Causal evaluation method and device for advertisement putting and computer readable storage medium

Abstract

The invention discloses a causal evaluation method for advertisement putting. The method comprises the following steps: acquiring a plurality of data sets of advertisements put on a take-out platform by merchants operating on the take-out platform; inputting the plurality of data sets into a causal graph mining model set to obtain a first candidate causal graph set; filtering the first candidate causal graph set based on an expert knowledge base to obtain a second candidate causal graph set; obtaining the weight of the path according to the relationship between the node and the path of each second candidate causal graph in the second candidate causal graph set, and aggregating the plurality of second candidate causal graphs based on the weight of the path to obtain a causal assessment large graph; and based on the causal evaluation large graph, evaluating the causal relationship between the advertising parameters and the effect indexes of the merchant advertising. According to the method, the accuracy of advertisement effect evaluation can be improved through accurate causal relationship analysis, the advertisement putting efficiency can be improved, and the putting strategy can be optimized.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to a method, device and computer-readable storage medium for causal evaluation of advertising placement. Background Technology

[0002] The rapid development of the internet has driven the growth in the number of users of food delivery platforms and the demand for online advertising. Advertising can significantly enhance the accuracy of food delivery service recommendations and improve the promotional effect.

[0003] In the food delivery industry, the effectiveness of advertising is often a key focus for merchants. However, the factors influencing advertising effectiveness are varied. Current technologies for analyzing the effectiveness of merchant advertising still rely on simple attribution methods, evaluating merchants' advertising investment based on direct reactions after advertising campaigns. For example, a decrease in order volume is attributed to insufficient advertising investment. This simplistic attribution method often fails to delve into the true reasons behind advertising effectiveness, thus making it difficult to provide merchants with accurate decision-making information. Summary of the Invention

[0004] This application provides a method, apparatus, device, and computer-readable storage medium for causal evaluation of advertising placement. The technical solution is as follows:

[0005] In a first aspect, embodiments of this application provide a method for causal evaluation of advertising placement. The method includes: acquiring multiple data sets of merchants operating on a food delivery platform who have placed advertisements on the platform, wherein each data set includes a merchant's placement parameters and performance indicators within a certain time period; inputting the multiple data sets into a causal graph mining model set to obtain a first candidate causal graph set, wherein the graph mining model set includes multiple causal graph mining models; filtering the first candidate causal graph set based on an expert knowledge base to obtain a second candidate causal graph set; obtaining the path weights based on the relationship between nodes and paths in each second candidate causal graph in the second candidate causal graph set; aggregating multiple second candidate causal graphs based on the path weights to obtain a large causal evaluation graph; and evaluating the causal relationship between the merchant's advertising placement parameters and performance indicators based on the large causal evaluation graph.

[0006] In one possible implementation, the method further includes, after the step of acquiring multiple data sets of merchants operating on the food delivery platform and placing advertisements on the platform, the method further includes processing the data in the data sets, including data cleaning, anomaly identification, and null value imputation.

[0007] In one possible implementation, the method further includes: in the step of inputting multiple data sets into a causal graph mining model set to obtain a first candidate causal graph set: for N data sets, they are respectively input into M causal graph mining models in the causal graph mining model set to obtain a first candidate causal graph set, which includes N*M first candidate causal graphs.

[0008] In one possible implementation, the method further includes: filtering the first candidate causal graph set based on an expert knowledge base to obtain a second candidate causal graph set, including: generating a reference causal graph based on the description of the causal relationship between the delivery parameters and performance indicators in the expert knowledge base; matching each first candidate causal graph in the first candidate causal graph set with the reference causal graph to obtain a matching score; and filtering the first candidate causal graph set according to a set matching score threshold to obtain a second candidate causal graph set.

[0009] In one possible implementation, the method further includes: obtaining the weight of a path based on the relationship between nodes and paths in each second candidate causal graph in the second candidate causal graph set; aggregating multiple second candidate causal graphs based on the path weight to obtain a large causal evaluation graph; including: splitting the links of each second candidate causal graph in the second candidate causal graph set, wherein the links include node pairs and paths between node pairs; obtaining the weight of the link in the second candidate causal graph set based on the frequency of the link's occurrence in the second candidate causal graph; aggregating the second candidate causal graphs in the second candidate causal graph set and simplifying them based on the weight to obtain a large causal evaluation graph.

[0010] In one possible implementation, the method further includes: after obtaining the weight of a path based on the relationship between nodes and paths in each second candidate causal graph in the second candidate causal graph set, and aggregating multiple second candidate causal graphs based on the path weight to obtain a causal evaluation graph, the method further includes: processing the causal evaluation graph using an expert knowledge base based on the placement parameters and performance indicators to be determined to obtain a causal evaluation subgraph; then, evaluating the causal relationship between the placement parameters and performance indicators of the merchant's advertising based on the causal evaluation graph and / or the causal evaluation subgraph.

[0011] In one possible implementation, the method further includes: processing the causal evaluation graph based on the delivery parameters and effect indicators to be determined using an expert knowledge base to obtain a causal evaluation subgraph, including: determining relevant causal rules in the expert knowledge base based on the delivery parameters and effect indicators to be determined; identifying delivery parameters and intermediate variable factors affecting the effect indicator target according to the causal rules, and layering the causal evaluation graph; for each layer of the causal evaluation graph, removing nodes and / or paths that do not conform to the causal rules to obtain a causal evaluation subgraph.

[0012] In one possible implementation, the method further includes: processing the large causal evaluation graph based on the delivery parameters and effect indicators to be determined using an expert knowledge base to obtain a causal evaluation subgraph; the method further includes: identifying the circular dependency paths in the causal evaluation subgraph; simplifying the circular dependency paths so that each node in the causal evaluation subgraph forms a single dependency relationship; and updating the weights of each path in the simplified causal evaluation subgraph.

[0013] Secondly, embodiments of this application provide a server, which includes a processor and a memory. The memory stores at least one piece of program code, which is loaded and executed by the processor to enable the server to implement the causal evaluation method for advertising placement as described in the first aspect and any implementation thereof.

[0014] Thirdly, embodiments of this application also provide a computer program or computer program product that stores at least one computer instruction, which is loaded and executed by a processor to enable the computer to implement any of the above-mentioned causal evaluation methods for advertising placement.

[0015] Fourthly, a computer-readable storage medium is also provided, which stores at least one piece of program code, which is loaded and executed by a processor to enable a computer to implement any of the above-described methods for causal evaluation of advertising placement.

[0016] The technical solution provided in this application provides at least the following beneficial effects: The technical solution uses a cause-effect graph evaluation method, which can obtain the complex causal relationships behind advertising effectiveness through data analysis, providing merchants with more accurate data support and decision-making information. This improves advertising efficiency, enhances the interpretability of advertising strategies, reduces decision-making difficulty and risk, and thus achieves more efficient and precise advertising promotion in fierce market competition. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the implementation environment of a causal evaluation method for advertising placement provided in an embodiment of this application;

[0019] Figure 2 This is a schematic flowchart illustrating a causal evaluation method for advertising placement provided in an embodiment of this application;

[0020] Figure 3 This is a schematic diagram of the structure of a server provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0022] Figure 1 This is a schematic diagram illustrating the implementation environment of a causal evaluation method for advertising placement provided in this application embodiment, such as... Figure 1 As shown, the implementation environment includes: electronic device 101 and server 102.

[0023] Electronic device 101 may be at least one of a smartphone, game console, desktop computer, tablet computer, e-book reader, MP3 (Moving Picture Experts Group Audio Layer III) player, MP4 (Moving Picture Experts Group Audio Layer IV) player, and laptop computer. Electronic device 101 is used to acquire first location information and target content, and send the first location information and target content to server 102.

[0024] Electronic device 101 can refer to one of a plurality of electronic devices. This embodiment uses electronic device 101 as an example only. Those skilled in the art will know that the number of electronic devices 101 can be more or less. For example, there may be only one electronic device 101, or there may be dozens or hundreds of electronic devices 101, or even more. This application embodiment does not limit the number or type of electronic devices.

[0025] Server 102 can be a single server, a server cluster consisting of multiple servers, or any of the following: a cloud computing platform or a virtualization center. This embodiment of the application does not limit this. Server 102 communicates with electronic device 101 via a wired or wireless network. Server 102 is used to generate multiple candidate acquisition schemes based on the first location information and the target content, and send the multiple candidate acquisition schemes to electronic device 101. Of course, server 102 may also have other functions, which are not limited in this embodiment of the application.

[0026] Based on the above implementation environment, this application provides a causal evaluation method for advertising placement, in order to... Figure 2 The schematic flowchart shown in this application embodiment illustrates a causal evaluation method for advertising placement. This method can be implemented by... Figure 1 The interaction between electronic device 101 and server 102 is described. For example... Figure 2 As shown, the method may include the following steps:

[0027] In step 201, multiple data sets of merchants operating on the food delivery platform and placing advertisements on the platform are obtained. Each data set includes the advertising parameters and performance indicators of a merchant within a certain time period.

[0028] In response to merchants selecting advertising products on the merchant interface of the food delivery platform, the system retrieves relevant data sets. Merchants choose advertising products suitable for their operational food delivery stores and can configure or select advertising information such as the advertising time range, parameters, and expected performance metrics. Upon confirmation from the merchant, the system uploads the merchant's user ID and / or store ID, along with the selected advertising products and information, to the server. Based on this information, the server retrieves one or more relevant data sets, each containing the merchant's advertising parameters and performance metrics for a specific time period.

[0029] For example, a merchant selects the "Afternoon Tea Promotion" advertising product on the merchant's interface and configures the following information: promotion period, bid, and expected performance metrics. After the merchant completes the configuration and confirms the operation, the server receives the above information and, based on this information, retrieves the following datasets for each merchant: Data Set 1: Time-based Competitive Advertising Dataset, containing the merchant's bid, ad impressions, click-through rate, competing brands, and product types during that time period. This dataset can be used to analyze the advertising performance of other merchants during that time period, including bids, ad placement, and click-through rate. Data Set 2: User Behavior Dataset, containing user search and browsing habits during the afternoon tea period, user clicks and conversions on afternoon tea-related ads, and user preferences. This dataset can be used to analyze user behavior patterns during the afternoon tea period to optimize advertising effectiveness. Data Set 3: Time-based Sales Dataset, containing the merchant's historical order volume during the afternoon tea period, changes in order volume during the advertising campaign, and the distribution of orders from new and returning customers during the afternoon tea period. This dataset can be used to compare and analyze sales data before and after the advertising campaign to evaluate advertising effectiveness.

[0030] This step allows merchants to select advertising products according to their own needs, flexibly build corresponding datasets, improve the personalization and adaptability of data processing, and thus provide merchants with more accurate promotion strategies and optimization suggestions.

[0031] Optionally, after obtaining the dataset, the data in the dataset can be processed, including but not limited to data cleaning, anomaly detection, and null value imputation.

[0032] Optionally, the constructed dataset can be divided into high-quality datasets and unprocessed datasets. Distinguishing data quality ensures the relevance of subsequent processing steps and improves data processing efficiency. Preprocessing of the unprocessed dataset can include: removing irrelevant and erroneous data through data cleaning; detecting and processing outliers through anomaly identification, such as removing them or replacing them with boundary values. Removing extreme values ​​from the dataset avoids adverse effects on model training and result prediction. This can also include null value imputation, using methods such as mean imputation, median imputation, and prediction imputation. By filling in missing values ​​in the dataset, data integrity and the accuracy of subsequent analysis are ensured.

[0033] In step 202, multiple datasets are input into a set of causal graph mining models to obtain a first candidate causal graph set, wherein the set of graph mining models includes multiple causal graph mining models.

[0034] Graph mining is a branch of data mining that focuses on extracting useful information and knowledge from graphs. A causal graph is a graphical representation that shows the causal relationships between variables. The causal graph mining models in this application's collection include, but are not limited to: Structural Equation Modeling (SEM), Bayesian Networks, the Peter-Clark Algorithm (PC), the Fast Causal Inference Algorithm (FCI), the Greedy Equivalence Search (GES), Direct Linear Non-Gaussian Causal Models (DirectLiNGAM), and Directed Acyclic Graphs with No Tears (DAGWithNoTears), etc., and this application makes no limitation thereto.

[0035] Optionally, each dataset can be input into a different causal graph to obtain a first candidate causal graph set. For example, for N datasets, each dataset can be input into one of the M causal graph mining models in the causal graph mining model set to obtain a first candidate causal graph set, which includes N*M first candidate causal graphs.

[0036] Optionally, after determining the set of graph mining models, standardization can be performed on the data inputs and causal graph outputs.

[0037] Regarding data input, standardized application programming interfaces (APIs) unify the data format of the dataset into metadata. Cause-effect graph mining models can be accessed via APIs. By designing a unified input interface for different cause-effect graph models, regardless of the source of the input data, format conversion and input can be performed through this interface, ensuring that all APIs adhere to the same input / output specifications. Metadata refers to the unified definition of variable metadata standards and encoding methods, including variable names, types, descriptions, and possible value ranges.

[0038] Regarding data output, the resulting causal graph will be standardized and integrated, which may include the following steps:

[0039] Define a unified standard: Define a unified standard format for cause-effect graphs, including the basic elements of cause-effect graphs, such as nodes, edges, weights, etc., as well as any necessary metadata information, such as parameter naming, etc.

[0040] Model output conversion: Convert the outputs of different cause-effect graph models into a unified standard format, including renaming variables, adjusting parameter representations, and standardizing edge directions and types.

[0041] The model output transformation may include:

[0042] The process involves analyzing the causal graph and standardizing the handling of nodes and edges to ensure that all model outputs have consistent definitions for nodes and edges. For example, nodes may represent variables, and edges may represent causal relationships between variables.

[0043] Standardize the attributes of nodes and edges, for example, by unifying the representation of weights and directions, such that all models use the same symbols to represent variables and directions.

[0044] Transform weights and parameters: Convert model-specific weights and parameters into standardized values ​​to facilitate comparison and analysis between different models.

[0045] Formatted output: Format the converted data into an easy-to-understand format, such as JSON, XML, or a specific chart format.

[0046] The technical benefits of unifying data input and output formats include improved performance in the following areas: interoperability: standardized output allows results from different models to be interoperable and compared; readability: formatted output is easier to process in subsequent steps; integration: standardized output can be more easily integrated into other systems, such as automated decision support systems; and reusability: a unified output format improves the reusability of data in different analyses.

[0047] In step 203, the first candidate causal graph set is filtered based on the expert knowledge base to obtain the second candidate causal graph set. By filtering the first candidate causal graph set based on the expert knowledge base, the accuracy of causal inference can be effectively improved, hypotheses that do not conform to prior knowledge can be eliminated, ensuring that the causal graph is consistent with the actual situation, and enhancing the explanatory power and application value of the model.

[0048] Optionally, step 203 may include:

[0049] A reference causal graph is generated based on descriptions of causal relationships between deployment parameters and performance metrics from an expert knowledge base. This step first extracts detailed descriptions of the causal relationships between deployment parameters and performance metrics from the expert knowledge base, forming a structured reference causal graph g'. These descriptions, based on expert experience and domain knowledge, ensure the accuracy and reliability of the reference causal graph g'. The reference causal graph includes all important causal relationships that experts believe exist in the dataset, such as direct causal effects, moderating effects, and time dependencies.

[0050] Each first candidate causal graph in the first candidate causal graph set is matched with the reference causal graph to obtain a matching score. In this step, each causal graph g in the first candidate causal graph set is matched with the generated reference causal graph g'. This process may include the following steps: parsing the nodes and edges in each first candidate causal graph and parsing the nodes and edges in each reference causal graph, comparing the corresponding elements. The matching relationship between each first candidate causal graph and the nodes and edges in each reference causal graph is calculated, and a quality score q(g) is calculated for each first candidate causal graph to evaluate the matching degree.

[0051] Based on a set matching score threshold, the first candidate causal graph set is filtered to obtain a second candidate causal graph set. In this step, for each first candidate causal graph in the set, its quality score q(g) is compared with a threshold τ. If the quality score q(g) is greater than or equal to the threshold τ, the first candidate causal graph is considered to have high confidence and is retained in the second candidate causal graph set. If the quality score is lower than the threshold τ, the causal graph is considered to have insufficient confidence and may introduce erroneous causal relationships, therefore it is not retained in the second candidate causal graph set.

[0052] The above steps ensure that the causal graphs in the second candidate causal graph set are not only statistically significant but also consistent with the expert's prior knowledge, thereby improving the effectiveness and credibility of the final causal graph. This filtering process helps reduce the risk of graph contamination and ensures that the final constructed causal graph accurately reflects the true causal relationships in the dataset.

[0053] In step 204, the weight of each path is obtained based on the relationship between nodes and paths in the second candidate causal graph set. Multiple candidate causal graphs are then aggregated based on these weights to obtain a large causal evaluation graph. This step, through weight calculation and aggregation, improves the accuracy and confidence of causal relationships in the large causal evaluation graph, ensuring that the final graph comprehensively reflects the consensus of multiple causal graphs, providing more reliable data support for decision-making, and enhancing the practicality and effectiveness of causal inference.

[0054] Optionally, step 204 may include:

[0055] The links of each second candidate causal graph in the set of second candidate causal graphs are decomposed, where each link includes node pairs and the paths between node pairs. For each second candidate causal graph g, all links are identified and extracted. Each link can be defined as being decomposed into...<source,target,support> The causal sequence is a triple sequence, for example, the link <bid, bid competitiveness, 1>. Here, `source` refers to the starting point of the causal relationship, i.e., the variable or node believed to lead to certain effects; `source` is the bid. `target` is the opposite of `source`; `target` refers to the ending point of the causal relationship, i.e., the variable or node affected by `source`; in the above example, `target` is "bid competitiveness". `support` is a numerical value representing the support or confidence of the causal relationship between `source` and `target`. In this example, `support` is 1, indicating that the causal relationship between bid and bid competitiveness occurs once in this second candidate causal graph `g`.

[0056] The weight of a link within the set of second-candidate causal graphs is obtained based on its frequency of occurrence. This can include the following process: for each link, its frequency of occurrence across all second-candidate causal graphs is counted, and the support value in the Triple sequence is updated. The weight of a link is determined by its support value in the set of second-candidate causal graphs; the higher the value, the greater the weight.

[0057] The second candidate causal graphs in the second candidate causal graph set are aggregated and simplified based on the weight to obtain a large causal evaluation graph.

[0058] Optionally, the aggregation step may include: merging the Triple sequences: merging the Triple sequences of the second candidate causal graph set and the reference causal graph g' based on expert experience respectively to obtain the Triple sequence set of all second candidate causal graphs and the Triple sequence set of the reference causal graph, and then merging these two sets.

[0059] Optionally, simplification based on this weight may include: setting simplification principles, and simplifying based on these principles and the weight. These simplification principles may include one or more of the following: expert experience weight principle, maximum path principle, shortest path principle, causal directionality principle, etc.

[0060] Specifically, the expert experience weighting principle applies: Additional weights are assigned to each Triple based on expert experience. During the simplification process, causal relationships deemed more important by experts are prioritized.

[0061] In a preferred embodiment, during the merging of the sets of Triple sequences of all second candidate causal graphs and the sets of Triple sequences of the reference causal graph, the weight ratio coefficient of the Triple sequences of the reference causal graph is set to be greater than the weight ratio coefficient of the Triple sequences of the second candidate causal graphs. For example, during the merging process, the weight ratio coefficient of the support value of the Triple sequences of the second candidate causal graphs is set to 1, and the weight ratio coefficient of the support value of the Triple sequences of the reference causal graph is set to 10 before merging.

[0062] In another preferred embodiment, an identifier is added to the Triple sequence of the reference causal graph, which indicates that the Triple sequence is preserved during simplification.

[0063] The beneficial effects of adopting the expert experience weighting principle include: Enhancing the priority of expert knowledge: By assigning higher weights to the Triples in the reference causal graph, it ensures that causal relationships deemed important by experts receive greater attention and retention during the aggregation process, thereby improving the accuracy and influence of expert knowledge in the causal evaluation graph. Reducing misjudgments and noise: Expert experience can help identify and retain key causal relationships that data-driven methods might overlook or misjudge, reducing the impact of misjudgments and noise on the final causal graph. Improving model interpretability: Combining expert experience makes the causal evaluation graph more aligned with business logic, improving the model's interpretability and enabling non-technical users to understand and accept the model's output.

[0064] The maximum path principle states that when multiple causal paths point to the same target node, only the path with the highest total path weight is retained. This helps identify the most important causal paths and simplifies the graph structure. The shortest path principle states that when multiple causal paths have similar weights, the shortest path is retained first. The causal directionality principle prioritizes retaining triples that conform to a known causal direction, even if their weights are not the highest. This helps maintain the logical consistency of causal relationships.

[0065] Constructing a causal assessment graph may include: Integrating key causal relationships: Integrating simplified key causal relationships into a large causal graph to form the causal assessment graph. Labeling key nodes: Labeling key nodes in the causal assessment graph; these nodes may be important mediating or outcome variables. Validation and adjustment: Validating the causal assessment graph to ensure its logical consistency and adjusting it as needed.

[0066] Through the above steps, the set of second candidate causal graphs can be aggregated into a large causal evaluation graph G'. This graph comprehensively considers the node and path relationships of multiple causal graphs and uses weights to reflect the confidence level of different causal relationships. This aggregation method not only improves the accuracy of causal relationships but also provides a solid foundation for subsequent causal inference and analysis.

[0067] In step 205, based on the aforementioned causal evaluation graph, the causal relationship between the merchant's advertising placement parameters and performance metrics is evaluated. The causal evaluation graph constructed based on the above steps can provide a detailed assessment of the causal relationship between the merchant's advertising placement parameters and performance metrics, and offer precise optimization suggestions to the merchant through in-depth analysis of the interactions between each parameter and performance metric. For example, the causal graph can analyze the correlation between parameters such as the critical period of advertising placement, target audience, and creative content, and key performance metrics such as click-through rate, conversion rate, and customer acquisition cost. By precisely quantifying these causal relationships, it reveals which parameters have a greater impact on advertising effectiveness.

[0068] In the context of online food delivery platforms, the technical effects of this solution include, but are not limited to:

[0069] By constructing a comprehensive causal assessment graph, we can provide merchants with the following advertising evaluation diagnoses and recommendations:

[0070] Causal Path Analysis of Advertising Effectiveness: Utilizing a causal evaluation graph, this study analyzes the direct and indirect causal relationships between advertising campaigns and outcomes such as order volume, brand awareness, and user satisfaction. It identifies which advertising channels, content, or strategies have the greatest positive impact on order volume.

[0071] Customized advertising recommendations: Based on the cause-and-effect graph analysis results, we provide merchants with customized advertising strategies, such as increasing investment in high-conversion channels and optimizing ad content to improve user engagement.

[0072] Merchants are advised to pay attention to intermediate variables that have a significant causal impact on order volume, such as user click-through rate and website dwell time.

[0073] Advertising Budget Optimization: A causal evaluation chart helps businesses understand the relationship between different advertising inputs and outputs, enabling them to allocate advertising budgets more effectively. It identifies inefficient aspects of advertising campaigns and recommends reducing or discontinuing investment in campaigns that have no significant causal impact on order volume.

[0074] This solution, compared to simple attribution methods, introduces a causal evaluation graph, enabling data analysis to uncover the complex causal relationships behind advertising effectiveness. This provides businesses with more accurate data support and decision-making information, improving advertising efficiency, enhancing the interpretability of advertising strategies, and reducing decision-making difficulty and risk, thereby achieving more efficient and precise advertising promotion in a highly competitive market.

[0075] In one alternative implementation, after obtaining the causal evaluation big picture, the method may further include:

[0076] Based on the placement parameters and performance indicators of the target audience whose causal relationship is to be determined, the causal evaluation graph is processed using an expert knowledge base to obtain a causal evaluation subgraph. Based on the causal evaluation graph and / or the causal evaluation subgraph, the causal relationship between the placement parameters and performance indicators of the merchant's advertising is evaluated.

[0077] Setting delivery parameters and performance metrics allows for a clear definition of advertising campaign goals and expected outcomes. This is essentially a problem-driven approach to identify which delivery parameters have a causal relationship with the performance metrics. This method employs a problem-driven heuristic factor hierarchical approach, which improves the focus of the analysis: by hierarchically classifying factors for a specific problem, the optimized solution can more accurately pinpoint the key factors affecting advertising effectiveness, avoiding information overload and distraction that may result from comprehensive analysis; and it enhances decision-making efficiency: the heuristic method simplifies the analysis of complex causal relationships, enabling decision-makers to more quickly identify the most important advertising delivery parameters, thereby improving decision-making efficiency.

[0078] Optionally, the steps of processing the large causal evaluation graph based on the deployment parameters and effect indicators to be determined, using an expert knowledge base, to obtain a causal evaluation subgraph may include:

[0079] Based on the placement parameters and performance metrics that establish causal relationships, the relevant causal rules in the expert knowledge base are determined. First, relevant causal rules need to be extracted from the expert knowledge base, which may include: identifying key factors: based on specific problems in ad placement, such as insufficient exposure or low conversion rates, determine the key placement parameters affecting these performance metrics, such as budget, ad content, and placement time. Extracting causal rules: extract causal rules related to these key factors from the expert knowledge base. For example, if budget is a key factor, extract causal rules between budget and exposure, and the number of times the budget is exceeded. The specific extraction method can also adopt the Triple sequence processing method described in the above implementation, which will not be elaborated here.

[0080] Based on this causal rule, the delivery parameters and intermediate variable factors affecting the performance indicator are identified, and the causal evaluation graph is layered. After determining the causal rule, the next step is to layer the causal evaluation graph. Layering principles are established: factors can be divided into different levels, such as attribution layer, phenomenon layer, transmission layer, and result layer, based on a problem-driven heuristic factor layering algorithm, ensuring that the directional constraints of causal relationships are observed and preserved. Intermediate variables are identified: intermediate variable factors affecting the performance indicator are identified in the causal graph; these factors may include user behavior, market trends, etc. Layered causal graphs are constructed: according to the layering principles, a causal graph is constructed for each layer, where only factors at higher levels are allowed to point to factors at lower levels, ensuring the directional constraints of causal relationships.

[0081] For each layer of the causal evaluation graph, nodes and / or paths that do not conform to the causal rules are removed to obtain a causal evaluation subgraph. Optionally, the Tarjan algorithm, which searches for strongly connected components in a directed graph, can be applied. The Tarjan acyclic graph construction algorithm is used to break cycles in the causal evaluation graph, prioritizing the removal of edges with low support to ensure the accuracy of the causal graph. Nodes and paths that do not conform to the rules are removed: Based on the causal rules and layering principles, nodes and paths that do not conform to the causal directionality constraints are removed; for example, paths pointing from lower-level factors to upper-level factors are not allowed. A causal evaluation subgraph is generated: Through the above steps, a simplified causal evaluation subgraph is finally obtained, which only contains the causal relationships between the delivery parameters and effect indicators related to the causal relationship to be determined.

[0082] By following the steps above, a causal evaluation subgraph for a specific problem can be extracted from a complex causal evaluation graph, providing businesses with more accurate and actionable advertising strategy recommendations.

[0083] Optionally, after processing the causal evaluation subgraph based on the deployment parameters and effect indicators to be determined using an expert knowledge base, the method may further include:

[0084] Circular dependencies in the causal evaluation subgraph are identified. For example, a depth-first search (DFS) or breadth-first search (BFS) algorithm can be used to traverse the subgraph, recording the visit status and order of each node. During the traversal, the entry and exit times of nodes are recorded to determine if circular dependencies exist. A circular dependency exists when the entry time of a node is less than or equal to the entry time of a subsequent node, and the exit time of that subsequent node is infinite. After identifying all circular dependencies, these paths are marked for subsequent processing.

[0085] The circular dependency path is simplified to ensure that all nodes in the causal evaluation subgraph have a single dependency. For example, for the identified circular dependency path, the dependencies between nodes in the path are analyzed to determine the key nodes of the circular dependency. Using cut vertex or cut edge theory from graph theory, the nodes or edges of the circular dependency can be broken by removal or redirection operations. After removal or redirection operations, it is ensured that each node in the causal evaluation subgraph has at most one predecessor and one successor node, thus forming a single dependency.

[0086] The weights of each path in the simplified causal evaluation subgraph are updated. After simplifying the circular dependency paths, the weights of each path in the causal evaluation subgraph are recalculated to reflect the causal strength between nodes. Path weights are adjusted based on rules and experience from the expert knowledge base, considering factors including, but not limited to, path length, node importance, and historical data. By updating the weights, the causal relationship between advertising parameters and performance metrics can be more accurately assessed, providing data support for subsequent advertising strategies.

[0087] By following the steps above, we can effectively address the circular dependency problem in the causal evaluation subgraph, improve the accuracy and reliability of causal inference, and thus better guide advertising placement decisions.

[0088] This innovative solution analyzes the effectiveness of advertising on online food delivery platforms. It constructs a preliminary causal graph through data preprocessing and integrates it with an expert knowledge base to create a precise causal evaluation graph. Based on this, it generates causal evaluation subgraphs for specific problems and ensures single dependencies between nodes by identifying and simplifying circular dependency paths. Weights are then updated to optimize advertising strategies. This solution significantly improves the targeting of analysis, enhances the scientific basis of decision-making, optimizes advertising effectiveness, reduces campaign risks, and improves resource utilization efficiency, thereby enhancing merchants' market competitiveness in a fiercely competitive market.

[0089] In an optional embodiment, this application also provides a causal evaluation device for advertising placement, which may include:

[0090] The data acquisition module is used to acquire multiple data sets of advertisements placed by merchants operating on the food delivery platform. Each data set includes the advertising parameters and performance indicators of a merchant within a certain time period.

[0091] The causal graph generation module is used to input multiple datasets into the causal graph mining model set to obtain a first candidate causal graph set, wherein the graph mining model set includes multiple causal graph mining models;

[0092] The cause-effect graph filtering module is used to filter the first candidate cause-effect graph set based on the expert knowledge base to obtain the second candidate cause-effect graph set.

[0093] The causal graph aggregation module is used to obtain the weight of the path based on the relationship between the nodes and paths of each second candidate causal graph in the second candidate causal graph set, and to aggregate multiple second candidate causal graphs based on the weight of the path to obtain a large causal evaluation graph.

[0094] The evaluation module is used to evaluate the causal relationship between the advertising parameters and performance indicators of merchants based on this causal evaluation big picture.

[0095] Figure 3 This is a schematic diagram of the server structure provided in the embodiments of this application. The server 1000 can vary considerably due to different configurations or performance. It may include one or more processors (Central Processing Units, CPUs) 1001 and one or more memories 1002. Each memory 1002 stores at least one line of program code, which is loaded and executed by the processors 1001 to implement the causal evaluation method for advertising placement provided in the various method embodiments described above. Of course, the server 1000 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server 1000 may also include other components for implementing device functions, which will not be elaborated upon here.

[0096] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one piece of program code that is loaded and executed by a processor to enable a computer to implement any of the above-described methods for causal evaluation of advertising placement.

[0097] Optionally, the aforementioned computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0098] In an exemplary embodiment, a computer program or computer program product is also provided, which stores at least one computer instruction, which is loaded and executed by a processor to enable the computer to implement any of the above-described methods for causal evaluation of advertising placement.

[0099] It should be noted that all actions involving the acquisition of signals, information, or data in this application are carried out in compliance with the data protection laws and policies of the country where the application is located, and with authorization from the owner of the relevant device.

[0100] It should be understood that "multiple" as used in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0101] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0102] The above are merely exemplary embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A method for causal evaluation of advertising placement, characterized in that, The method includes: Obtain multiple data sets of advertisements placed by merchants operating on the food delivery platform, wherein each data set includes the advertising parameters and performance indicators of a merchant within a certain time period; Multiple datasets are input into a set of causal graph mining models to obtain a first candidate causal graph set, wherein the set of graph mining models includes multiple causal graph mining models; The first candidate causal graph set is filtered based on the expert knowledge base to obtain the second candidate causal graph set; The weight of a path is obtained based on the relationship between the nodes and paths of each second candidate causal graph in the set of second candidate causal graphs. Multiple second candidate causal graphs are then aggregated based on the weight of the paths to obtain a large causal evaluation graph. Based on the aforementioned causal evaluation big picture, the causal relationship between the advertising parameters and performance indicators of merchants is evaluated.

2. The method according to claim 1, characterized in that, After the step of obtaining multiple datasets of advertisements placed on the food delivery platform by merchants operating on the platform, the method further includes: The data in the dataset is processed, including data cleaning, anomaly detection, and null value filling.

3. The method according to claim 1, characterized in that, In the step of inputting multiple data sets into the causal graph mining model set to obtain the first candidate causal graph set: for N data sets, they are respectively input into M causal graph mining models in the causal graph mining model set to obtain the first candidate causal graph set, which includes N*M first candidate causal graphs.

4. The method according to claim 1, characterized in that, The first candidate causal graph set is filtered based on an expert knowledge base to obtain a second candidate causal graph set, which includes: A reference cause-effect diagram is generated based on the description of the causal relationship between deployment parameters and performance indicators in the expert knowledge base; Each first candidate causal graph in the first candidate causal graph set is matched with the reference causal graph to obtain a matching score; Based on the set matching score threshold, the first candidate causal graph set is filtered to obtain the second candidate causal graph set.

5. The method according to claim 1, characterized in that, The step involves obtaining the path weights based on the relationships between nodes and paths in each second candidate causal graph set, and then aggregating multiple second candidate causal graphs based on these path weights to obtain a large causal evaluation graph, including: The links of each second candidate causal graph in the second candidate causal graph set are split, wherein the links include node pairs and paths between node pairs; The weight of the link in the second candidate causal graph set is obtained based on the frequency of the link's occurrence in the second candidate causal graph. The second candidate causal graphs in the second candidate causal graph set are aggregated and simplified based on the weights to obtain a large causal evaluation graph.

6. The method according to any one of claims 1 to 4, characterized in that, After the steps of obtaining the path weights based on the relationships between nodes and paths in each second candidate causal graph set, and aggregating multiple second candidate causal graphs based on the path weights to obtain a large causal evaluation graph, the method further includes: Based on the deployment parameters and effect indicators of the causal relationship to be determined, the causal evaluation big graph is processed using an expert knowledge base to obtain a causal evaluation subgraph. Then, based on the aforementioned causal evaluation big graph and / or the aforementioned causal evaluation subgraph, the causal relationship between the merchant's advertising placement parameters and performance indicators is evaluated.

7. The method according to claim 6, characterized in that, The deployment parameters and effect indicators based on the causal relationship to be determined are processed using an expert knowledge base to obtain a causal evaluation sub-graph, including: Based on the deployment parameters and effect indicators that determine the causal relationship, the relevant causal rules in the expert knowledge base are determined; Based on the causal rules, the delivery parameters and intermediate variable factors that affect the performance indicators are identified, and the causal evaluation graph is layered. For each layer of the causal evaluation graph, nodes and / or paths that do not conform to the causal rules are removed to obtain a causal evaluation subgraph.

8. The method according to claim 6, characterized in that, After processing the causal evaluation subgraph using an expert knowledge base on the deployment parameters and effect indicators based on the causal relationship to be determined, the method further includes: Identify the circular dependency paths in the causal evaluation subgraph; The circular dependency path is simplified so that a single dependency relationship is formed between the nodes of the causal evaluation subgraph; The weights of each path in the simplified causal evaluation subgraph are updated.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one piece of program code, which is loaded and executed by a processor to enable the computer to implement the method as described in any one of claims 1 to 8.

Images