input terms; the divergence reflects the space consumption of different candidate slot bundles. The degree of distribution shift on; when When the preset low dispersion condition is met, it indicates that the candidate differences have little impact on actual ad spend, and it is possible to automatically fill candidate slot bundles with higher fidelity and lower account competition risk; when When the preset low dispersion condition is not met, a set of follow-up questions is generated. Each follow-up question corresponds to an undetermined slot, such as time range, store target, audience range, material type or conversion goal. The selection of follow-up questions is based on the following formula: ; in, Indicate the problem The follow-up question is the benefit of reducing divergence due to unit interaction cost; This indicates a follow-up question regarding the set of candidate slots. The overall consumption distribution divergence; Indicates user's opinion on the issue One possible answer; This indicates that when a user answers a question The expected divergence of the remaining candidate slot bundle set can be obtained by taking a weighted average of the divergences corresponding to all possible answers, with the weight being the probability of that answer appearing. Indicate the problem The interaction cost, ranging from 0.1 to 1, can be determined by factors such as the number of question options, whether the user needs to view external information, and whether the question involves budget or permissions. When there are 2 question options, the cost is 0.1 to 0.2; when there are 3 to 5 options, the cost is 0.2 to 0.4; when there are more than 5 options, the cost is 0.4 to 0.6; if external information needs to be viewed, the cost increases by 0.2 to 0.3; if budget or permission adjustments are involved, the cost increases by 0.3 to 0.5. choose For more complex issues, follow up with the user. If multiple candidate slot bundles still exist after the user's response, recalculate. and The process continues until the candidate set meets the low dispersion condition, or the candidate slot bundle can be further distinguished by subsequent account competition processing. This process can avoid excessive questioning due to ambiguity in ordinary text, and can also avoid blindly and automatically filling in slots with significant consumption impact.
[0027] Step S4: Slot orthogonalization based on the competition graph within the account.
[0028] When handling traffic competition between new candidate slot bundles and existing plans in an account, the criterion is not whether the plan names or fields are the same, but rather the degree of overlap between the candidate plan and the existing plan in terms of traffic usage.
[0029] Obtain a collection of existing advertising campaigns within the account that are currently running, awaiting launch, or recently paused but still relevant for reference. For each existing advertising campaign, extract its geographic location, time period, target audience, creative semantics, conversion goals, bidding method, and budget release schedule, and convert them into a traffic usage vector. For candidate slot bundles The same method is used to convert it into a candidate traffic occupancy vector. Each dimension of the traffic usage vector corresponds to a traffic feature, with the dimension value representing the coverage or intensity of that feature. The geographic dimension can be represented as a vector, with each element corresponding to an administrative region. A value of 1 indicates that the region is within the targeting range, while 0 indicates that it is not. The time period dimension can be represented as a vector, with each element corresponding to an hourly time period. A value of 1 indicates that the time period is available for targeting, while 0 indicates that it is not. The audience dimension can be represented as a vector, with each element corresponding to an audience tag, and the element value representing the weight of that tag. The overlap between candidate slot bundles and existing plans in the account is calculated using the following formula: ; in, Indicates the first The maximum competitive overlap of a candidate slot bundle relative to existing plans within the account; This indicates that the account already has a collection of advertising campaigns; This refers to the set of dimensions used to determine whether to participate in the competition, including region, time period, target audience, semantic content, conversion goals, bidding method, and budget release schedule. Representing dimensions The competition weight ranges from 0 to 1, and the sum of the weights of all dimensions is 1. It can be configured according to the impact of this dimension on traffic acquisition and cost increase. The weight of the audience and conversion target dimensions can be set to 0.15~0.25, the weight of the region and time period dimensions can be set to 0.1~0.2, and the weight of the material semantics, bidding method, and budget release rhythm dimensions can be set to 0.05~0.15. Indicates candidate slot bundle In dimensions The flow occupancy vector on; Indicates that there is already a plan. In dimensions The flow occupancy vector on; Indicates the two in dimension The overlap of regions and time periods is determined by the intersection-union ratio, which is the number of elements in the intersection of two vectors divided by the number of elements in the union. The overlap of the population is determined by the label overlap rate, which is the number of common non-zero elements in two vectors divided by the total number of non-zero elements in the two vectors. The semantic overlap of the material is determined by the cosine similarity. The overlap of the conversion target is 1 when the two conversion targets are completely consistent, otherwise it is 0. when When the low competition condition is met, retain the candidate slot bundle; when When the low-competition condition is not met, identify the dominant dimension that causes increased competition overlap. The dominant dimension can be identified through the items in each dimension. The size is determined, and the dimension with the largest value is the dominant dimension. Then, the dimension with smaller semantic loss and the ability to reduce competitive overlap is selected for orthogonalization correction. Semantic loss refers to the degree of semantic difference between the corrected slot and the original slot. It can be calculated using the same semantic difference function as in S2. The dimension with the smallest semantic loss is selected for correction. Orthogonalization correction includes generating slots for audience exclusion, off-peak hours, geographic contraction, creative replacement, conversion target isolation, or budget release rhythm adjustment. If there are explicitly excluded slots in the candidate slot bundle, such as avoiding traffic from a certain plan or not competing with a certain plan, the weight of the avoided plan or plan type in the competition calculation is increased, and slots that are staggered from it are generated first. After each orthogonalization correction, reversible API compilation and competition overlap calculation are re-executed until the candidate slot bundle meets the execution fidelity and low competition conditions, or enters the minimum follow-up process. For exclusion slots, the following execution processing is adopted: object exclusion slots are compiled into exclusion fields when the platform supports them; when the platform does not support direct exclusion, approximate exclusion is achieved through candidate group complementation, alternative audiences, or scope shrinking; time exclusion slots are converted into a set of available time periods, and the excluded time periods are not included in the consumption intensity calculation when predicting consumption distribution; competitive exclusion slots are converted into avoidance constraints in the account competition graph, so that the candidate slot bundle is staggered from the excluded campaign in at least one key traffic dimension; semantic exclusion slots are converted into semantic filtering conditions for creative materials, filtering conditions for landing pages, or filtering conditions for conversion components; the nodes of the intra-account competition graph are the advertising campaigns and candidate slot bundles within the account, and the edges are the competition overlap values between two nodes. If the competition overlap value exceeds a preset threshold, an edge is added, and the weight of the edge is the competition overlap value.
[0030] Step S5: Target slot bundle decision and output.
[0031] The target slot bundle is determined from multiple candidate slot bundles, and the decision takes into account platform execution fidelity, consumption distribution uncertainty, intra-account competition overlap, and exclusion of slot satisfaction.
[0032] Candidate slots are denoted as a set For each candidate slot bundle, calculate the execution fidelity. Account competition overlap value Candidate uncertainty and exclusion slot violation degree ;in, It can be determined by the average consumption distribution divergence between the candidate slot bundle and other candidate slot bundles in the same group. The calculation method is the average value of the consumption distribution KL divergence between the candidate slot bundle and all other candidate slot bundles in the same group, with a value range of 0 to 1. This indicates the degree to which a candidate slot bundle fails to meet the exclusion criteria, ranging from 0 to 1. The calculation method is to divide the number of exclusion slots in the candidate slot bundle that failed to meet the criteria by the total number of exclusion slots, and then multiply by the weight of each exclusion slot. User-specified exclusion slot weights range from 0.8 to 1, while automatically generated exclusion slot weights range from 0.4 to 0.7. A value of 1 indicates complete failure to meet the criteria, 0.2 to 0.8 indicates partial compliance, and 0 indicates complete compliance. This also applies when a specified group of people, a specified plan, or a specified time period cannot be excluded. Take the larger value; The target slot bundle is determined according to the following formula: ; in, This indicates the final output target slot bundle; Represents the candidate set The first in One candidate slot bundle; , , and These represent the weights for execution fidelity, account competition overlap, candidate uncertainty, and exclusion slot violation, respectively, ranging from 0 to 1. The sum of the four weights is 1. Preferred settings are configured according to the ad account security policy, delivery goals, and user permissions. Accounts prioritizing the accuracy of delivery intent can... Set to 0.3~0.4; highly competitive accounts can... Set it to 0.2~0.3. Accounts with many ambiguities can... Setting it to 0.2~0.3 excludes accounts with strict intent requirements. Set to 0.1~0.2; because Both the corresponding weights are dimensionless values. The items in the parentheses can be compared. The left and right sides of the formula conform to the candidate selection relationship. like If the execution fidelity does not meet the preset conditions, the non-executable reason obtained from the reversible API compilation will be output, and the minimum follow-up item corresponding to the non-executable slot will be generated; if If the account contention overlap value does not meet the preset condition, then slot orthogonalization processing will continue; if If the violation of the excluded slot does not meet the preset conditions, the confirmed slot that conflicts with the excluded slot will be corrected first; the target slot bundle will only be output when the preset conditions are met for execution fidelity, consumption uncertainty, account contention and the exclusion slot satisfaction. The target slot bundle includes at least the target audience slot, conversion target slot, budget slot, geographic slot, time slot, audience slot, creative slot, bid slot, exclusion slot, platform executable parameter reference, reverse semantic restoration result, account competition correction result, and follow-up question processing record. Through the above target slot bundle, natural language advertising intent can be converted into structured input that can be received by the subsequent advertising campaign creation process.
[0033] Step S6: Execution and output of advertising based on target slot bundles.
[0034] Target slot bundle output by S5 As input, the affirmative slots, excluded slots, orthogonalization correction records, and platform executable parameter references are converted into real advertising delivery tasks. This step completes the implementation configuration of the planning layer, unit layer, creative layer, and conversion path, and outputs advertising delivery results that can be launched or have been launched.
[0035] S61. Deployment tier breakdown: For target slot bundle The campaign is divided into different tiers. Based on the scope of each slot, the target audience, marketing objectives, budget, campaign duration, and account-level exclusions are placed in the campaign tier. The region, audience, campaign time, bidding method, budget release method, and account competition avoidance configuration are placed in the ad group tier. The creatives, landing pages, conversion components, and creative disabling conditions are placed in the creatives tier. Slots whose scope covers the entire advertising campaign are placed in the campaign tier, slots whose scope covers a single ad group are placed in the ad group tier, and slots whose scope covers a single creative are placed in the creatives tier. like If a product or conversion goal is included, it is split into multiple advertising targets based on its scope. Each advertising target inherits and shares the slots, and only covers the local slots related to its own scope, so as to avoid the local conditions of a certain store, product or conversion goal being incorrectly applied to other advertising targets. S62. Execution of regional slots and time slots: The system performs execution-oriented processing on geographic slots and time slots. For geographic slots such as store perimeter, service radius, and local in-store area, it reads the store address or latitude and longitude and converts them into geofences, store packages, business districts, or administrative region sets supported by the target advertising platform. The conversion from service radius to geofence can be achieved through polar coordinate conversion. A circular area is generated with the store's latitude and longitude as the center and the service radius as the radius. If the platform does not support circular geofences, the circular area is approximated as a polygonal geofence composed of multiple polygon vertices, with the number of vertices set to 8 to 32. If the target platform does not support the precise geographic range expressed in natural language, an executable geographic field that can cover the advertising intent is selected, and the substitution relationship is written into the reverse semantic reconstruction result. For non-service areas, areas already covered by plans, or regions that need to be excluded, convert them into region exclusion configurations, or implement region exclusion by splitting unit layer objects; for time slots such as weekends, business hours, pre-live broadcast warm-up, live broadcast exclusion, and no form collection during non-business hours, convert them into platform advertising calendars; if the platform does not support complex time slot combinations, split them into multiple unit layer objects so that different advertising time slots correspond to different budgets, bids, and exclusion configurations. S63. Targeting configuration for audience slots and exclusion slots: Targeting is configured for audience slots and exclusion slots. For natural language audience slots such as young families, high-intent users, live-streaming interactive users, potential customers near stores, and people who have recently viewed product pages, these are converted into audience tags, interest tags, interaction behaviors, geographic relationships, CRM status, or remarketing audiences supported by the platform. The conversion can be achieved through semantic mapping. A semantic mapping table is established between natural language audience descriptions and platform audience tags. The semantic features of the input audience description are extracted, and similarity is calculated between the semantic features of the tags in the mapping table. One or more platform tags with the highest similarity are selected as the conversion result. For exclusion slots such as existing customers, users who have left their information, users who have made a purchase, people who have made reservations and interacted during the live broadcast, and people who have already been covered by the plan, priority is given to generating exclusion audience configurations; if the target platform does not support direct exclusion, approximate exclusion is achieved by narrowing the certain audience, selecting the set of people who did not hit the exclusion target, splitting the unit layer object, or lowering the priority of the relevant candidate slots, and the reasons for the exclusion not being fully executed are recorded. S64, Budget Slots, Bidding Slots, and Orthogonalization Modification Execution: The orthogonalization correction records for budget slots, bidding slots, and account competition are processed for execution. For budget or bidding semantics such as daily budget, total budget, cost control, small-scale trial, lead generation, and volume expansion, they are converted into budget fields, bidding methods, cost control fields, and budget release methods supported by the target advertising platform; budget slots are prioritized to be bound to the campaign layer, and bidding methods are prioritized to be bound to the ad group layer. If S4 generates orthogonalization corrections such as off-peak delivery, geographic shrinkage, audience exclusion, creative replacement, or conversion target isolation, the budget and bidding configuration will be reallocated according to the corrected available traffic range to avoid the excluded time periods, audiences, or regions still occupying the budget; for orthogonalization slots generated by avoiding existing plans, the avoidance plan identifier, avoidance dimension, and corrected executable configuration will be retained. S65. Material slots are linked to the conversion path: Bind the creative slots and conversion paths for ad placement; for short video customer acquisition plans, convert creative slots into ad-supported short video creative tags and bind them to form, private message, phone consultation or landing page conversion components; for live streaming appointment plans, bind creative slots, live streaming room tags and appointment components; for in-store conversion or car test drive lead plans, bind creatives, store pages, map pages, test drive appointment forms and lead feedback fields. If the exclusion slot includes rejecting old materials, live stream materials, materials highlighting certain selling points, or materials shared with existing plans, then filtering is first performed based on material tags, material review status, material history binding relationships, and material semantic categories before generating material layer objects. Material semantic filtering can be achieved through material semantic vectors. All materials in the account are pre-annotated semantically and semantic vectors are generated. The filtering conditions are converted into corresponding semantic vectors, similarity is calculated, and materials exceeding the preset threshold are filtered out. If the material review status is unavailable for delivery, the corresponding delivery object is retained but the material is not launched, or the minimum follow-up questions related to the material slot are triggered. S66. Interface parameter generation and consistency verification before submission: After the ad placement object is created, the interface parameters are generated according to the creation order of the target advertising platform. Usually, the conversion asset is confirmed or created first, then the campaign layer object is created, then the unit layer object is created, then the creative layer object, landing page object and conversion component are bound, and finally the launch status is set. Each interface call records the campaign identifier, unit identifier, creative identifier, review status and launch status returned by the platform. If the target advertising platform requires the creative to be approved before launch, the created ad placement object will first be kept in the pending launch or paused state. The launch interface will be called after the review status meets the platform requirements. Before submitting the API, a consistency check is performed on the ad delivery targets. This check includes verifying whether the budget field is compatible with the bidding method, whether the geographic range is empty, whether audience targeting is completely offset by excluded slots, whether the creative type matches the conversion goal, whether the landing page matches the target audience, whether the conversion component matches the marketing goal, whether excluded and confirmed audiences simultaneously hit the same targetable set, and whether the account's competition avoidance configuration still meets the low competition conditions. If inconsistencies are found, the problem is traced to... The corresponding slot in the code, and re-execute reversible API compilation, account race orthogonalization or minimal follow-up based on the problem type; S67. Create an interface or initiate an interface call: After user confirmation, the target advertising platform's creation or launch interface is invoked. If the platform returns success for creation or launch, the advertising delivery results are output. The advertising delivery results include at least the advertising plan identifier, advertising unit identifier, creative identifier, delivery status, review status, confirmed effective slots, excluded effective slots, account competition correction records, platform return information, and reasons for incompletely effective slots. For exclusion slots that have already taken effect, record how they are implemented on the platform, such as excluding specific audiences, prohibiting certain time periods, disabling certain creatives, removing certain regions, or avoiding existing plans. For orthogonalization correction slots, record the slot before correction, the slot after correction, and the existing plans that are being avoided. S68, Anomaly Location, Slot Writeback and Redeployment: If the platform returns errors such as creation failure, field error, insufficient permissions, budget limitation, unavailable creatives, narrow targeting, missing conversion components, or review exceptions, then reverse targeting will be performed based on the fields returned by the platform. The corresponding slot in the [processing context] is used to write the error type into the execution status of that slot; For errors related to non-executable fields, re-execute the reversible API compilation and repair in S2; for errors related to account competition or failure to implement exclusion constraints, re-execute the slot orthogonalization process in S4; for errors caused by candidate ambiguity, re-execute the minimal follow-up process in S3. After repair, regenerate the ad delivery object and re-enter the submission process. Through the above processing, the target slot bundle Not only does it serve as a structured parameter output, but it also directly drives the creation of campaigns, configuration of ad units, binding of creatives, binding of conversion components, implementation of exclusion conditions, confirmation of campaign launch, and output of results in real advertising campaigns. As a result, natural language advertising intent can be converted into actual executable, verifiable, and traceable advertising campaign tasks in the advertising platform.
[0036] Through the detailed description of the above embodiments, the natural language advertising intent slot generation method of the present invention, by constructing a complete technical process including natural language parsing, reversible API compilation, consumption distribution divergence judgment, slot orthogonalization, target slot decision-making, and delivery execution, realizes end-to-end conversion from natural language advertising requests to executable tasks on the advertising platform. The present invention simultaneously identifies and handles conflicts between positive and negative intents during the slot generation stage, ensures the execution fidelity of advertising intent through a bidirectional compilation mechanism, optimizes user follow-up strategies based on actual consumption impact, and introduces in-account traffic competition analysis for slot correction. The final output target slot bundle can directly drive the creation, configuration, and launch of advertising plans, providing a complete technical implementation solution for the automation and intelligent processing of advertising delivery.
[0037] The preset parameters in the above formulas shall be set by those skilled in the art according to the actual situation.
[0038] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.
[0039] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0040] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0041] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0042] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0043] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0044] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for generating natural language delivery intent slots, characterized in that, The method flow is as follows: Step S1: Receive a natural language delivery request input by the user, perform sentence segmentation, phrase boundary recognition, action word recognition, and dominance relationship recognition on the natural language delivery request, and extract delivery semantic fragments and generate a candidate slot bundle containing positive slots and excluded slots when the request type belongs to plan creation, copy creation, or plan variant creation. Step S2: Read the interface constraints of the target advertising platform, compile the candidate slot bundle into a set of parameters that the platform can accept, and reverse the parameter set to obtain the reverse semantic slot bundle. Determine the execution fidelity based on the semantic difference between the candidate slot bundle and the reverse semantic slot bundle. Step S3: Predict the consumption distribution for candidate slot bundles that meet the execution fidelity conditions, and determine whether to generate a minimum follow-up based on the consumption distribution divergence between different candidate slot bundles. Step S4: Obtain the set of existing advertising campaigns in the account, convert the candidate slot bundles and existing advertising campaigns into traffic occupancy vectors respectively, and perform slot orthogonalization processing on the candidate slot bundles based on the competition overlap value; Step S5: Determine and output the target slot bundle from the candidate slot bundles based on execution fidelity, account contention overlap value, candidate uncertainty, and excluded slot violation degree; Step S6: Based on the target slot bundle, perform the following: split the delivery level, execute the regional slots and time slots, configure the audience slots and exclusion slots for delivery targeting, execute the budget slots and bid slots, bind the creative slots and conversion links, and perform consistency verification of interface parameters before submission. After user confirmation, call the target advertising platform's creation interface or launch interface to output the advertising delivery results.
2. The method for generating natural language delivery intent slots according to claim 1, characterized in that, The semantic fragments of the campaign include the target audience, campaign objective, budget, region, time, demographics, creative materials, bidding method, conversion components, landing page, and exclusions. When there are multiple feasible interpretations for the same semantic fragment, candidate slot items are generated separately and combined into a candidate slot bundle according to slot type, scope of application, and semantic compatibility.
3. The method for generating natural language delivery intent slots according to claim 2, characterized in that, The exclusion slots include exclusion dimensions, exclusion objects, scope of application, triggering text fragments, execution operators, and priorities. For explicit negation expressions, the exclusion scope is determined according to the dependency relationship between the negation word and the dominated object. For indirect exclusion expressions, the object to be avoided is identified, and competitive exclusion slots, crowd exclusion slots, or time-period exclusion slots are generated according to the type of the object to be avoided.
4. The method for generating natural language delivery intent slots according to claim 3, characterized in that, When affirmative slots and exclusion slots conflict on the same dimension, they are resolved according to source priority. Specifically, exclusion slots explicitly given by the user take precedence over automatically supplemented affirmative slots, platform mandatory interface constraints take precedence over normal delivery preferences, and affirmative slots explicitly given by the user take precedence over default supplemented slots. When conflicts cannot be resolved by priority, multiple candidate slot bundles are retained for consumption distribution divergence judgment.
5. The method for generating natural language delivery intent slots according to claim 1, characterized in that, The interface constraints are determined by the interface field description, account available capabilities, field enumeration returned by the platform, field mandatory relationships, field mutual exclusion relationships, and interface pre-validation results; the forward compilation includes field naming conversion, field type conversion, enumeration value conversion, mandatory field completion, mutual exclusion field deletion, approximate replacement of fields not supported by the platform, and execution conversion of excluded slots.
6. The method for generating natural language delivery intent slots according to claim 5, characterized in that, The reverse restoration includes restoring the promotion target, goal, budget, region, audience, time, creative, conversion component and exclusion conditions based on the reverse correspondence between platform fields and delivery semantics; when a natural language slot is deleted, replaced by a default field or replaced by a similar field in the forward compilation, it is recorded as missing, downgraded or semantically offset in the reverse semantic slot bundle, and the execution fidelity is determined based on the missing, downgraded or semantically offset.
7. The method for generating natural language delivery intent slots according to claim 6, characterized in that, When the execution fidelity does not meet the preset fidelity conditions, the candidate slot bundle is repaired. The repair includes splitting the non-executable slot into multiple executable sub-slots, converting semantic audiences not supported by the platform into executable audience combinations, converting geographical radii that cannot be directly expressed into geofences or administrative region combinations supported by the platform, or splitting incompatible conversion targets and material types into different candidate slot bundles. The repaired candidate slot bundles are then re-compiled and reconstructed.
8. The method for generating natural language delivery intent slots according to claim 1, characterized in that, When predicting consumption distribution, geographic units, time units, audience units, creative categories, and conversion goals are collectively determined as consumption units. Based on the estimated traffic returned by the target platform's traffic estimation interface, the account's recent average consumption intensity in the same or similar consumption units, candidate bid levels, budget release methods, and creative availability status, the consumption intensity of candidate slot bundles in each consumption unit is determined and normalized to obtain the consumption distribution.
9. A method for generating natural language delivery intent slots according to claim 8, characterized in that, When the overall consumption distribution divergence of the candidate slot bundle set meets the preset low divergence condition, the candidate slot bundle with higher fidelity and lower account competition risk is automatically supplemented; when the overall consumption distribution divergence does not meet the preset low divergence condition, a set of follow-up questions is generated for the undetermined slots, and the follow-up questions are selected according to the divergence reduction benefit corresponding to the unit interaction cost, until the candidate slot bundle set meets the preset low divergence condition, or the candidate slot bundles are further distinguished by subsequent account competition processing.
10. The method for generating natural language delivery intent slots according to claim 1, characterized in that, Step S6 includes: splitting the target slot bundle into advertising targets corresponding to different delivery levels based on the scope of each slot in the target slot bundle; performing platform execution processing on the split advertising targets to convert the regional slots, time slots, audience slots, exclusion slots, budget slots, bidding slots, creative slots, conversion component slots, and account competition correction records into delivery configurations supported by the target advertising platform; generating interface parameters according to the object creation order of the target advertising platform, and verifying the consistency between delivery configurations and the executable status of exclusion slots and account competition avoidance configurations before submission; if the verification result does not meet the submission conditions, the verification exception is located to the corresponding slot in the target slot bundle, and re-execute reversible API compilation, slot orthogonalization processing, or minimal follow-up processing according to the exception type; if the verification result meets the submission conditions, the creation interface or launch interface of the target advertising platform is called after user confirmation, and the advertising delivery result including platform object identifier, delivery status, review status, actual effective slots, account competition correction records, platform return information, and reasons for incompletely effective slots is output.
