Marketing activity anomaly analysis method and device based on large language model

Abstract

The invention discloses a marketing activity anomaly analysis method and device based on a large language model, electronic equipment and a storage medium, and the method comprises the steps: obtaining to-be-analyzed order subsidy intensity data for to-be-analyzed order data, and carrying out the analysis and detection of the to-be-analyzed order subsidy intensity data through a preset statistical analysis model, and obtaining analysis detection result information to judge whether the subsidy strength data of the to-be-analyzed order is abnormal or not, actually, checking whether the marketing activity related to the to-be-analyzed order is abnormal or not, and when the subsidy strength data of the to-be-analyzed order is abnormal, analyzing the associated marketing activity data by adopting a large language model. Abnormal marketing activities and abnormal orders can be rapidly determined, and rapid positioning of the abnormal marketing activities is realized. The preset statistical analysis model is used to obtain the analysis detection result information so as to determine whether there is an abnormality, and there is no need to manually set an early warning threshold value, thereby improving the positioning efficiency and accuracy of the abnormal marketing activity.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a marketing campaign anomaly analysis method based on a large language model, a marketing campaign anomaly analysis device based on a large language model, electronic equipment, and computer storage media. Background Technology

[0002] As living standards continue to improve, the demand for online orders is increasing. At the same time, various marketing activities have been set up for online products in order to increase sales.

[0003] Because various marketing campaigns are set up for online products, many online orders also enjoy subsidies from these campaigns. In order to monitor orders on the platform and ensure that the subsidies set up for these campaigns are reasonable and do not cause asset losses, it is necessary to set some early warning thresholds to detect whether there are any anomalies in the marketing campaigns. However, this method requires manual setting of thresholds, which leads to low accuracy in warning of abnormal activities. At the same time, the existing methods for detecting marketing campaigns do not have a highly efficient detection chain, which also leads to low efficiency in locating anomalies in a large number of marketing campaigns. Therefore, how to improve the efficiency and accuracy of locating anomalies in a large number of marketing campaigns has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides a marketing campaign anomaly analysis method based on a large language model, which can improve the efficiency and accuracy of locating anomalies in massive marketing campaigns. This application also provides a marketing campaign anomaly analysis device, electronic device, and computer storage medium based on a large language model.

[0005] In a first aspect, this application provides a marketing campaign anomaly analysis method based on a large language model, comprising: obtaining order data to be analyzed for ordered goods and marketing campaign data to be analyzed for the goods; obtaining subsidy intensity data for the orders to be analyzed based on the order data to be analyzed; analyzing and detecting the subsidy intensity data for the orders to be analyzed using a preset statistical analysis model to obtain analysis and detection result information; determining whether the subsidy intensity data for the orders to be analyzed is abnormal based on the analysis and detection result information; if the subsidy intensity data for the orders to be analyzed is abnormal, analyzing the associated marketing campaign data using a large language model to determine the abnormal marketing campaign and abnormal orders; the abnormal marketing campaign is an abnormal campaign within the associated marketing campaign, and the abnormal order is an abnormal order within the order to be analyzed; the associated marketing campaign data is the marketing campaign data determined to be associated with the order data to be analyzed in the marketing campaign data to be analyzed; and analyzing the abnormal marketing campaign data and abnormal order data using the large language model to obtain anomaly report information for the abnormal marketing campaign.

[0006] Secondly, this application provides a marketing activity anomaly analysis device based on a large language model, comprising: a data acquisition unit, used to acquire order data to be analyzed for ordered goods and marketing activity data to be analyzed for the goods; an order subsidy intensity data acquisition unit, used to acquire order subsidy intensity data to be analyzed based on the order data to be analyzed; a statistical analysis model processing unit, used to analyze and detect the order subsidy intensity data to be analyzed using a preset statistical analysis model, and obtain analysis and detection result information; and determine whether the order subsidy intensity data to be analyzed is abnormal based on the analysis and detection result information; a large language model first processing unit, used to analyze related marketing activity data using a large language model if the order subsidy intensity data to be analyzed is abnormal, and determine abnormal marketing activities and abnormal orders; the abnormal marketing activity is an abnormal activity in related marketing activities, and the abnormal order is an abnormal order in the order to be analyzed; the related marketing activity data is the marketing activity data determined in the marketing activity data to be analyzed that is associated with the order data to be analyzed; and a large language model second processing unit, used to analyze the abnormal marketing activity data and abnormal order data using the large language model, and obtain anomaly report information for the abnormal marketing activity.

[0007] Thirdly, this application provides an electronic device, including: a processor; and a memory for storing a computer program. After the electronic device is powered on and runs the computer program through the processor, it executes the above-mentioned marketing activity anomaly analysis method based on a large language model.

[0008] Fourthly, this application provides a computer storage medium storing computer execution instructions, which are executed by a processor to perform the above-mentioned marketing activity anomaly analysis method based on a large language model.

[0009] Compared with the prior art, this application has the following advantages: The marketing campaign anomaly analysis method based on a large language model provided in this application includes: obtaining order data for ordered goods to be analyzed and marketing campaign data for the goods to be analyzed; obtaining subsidy intensity data for the orders to be analyzed based on the order data; analyzing and detecting the subsidy intensity data for the orders to be analyzed using a preset statistical analysis model to obtain analysis and detection result information; determining whether there are anomalies in the subsidy intensity data for the orders to be analyzed based on the analysis and detection result information; if there are anomalies in the subsidy intensity data for the orders to be analyzed, analyzing the associated marketing campaign data using a large language model to determine the abnormal marketing campaigns and abnormal orders; the abnormal marketing campaigns are the abnormal campaigns within the associated marketing campaigns, and the abnormal orders are the abnormal orders within the orders to be analyzed; the associated marketing campaign data are the marketing campaign data determined to be associated with the order data to be analyzed in the marketing campaign data; and analyzing the abnormal marketing campaign data and abnormal order data using a large language model to obtain anomaly report information for the abnormal marketing campaigns. This method involves obtaining subsidy intensity data for the orders to be analyzed, then using a pre-defined statistical analysis model to analyze and detect this data. The analysis results are used to determine if any anomalies exist in the subsidy intensity data. Anomalies in the subsidy intensity data can be used to check if the marketing activities associated with the orders are abnormal. Therefore, anomalies in the subsidy intensity data are equivalent to confirming anomalies in the marketing activities associated with the orders. When anomalies are found, a large language model is used to analyze the associated marketing activity data, quickly identifying abnormal marketing activities and orders. This enables rapid location of abnormal marketing activities. Furthermore, the pre-defined statistical analysis model provides analysis results for the subsidy intensity data to facilitate anomaly detection, eliminating the need for manually setting warning thresholds and improving the efficiency and accuracy of locating abnormal marketing activities. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0011] Figure 1 A flowchart of the marketing campaign anomaly analysis method based on a large language model provided in the first embodiment of this application.

[0012] Figure 2 This is a schematic diagram of the framework of the marketing campaign anomaly analysis method based on a large language model provided in the first embodiment of this application.

[0013] Figure 3 A schematic diagram of a marketing campaign anomaly analysis device based on a large language model provided in the second embodiment of this application.

[0014] Figure 4 A schematic diagram of an electronic device provided in the third embodiment of this application. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions of this application, the application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. However, this application can be implemented in many other ways different from those described below. Therefore, based on the embodiments provided in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0016] It should be noted that the terms "first," "second," "third," etc., in the claims, specification, and drawings of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. Such data are interchangeable where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown or described in this application. Furthermore, the terms "comprising," "having," and their variations are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0017] It should be understood that in the embodiments of this application, "at least one" means one or more, and "more than one" means two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship. "Contains A, B and / or C" means containing any one, two, or three of A, B, and C.

[0018] It should be understood that in the embodiments of this application, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0019] This application provides a marketing campaign anomaly analysis method, a marketing campaign anomaly analysis device, an electronic device, and a computer storage medium based on a large language model. The following specific embodiments illustrate the marketing campaign anomaly analysis method, the marketing campaign anomaly analysis device, the electronic device, and the computer storage medium based on a large language model.

[0020] First Embodiment This embodiment provides a marketing campaign anomaly analysis method based on a large language model. Please refer to [link / reference] for details. Figure 1 The flowchart is a method for analyzing marketing campaign anomalies based on a large language model, provided in the first embodiment of this application.

[0021] The marketing campaign anomaly analysis method based on a large language model according to this application includes the following steps.

[0022] Step S101: Obtain the order data to be analyzed for the ordered goods and the marketing campaign data to be analyzed for the goods.

[0023] The marketing campaign anomaly analysis method based on a large language model in this embodiment is actually used to detect anomalies in massive marketing campaigns. Marketing campaigns are mainly set up for online products. Marketing campaign data refers to data related to marketing campaigns, such as "Double Eleven" campaigns, "New Year's Shopping Festival" campaigns, etc. The scope of marketing campaign data is quite broad, including data such as campaign subsidy data (e.g., 10 yuan off for every 100 yuan spent, where 10 yuan is the subsidy data), campaign subsidy intensity data (mainly referring to the subsidy intensity data enjoyed by orders containing products from the marketing campaign), campaign identification data, etc. In fact, marketing campaign data includes many types, which will not be listed here. Order data refers to data related to online orders. Order data mainly includes the products included in the order, the actual cost information of the order, the subsidy information of the order, the subsidy intensity data of the order (which can be obtained by dividing the subsidy information of the order by the actual cost information of the order), order identification data, etc. In fact, order data also includes many types, which will not be listed here.

[0024] Since marketing campaigns are closely linked to orders—each order corresponds to multiple marketing campaigns (e.g., each order enjoys multiple marketing campaigns), and each marketing campaign corresponds to multiple orders (e.g., each marketing campaign corresponds to multiple orders enjoying that campaign)—this embodiment can determine whether order data is abnormal based on its presence. If order data is abnormal, the corresponding marketing campaign data can then be anomaly located. It's understood that abnormal order data refers to abnormal order data associated with the marketing campaign. Since both order data and marketing campaign data correspond to subsidy intensity data, anomalies in subsidy intensity data also indicate anomalies in the corresponding order and marketing campaign data. Therefore, this embodiment determines whether order data is abnormal based on whether the order subsidy intensity data is abnormal. Subsequently, if order subsidy intensity data is abnormal, the abnormal marketing campaign can be quickly located based on the marketing campaign data associated with the abnormal order data.

[0025] Therefore, in this embodiment, it is necessary to obtain both order data and marketing activity data simultaneously. The order data is used as the order data to be analyzed, and the marketing activity data is used as the marketing activity data to be analyzed.

[0026] Given the large volume of orders and marketing activities, order data can be categorized according to preset categories to obtain order data for analysis. One method to obtain order data for ordered goods is to acquire multi-source order data; then, this multi-source order data is categorized according to preset categories to obtain the order data for analysis; preset categories include at least one of industry, city, brand, and store.

[0027] Multi-source order data is actually massive amounts of order data, which is obtained from various order ordering applications (or various order ordering platforms).

[0028] After obtaining multi-source order data, the store attribute data of the stores to which the multi-source orders belong can also be obtained simultaneously. Based on the store attribute data, the multi-source order data is classified and processed according to preset categories, and the classified order data is used as the order data to be analyzed. The preset categories include industry, city, brand, and store. The preset categories can also be other situations listed above. After classification and processing, the actual classified order data includes order data from various industries, cities, brands, and stores. That is, the order data from each industry, city, brand, and store constitutes one type of order data to be analyzed.

[0029] Since store attribute data includes the store's industry, city, and brand, it is possible to classify multi-source order data according to preset categories based on store attribute data. In this embodiment, the order data of a specific store is used as an example for analysis. It should be noted that the order data to be analyzed is actually the order data corresponding to orders generated within the current time period, such as the order data of store 1 in the last 20 minutes. Of course, it can be understood that the current time period can be set according to requirements, such as the order data of the store in the last hour.

[0030] The marketing campaign data to be analyzed is actually data configured for marketing campaigns. This data consists of massive amounts of marketing campaign data that have not been categorized. As mentioned earlier, the "10 RMB off for every 100 RMB spent" configuration for the "Double Eleven" campaign is one such configuration.

[0031] Step S102: Obtain the subsidy intensity data of the orders to be analyzed based on the order data to be analyzed.

[0032] After obtaining the order data to be analyzed, in order to identify anomalies in the subsidy data associated with marketing campaigns within the order data, and then to pinpoint the anomalies in the corresponding marketing campaign data, it is necessary to obtain the subsidy intensity data of the orders to be analyzed. The subsidy intensity data of the orders to be analyzed is an important indicator used to characterize the subsidy data associated with marketing campaigns within the order data.

[0033] In this embodiment, the subsidy intensity data of the orders to be analyzed can be obtained based on the order data to be analyzed. Specifically, the initial order subsidy intensity data of the orders to be analyzed is obtained based on the order data to be analyzed; there are multiple orders to be analyzed; then, the initial order subsidy intensity data is decomposed according to a preset subsidy dimension to obtain the order subsidy intensity data under each subsidy dimension; then, for multiple orders to be analyzed, the order subsidy intensity data under each subsidy dimension is aggregated to obtain the aggregated order subsidy intensity data under each subsidy dimension as the subsidy intensity data of the orders to be analyzed.

[0034] Taking the order data of Store 1 in the past 20 minutes as the order data to be analyzed as an example, assuming that Store 1 has 1000 orders in the past 20 minutes (in reality, it may be much more than 1000), we obtain the initial order subsidy intensity data for each of these 1000 orders. Then, according to the preset subsidy dimensions, we break down the initial order subsidy intensity data of each of the 1000 orders to obtain the order subsidy intensity data under each subsidy dimension for each order. The preset subsidy dimensions include platform subsidy dimension, brand subsidy dimension, store subsidy dimension, and agent subsidy dimension, that is, the subsidy for each order actually comes from one or more of multiple subsidy dimensions. Of course, the preset subsidy dimensions can also be other situations listed above. For example, suppose the first order has a total subsidy of 10 yuan, with 5 yuan from the platform subsidy, 2 yuan from the brand subsidy, 1 yuan from the store subsidy, and 2 yuan from the agent subsidy (the agent is the agent involved with the store). Assuming the actual cost of this order is 20 yuan, then the order subsidy intensity data for each subsidy dimension are 0.25, 0.1, 0.05, and 0.1 respectively. This is using the first order as an example. In reality, for all 1000 orders, it is necessary to obtain the order subsidy intensity data for each subsidy dimension. Then, the order subsidy intensity data for the platform subsidy dimension of these 1000 orders is aggregated (e.g., by averaging, or other methods). The order subsidy intensity data for the brand subsidy dimension is aggregated; the order subsidy intensity data for the store subsidy dimension is aggregated; and the order subsidy intensity data for the agent subsidy dimension is aggregated. These aggregated order subsidy intensity data for these four dimensions are used as the order subsidy intensity data to be analyzed.

[0035] The above-mentioned initial order subsidy intensity data for the orders to be analyzed is obtained based on the order data to be analyzed. This includes: obtaining order subsidy information and actual order cost information for each order to be analyzed; and obtaining the initial order subsidy intensity data based on the order subsidy information and actual order cost information. For example, if the total subsidy for the first order is 10 yuan and the actual cost is 20 yuan, then the initial order subsidy intensity data is 10 divided by 20, which equals 0.5. The initial order subsidy intensity data is then broken down according to preset subsidy dimensions to obtain the order subsidy intensity data under each subsidy dimension, such as breaking 0.5 down into 0.25, 0.1, 0.05, and 0.1.

[0036] Step S103: Use a preset statistical analysis model to analyze and detect the subsidy intensity data of the order to be analyzed, and obtain the analysis and detection results; based on the analysis and detection results, determine whether there are any anomalies in the subsidy intensity data of the order to be analyzed.

[0037] After obtaining the subsidy intensity data of the orders to be analyzed, a preset statistical analysis model is used to analyze and test the subsidy intensity data of the orders to be analyzed, and the analysis and test results are obtained.

[0038] In this embodiment, the method further includes: obtaining historical order subsidy intensity data; historical orders refer to orders within a historical time period; for example, obtaining historical orders from store 1 over the past 30 days. It should be noted that historical orders correspond to orders to be analyzed, that is, if the order data to be analyzed is the order data of a certain store, then the historical orders are also the historical orders of that store; if the order data to be analyzed is the order data of a certain industry, then the historical orders are also the historical orders of that industry; if the order data to be analyzed is the order data of a certain city, then the historical orders are also the historical orders of that city; if the order data to be analyzed is the order data of a certain brand, then the historical orders are also the historical orders of that brand.

[0039] Using a pre-set statistical analysis model to analyze and test the subsidy intensity data of the order to be analyzed, and obtaining the analysis and testing results, can refer to: using a pre-set statistical analysis model based on historical order subsidy intensity data to analyze and test the subsidy intensity data of the order to be analyzed, and obtaining the analysis and testing results.

[0040] More specifically, the preset statistical analysis model includes an exponentially weighted moving average model and a standardization model. Using the preset statistical analysis model to analyze and test the historical order subsidy intensity data based on the order subsidy intensity data to obtain analysis and testing results can refer to: using an exponentially weighted moving average model to analyze the changing trends of historical order subsidy intensity data to obtain predicted order subsidy intensity data corresponding to the current time; and then using a standardization model to standardize the predicted order subsidy intensity data based on the predicted order subsidy intensity data to obtain analysis and testing results.

[0041] The Exponentially Weighted Moving Average (EWMA) model is a statistical method widely used in time series analysis. Its core idea is to assign higher weights to recent observations and decreasing weights to earlier data points, allowing the model to more sensitively capture the latest trends in the data. In this scenario, the EWMA model can analyze the changing trends of historical order subsidy intensity data. For example, analyzing the changing trends of historical order subsidy intensity data for Store 1 over the past 30 days. In the specific analysis process, based on the exponentially weighted decay mechanism, higher weights are assigned to recent data (e.g., over these 30 days, historical order subsidy intensity data closer to the current time period is assigned higher weights; for example, yesterday's order subsidy intensity data is assigned a higher weight than the previous 29 days). This allows for the prediction of the current order subsidy intensity data (a theoretical value), which is then compared with the actual order subsidy intensity data to be analyzed. It should be noted that the historical order subsidy intensity data refers to the actual subsidy intensity data of historical orders. For details, please refer to the process of obtaining the subsidy intensity data of the order to be analyzed. However, the historical order subsidy intensity data is based on historical orders. It should also be noted that the subsidy intensity data of the order to be analyzed corresponds to the predicted order subsidy intensity data. For example, if the subsidy intensity data of the order to be analyzed is under the platform subsidy dimension, then the predicted order subsidy intensity data is also under the platform subsidy dimension.

[0042] The standardized processing model can refer to the standard score model, which is another Chinese expression for "Z-score." It is also a statistical concept, emphasizing the standardization process. The "Z-score" model standardizes the subsidy intensity data of the order to be analyzed by dividing the predicted subsidy intensity data by the standard deviation data. For example, the difference between the subsidy intensity data of the order to be analyzed and the predicted subsidy intensity data is divided by the standard deviation data to obtain the analysis results. The standard deviation data is calculated using the EWMA model to predict the subsidy intensity data of historical orders. It should be noted that the subsidy intensity data of historical orders predicted using the EWMA model is different from the actual subsidy intensity data of historical orders. The actual subsidy intensity data of historical orders is the actual subsidy intensity data, while the subsidy intensity data of historical orders predicted using the EWMA model is the theoretical subsidy intensity data.

[0043] Subsequently, based on the analysis and testing results, it is determined whether there are any anomalies in the subsidy intensity data of the orders to be analyzed. In fact, it is determined whether the standardized data is within the confidence interval, such as whether the analysis and testing results are within the 99.7% confidence interval, thereby effectively distinguishing between normal fluctuations and potential risks.

[0044] Step S104: If there are anomalies in the subsidy intensity data of the order to be analyzed, the large language model is used to analyze the related marketing activity data to identify the abnormal marketing activities and abnormal orders.

[0045] In practice, it is necessary to first use a large language model to analyze the marketing campaign data and the order data to be analyzed, and then identify the related marketing campaign data that is associated with the order data in the marketing campaign data to be analyzed.

[0046] After obtaining the order data to be analyzed, such as the order data of a specific store, the marketing activity data to be analyzed is also a large amount of data. Furthermore, the method in this embodiment first checks for anomalies in the order data, and then locates the anomalies in the corresponding marketing activity data if anomalies are found. Therefore, it is necessary to identify related marketing activity data associated with the order data to be analyzed from the marketing activity data to be analyzed. For example, when using the order data of store 1 from the past 20 minutes as the order data to be analyzed, it is necessary to identify the marketing activity data associated with the order data of store 1 from the massive marketing activity data. Assuming there are 100 marketing activities to be analyzed, and store 1's orders from the past 20 minutes have already benefited from 20 of these marketing activities, then the data from these 20 marketing activities are the related marketing activity data. In reality, the number of marketing activities to be analyzed is much larger; this example is for ease of understanding. In practice, the number of marketing activities to be analyzed far exceeds 100.

[0047] In this embodiment, a large language model is used to analyze the marketing campaign data and order data to be analyzed, and to identify related marketing campaign data associated with the order data in the marketing campaign data. Specifically, the analysis of the marketing campaign data and order data to be analyzed using a large language model, and the identification of related marketing campaign data associated with the order data in the marketing campaign data, can refer to the following: First, the large language model is used to analyze the marketing campaign data and order data to be analyzed, and to construct an identifier correspondence between the activity identifier of the marketing campaign data to be analyzed and the order identifier of the order data in the order data; then, based on the identification correspondence between the marketing campaign data to be analyzed, the order data to be analyzed, and the identification correspondence, the related marketing campaign data associated with the order data in the marketing campaign data is identified.

[0048] In fact, in this embodiment, since both the marketing activity volume and the order volume to be analyzed are large, a large language model is used to construct a network relationship between the marketing activities and the orders to be analyzed. The network relationship between the marketing activities and the orders to be analyzed is represented by the identification correspondence between the activity identifier (activity ID, ID is the identity document) of the marketing activity data and the order identifier (order ID) of the order data.

[0049] After obtaining the identifier mapping relationship, the associated marketing campaign data can be identified within the marketing campaign data and the order data to be analyzed, based on this mapping relationship. Alternatively, a large language model can be used to analyze the marketing campaign data and multi-source order data to construct an identifier mapping relationship between the campaign identifiers of the marketing campaign data and the order identifiers of the multi-source order data. Either method used to construct the identifier mapping relationship is acceptable.

[0050] In this embodiment, a large language model can refer to an artificial intelligence model (such as AI) that processes semantic analysis. A large language model is abbreviated as LLM, which stands for Large Language Model.

[0051] In this embodiment, abnormal marketing activities refer to abnormal activities within related marketing activities, and abnormal orders refer to abnormal orders among the orders to be analyzed.

[0052] Specifically, analyzing related marketing campaign data using a large language model to identify anomalous marketing campaigns and orders can involve: obtaining the subsidy intensity data for the campaigns to be analyzed corresponding to the related marketing campaign data; breaking down the subsidy intensity data for the campaigns to be analyzed according to preset campaign attribute dimensions to obtain the subsidy intensity data for each campaign attribute dimension; calculating the deviation of preset indicators for the marketing campaigns based on the subsidy intensity data for each campaign attribute dimension; using the large language model based on the indicator deviation to identify anomalous marketing campaigns, and then identifying and filtering orders associated with these anomalous marketing campaigns to determine anomalous orders. It should be noted that not all orders associated with anomalous marketing campaigns are anomalous orders; therefore, further filtering based on the actual situation is necessary to determine the truly anomalous orders. Orders associated with anomalous marketing campaigns can be orders that benefited from the anomalous marketing campaigns. Anomalous orders are also among the orders to be analyzed.

[0053] Based on the activity subsidy intensity data under each activity attribute dimension, the deviation of the preset indicators of the marketing activity is calculated. This can refer to: for each activity attribute dimension, based on the activity subsidy intensity data under the target activity attribute dimension, determining whether there are any anomalies in the marketing activity under the target activity attribute dimension; if so, calculating the deviation of the preset indicators of the marketing activity under the target activity attribute dimension; each activity attribute dimension can be used as the target activity attribute dimension.

[0054] Preset activity attribute dimensions such as marketing activity type, applicable channels, and effective date. Of course, preset activity attribute dimensions can also be other than those listed above.

[0055] Preset indicators include marketing campaign order volume, marketing campaign subsidy intensity data, and changes in order volume before and after the marketing campaign. These preset indicators have corresponding preset threshold conditions. If the indicator deviation does not meet the preset threshold conditions, a certain marketing campaign is accurately identified as an abnormal marketing campaign. In this embodiment, the subsidy intensity data of a large number of related marketing campaigns to be analyzed is decomposed according to preset activity attribute dimensions to obtain the activity subsidy intensity data under each activity attribute dimension. The influence weight of the marketing campaign on the subsidy trend is quantified, thereby initially determining whether there is an anomaly in the marketing campaign under a certain activity attribute dimension. Then, under the premise that the marketing campaign under that activity attribute dimension may be abnormal, the indicator deviation is further calculated to further confirm whether the marketing campaign under that activity attribute dimension is abnormal.

[0056] More specifically, obtaining the subsidy intensity data of the activity to be analyzed corresponding to the related marketing campaign data includes: obtaining initial activity subsidy intensity data based on the related marketing campaign data; there are multiple related marketing campaigns; the initial activity subsidy intensity data is broken down according to preset subsidy dimensions to obtain activity subsidy intensity data under each subsidy dimension; for multiple related marketing campaigns, the activity subsidy intensity data under each subsidy dimension is aggregated to obtain aggregated activity subsidy intensity data under each subsidy dimension as the activity subsidy intensity data to be analyzed.

[0057] Related marketing activities, such as all orders in Store 1 within the last 20 minutes, can be analyzed by referring to the method for obtaining the subsidy intensity data of the orders to be analyzed. This will not be elaborated here.

[0058] Furthermore, obtaining initial campaign subsidy intensity data based on associated marketing campaign data can be achieved by: obtaining subsidy cost information and actual payment information for orders involved in the campaign for each associated marketing campaign; and obtaining initial campaign subsidy intensity data based on the subsidy cost information and actual payment information for orders involved in the campaign.

[0059] For example, for Activity A, if only Order 1, Order 2, and Order 3 out of all orders placed in Store 1 within the last 20 minutes are eligible for Activity A, then the initial activity subsidy intensity data for Activity A can be obtained by dividing the aggregated subsidy cost information of Order 1, Order 2, and Order 3 by the aggregated actual cost information of Order 1, Order 2, and Order 3.

[0060] In this embodiment, there are multiple abnormal marketing activities. Since each abnormal marketing activity exhibits a deviation in metrics, the method further includes: using a large language model to determine the abnormality priority of the abnormal marketing activities based on the magnitude of the metric deviation; sorting the multiple abnormal marketing activities based on their abnormality priority to obtain sorting information; marking the abnormal marketing activities with processing level information based on the sorting information; and using the processing level information to issue alerts for the abnormal marketing activities. For example, the more severe the abnormality of a particular marketing activity, the higher its processing level, and therefore, it needs to be alerted with priority. It is understood that the sorting information obtained for the multiple abnormal marketing activities also includes abnormal orders associated with the abnormal marketing activities.

[0061] Step S105: Use a large language model to analyze abnormal marketing activity data and abnormal order data to obtain abnormal report information for abnormal marketing activities.

[0062] In this embodiment, a large language model is used to analyze abnormal marketing activity data and abnormal order data to obtain abnormal reporting information for abnormal marketing activities. This includes: obtaining abnormal subsidy intensity data based on abnormal marketing activity data and abnormal order data; obtaining abnormal subsidy intensity data based on abnormal activity subsidy intensity data and abnormal order subsidy intensity data; obtaining a multi-dimensional abnormal feature vector based on abnormal marketing activity data, abnormal order data, and abnormal subsidy intensity data; and inputting the multi-dimensional abnormal feature vector into the large language model to obtain abnormal reporting information.

[0063] It should be noted that the abnormal subsidy intensity data includes the intersection of abnormal activity subsidy intensity data and abnormal order subsidy intensity data.

[0064] In this embodiment, the anomaly report information includes the anomaly cause information of the abnormal marketing activity, the scope of impact information of the abnormal marketing activity (such as which orders are affected), and the handling suggestions for the abnormal marketing activity; at the same time, the anomaly report information may also include the abnormal location characteristics of the abnormal marketing activity.

[0065] It should be noted that task description information also needs to be input into the large language model so that it can understand its task. The large language model, upon obtaining multi-dimensional anomaly feature vectors, can automatically associate them with anomaly analysis results from historical anomaly marketing campaigns as a reference, and then generate anomaly analysis results for the currently input anomaly marketing campaign.

[0066] Once anomaly analysis results are obtained, they are output as anomaly report information. The difference between anomaly analysis results and anomaly report information is that anomaly analysis results are analysis results of abnormal marketing activities within a short period of time, while anomaly report information is a report that aggregates anomaly analysis results over a period of time (such as one day).

[0067] In this embodiment, the method further includes: providing the anomaly report information to the display terminal so as to display the anomaly report information on the display terminal for manual review.

[0068] To facilitate understanding of the overall framework of the marketing campaign anomaly analysis method based on a large language model in this embodiment, please refer to... Figure 2 This is a schematic diagram of the framework of the marketing campaign anomaly analysis method based on a large language model provided in the first embodiment of this application.

[0069] The framework 200 includes a data source acquisition module 201, an order dimension splitting module 202, a data source channel module 203, a model analysis module 204, and an anomaly analysis result processing module 205.

[0070] The data source acquisition module 201 is used to acquire multi-source order data, marketing activity data to be analyzed, and store attribute data from the data source channel module 203; the order dimension splitting module 202 classifies the multi-source order data based on the store attribute data to obtain the order data to be analyzed; the model analysis module 204 processes the order data to be analyzed and the marketing activity data to be analyzed to obtain anomaly analysis results for abnormal marketing activities, and provides them to the anomaly analysis result processing module 205; the anomaly analysis result processing module 205 generates anomaly report information for abnormal marketing activities based on the anomaly analysis result information. Figure 2 AI notifications refer to the use of intelligent agents to inform relevant personnel of abnormal marketing activities, and AI analysis reports are the abnormality reporting information.

[0071] In this embodiment, orders and activities are actually linked through subsidies (specifically, subsidy intensity data). Then, when an order is deemed abnormal, the abnormal marketing activity within the order is located. In this embodiment, the subsidy intensity data in the order data and the subsidy intensity data in the marketing activity data are broken down and aggregated to facilitate the location of abnormal marketing activities. The reasons for abnormal marketing activities can be varied: for example, an excessively high subsidy for a particular marketing activity, or an abnormal subsidy caused by the overlap of multiple marketing activities.

[0072] In this embodiment, the method obtains subsidy intensity data for the order data to be analyzed, then uses a preset statistical analysis model to analyze and detect the subsidy intensity data, obtaining analysis and detection results. Based on these results, it is determined whether the subsidy intensity data is abnormal. The abnormality of the subsidy intensity data allows us to check whether the marketing activities associated with the order are abnormal. Therefore, when the subsidy intensity data is abnormal, it is equivalent to confirming that the marketing activities associated with the order are abnormal. Since a large language model is used beforehand to identify related marketing activities in the marketing activity data associated with the order data, when the subsidy intensity data is abnormal, the large language model analyzes the related marketing activity data, quickly identifying abnormal marketing activities and abnormal orders. This enables rapid location of abnormal marketing activities. Furthermore, the preset statistical analysis model obtains analysis and detection results for the subsidy intensity data to facilitate the determination of abnormality, eliminating the need for manually setting warning thresholds and improving the efficiency and accuracy of locating abnormal marketing activities.

[0073] Second Embodiment Corresponding to the marketing campaign anomaly analysis method based on a large language model provided in the first embodiment of this application, the second embodiment of this application also provides a marketing campaign anomaly analysis device based on a large language model. Since the device embodiment is basically similar to the first embodiment, it is described simply, and relevant parts can be referred to in the description of the first embodiment. The device embodiments described below are merely illustrative.

[0074] Please refer to Figure 3 This is a schematic diagram of a marketing activity anomaly analysis device based on a large language model provided in the second embodiment of this application.

[0075] The marketing activity anomaly analysis device 300 based on a large language model includes: a data acquisition unit 301, used to acquire order data to be analyzed for ordered goods and marketing activity data to be analyzed for the goods; an order subsidy intensity data acquisition unit 302, used to acquire order subsidy intensity data to be analyzed based on the order data to be analyzed; a statistical analysis model processing unit 303, used to analyze and detect the order subsidy intensity data to be analyzed using a preset statistical analysis model, and obtain analysis and detection result information; and to determine whether there is an anomaly in the order subsidy intensity data to be analyzed based on the analysis and detection result information; and a large language model first processing unit. Unit 304 is used to analyze the associated marketing activity data using a large language model if the subsidy intensity data of the order to be analyzed is abnormal, and to identify abnormal marketing activities and abnormal orders; the abnormal marketing activity is an abnormal activity in the associated marketing activities, and the abnormal order is an abnormal order in the order to be analyzed; the associated marketing activity data is the marketing activity data identified in the marketing activity data to be analyzed that is associated with the order data to be analyzed; the second processing unit 305 of the large language model is used to analyze the abnormal marketing activity data and abnormal order data using the large language model to obtain abnormal report information for the abnormal marketing activity.

[0076] Third Embodiment Corresponding to the method of the first embodiment of this application, the third embodiment of this application also provides an electronic device.

[0077] The electronic device includes: a processor; and a memory for storing a computer program. After the electronic device is powered on and runs the computer program through the processor, it executes the method of the first embodiment. Figure 4 As shown, Figure 4 This is a schematic diagram of an electronic device provided according to a third embodiment of this application. The electronic device specifically includes: at least one processor 401, at least one communication interface 402, at least one memory 403, and at least one communication bus 404. Optionally, the communication interface 402 can be an interface for a communication module, such as an interface for a GSM module. The processor 401 may be a CPU, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of this application. The memory 403 may include high-speed RAM and may also include non-volatile memory, such as at least one disk storage device. The memory 403 stores a program, and the processor 401 calls the program stored in the memory 403 to execute the method of the first embodiment.

[0078] Fourth embodiment Corresponding to the method of the first embodiment of this application, the fourth embodiment of this application also provides a computer storage medium storing computer execution instructions, which are executed by a processor to perform the method of the first embodiment of this application.

[0079] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0080] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0081] 1. Computer-readable media includes both permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined in this application, computer-readable media does not include non-transitory computer-readable media, such as modulated data signals and carrier waves.

[0082] 2. Those skilled in the art will understand that embodiments of this application can provide methods, systems, or computer program products. Therefore, embodiments of this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] 3. This application embodiment may involve the use of user data. In practical applications, user-specific personal data may be used within the scope permitted by applicable laws and regulations of the country in which the application is located (e.g., with the user's explicit consent and effective notification to the user, etc.). Furthermore, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0084] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A marketing campaign anomaly analysis method based on a large language model, characterized in that, include: Obtain order data for ordered products to be analyzed, and marketing campaign data for products to be analyzed; Based on the order data to be analyzed, obtain the subsidy intensity data for the order to be analyzed; A preset statistical analysis model is used to analyze and test the subsidy intensity data of the orders to be analyzed, and the analysis and test results are obtained. Based on the analysis and detection results, it is determined whether there are any anomalies in the subsidy intensity data of the order to be analyzed; If the subsidy intensity data of the order to be analyzed is abnormal, a large language model is used to analyze the related marketing activity data to identify abnormal marketing activities and abnormal orders; the abnormal marketing activities are abnormal activities in related marketing activities, and the abnormal orders are abnormal orders in the order to be analyzed. The associated marketing campaign data refers to the marketing campaign data identified in the marketing campaign data to be analyzed that is associated with the order data to be analyzed; The large language model is used to analyze abnormal marketing activity data and abnormal order data to obtain abnormal report information for the abnormal marketing activities.

2. The method according to claim 1, characterized in that, Also includes: A large language model is used to analyze the marketing campaign data and the order data to be analyzed, and related marketing campaign data associated with the order data to be analyzed are identified in the marketing campaign data.

3. The method according to claim 2, characterized in that, The process involves using a large language model to analyze the marketing campaign data and the order data to be analyzed, and identifying related marketing campaign data associated with the order data in the marketing campaign data to be analyzed, including: A large language model is used to analyze the marketing campaign data and the order data to be analyzed, and to construct an identifier correspondence between the campaign identifier of the marketing campaign data and the order identifier of the order data. Based on the correspondence between the marketing campaign data to be analyzed, the order data to be analyzed, and the identifier, related marketing campaign data associated with the order data to be analyzed is determined from the marketing campaign data to be analyzed.

4. The method according to claim 1, characterized in that, The step of obtaining the subsidy intensity data of the order to be analyzed based on the order data to be analyzed includes: Based on the order data to be analyzed, the initial order subsidy intensity data of the orders to be analyzed is obtained; the number of orders to be analyzed is multiple. The initial order subsidy intensity data is broken down according to the preset subsidy dimensions to obtain the order subsidy intensity data under each subsidy dimension; For multiple orders to be analyzed, the order subsidy intensity data under each subsidy dimension is aggregated to obtain the aggregated order subsidy intensity data under each subsidy dimension as the order subsidy intensity data to be analyzed.

5. The method according to claim 4, characterized in that, The step of obtaining the initial order subsidy intensity data of the orders to be analyzed based on the order data to be analyzed includes: For each order to be analyzed, obtain information on order subsidy fees and actual order expenses; Based on the order subsidy information and the actual order payment information, the initial order subsidy intensity data of the order to be analyzed is obtained.

6. The method according to claim 1, characterized in that, Also includes: Obtain historical order subsidy intensity data; The step involves using a preset statistical analysis model to analyze and test the subsidy intensity data of the orders to be analyzed, and obtaining analysis and testing results, including: A preset statistical analysis model is used to analyze and detect the subsidy intensity data of the order to be analyzed based on the historical order subsidy intensity data, and the analysis and detection results are obtained.

7. The method according to claim 6, characterized in that, The preset statistical analysis model includes an exponentially weighted moving average model and a standardized processing model; The step involves using a preset statistical analysis model to analyze and test the subsidy intensity data of the orders to be analyzed based on the historical order subsidy intensity data, and obtaining analysis and testing results information, including: An exponentially weighted moving average model is used to analyze the changing trend of the historical order subsidy intensity data to obtain the predicted order subsidy intensity data corresponding to the current time. The standardized processing model is used to standardize the order subsidy intensity data to be analyzed based on the predicted order subsidy intensity data, thereby obtaining the analysis and detection results information.

8. A marketing campaign anomaly analysis device based on a large language model, characterized in that, include: The data acquisition unit is used to acquire order data for ordered goods to be analyzed and marketing campaign data for the goods to be analyzed. The unit for obtaining subsidy intensity data of orders to be analyzed is used to obtain subsidy intensity data of orders to be analyzed based on the data of orders to be analyzed. The statistical analysis model processing unit is used to analyze and detect the subsidy intensity data of the order to be analyzed using a preset statistical analysis model, and to obtain the analysis and detection results information. Based on the analysis and detection results, it is determined whether there are any anomalies in the subsidy intensity data of the order to be analyzed; The first processing unit of the large language model is used to analyze the related marketing activity data using the large language model if there are anomalies in the subsidy intensity data of the order to be analyzed, and to determine the abnormal marketing activities and abnormal orders; the abnormal marketing activities are abnormal activities in the related marketing activities, and the abnormal orders are abnormal orders in the order to be analyzed. The associated marketing campaign data refers to the marketing campaign data identified in the marketing campaign data to be analyzed that is associated with the order data to be analyzed; The second processing unit of the large language model is used to analyze the abnormal marketing activity data and abnormal order data using the large language model to obtain abnormal report information for the abnormal marketing activities.

9. An electronic device, characterized in that, include: processor; And a memory for storing a computer program, wherein after the electronic device is powered on and the computer program is run by the processor, it performs the method described in any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer execution instructions, which are executed by a processor to perform the method described in any one of claims 1-7.

Images