Intelligent early warning method, device and equipment and storage medium

Abstract

This invention relates to the field of artificial intelligence technology. It discloses an intelligent early warning method, apparatus, device, and storage medium. The method includes: acquiring an early warning configuration file corresponding to a user identifier and all deployment field data; calculating indicators for all deployment field data to obtain multiple indicator data; generating a matrix from all deployment field data and all indicator data to obtain a matrix to be processed; inputting the early warning configuration file and the matrix to be processed into an early warning hosting model, performing early warning condition detection on the matrix to be processed, and obtaining detection results; determining whether the detection results contain early warning items; and performing trend analysis and action recommendation on the matrix to be processed and all early warning items through a deployment action recommendation model to obtain early warning results. Therefore, this invention achieves automatic detection of early warning items through an early warning hosting model, automatic trend analysis and action recommendation, timely early warning and action issuance, avoiding escalation of losses, and saving deployment costs.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent early warning method, device, equipment, and storage medium. Background Technology

[0002] Currently, advertising plays a crucial role in existing commercial promotion activities, and the quality of advertising strategies is paramount. In current advertising practices, advertising optimization specialists often manually monitor multiple advertising campaigns in real time, calculate the monitoring metrics for each client, and adjust the advertising strategy for each ad based on the client's configuration requirements. Due to the lag and subjectivity of manual decision-making, problems such as budget overspending, missing the optimal bidding time, and unsatisfactory conversion results are prone to occur, ultimately resulting in wasted client costs. Summary of the Invention

[0003] This invention provides an intelligent early warning method, device, computer equipment, and storage medium, which realizes automatic trend analysis and action recommendation when there is an early warning project, recommends the action to be taken, can make early warnings and take actions in a timely manner, avoid the expansion of losses, and save deployment costs.

[0004] An intelligent early warning method includes:

[0005] Obtain the alert configuration file and all delivery field data corresponding to the user identifier;

[0006] Calculate metrics for all the aforementioned delivery fields to obtain multiple metric data;

[0007] A matrix is ​​generated from all the aforementioned delivery field data and all the aforementioned indicator data to obtain the matrix to be processed;

[0008] The warning configuration file and the matrix to be processed are input into the warning management model. The warning management model is used to detect the warning conditions of the matrix to be processed to obtain the detection results.

[0009] Determine whether any warning items are present in the test results;

[0010] If any warning item exists in the detection results, the matrix to be processed and the warning item in the detection results are input into the action recommendation model. The action recommendation model performs trend analysis and action recommendation on the matrix to be processed and all the warning items to obtain the warning result.

[0011] An intelligent early warning device includes:

[0012] The acquisition module is used to acquire the early warning configuration file and all field data corresponding to the user identifier;

[0013] The calculation module is used to perform indicator calculations on all the data in the delivery fields to obtain multiple indicator data.

[0014] The generation module is used to generate a matrix from all the aforementioned delivery field data and all the aforementioned indicator data to obtain a matrix to be processed.

[0015] The detection module is used to input the early warning configuration file and the matrix to be processed into the early warning management model, and to perform early warning condition detection on the matrix to be processed through the early warning management model to obtain the detection result;

[0016] The judgment module is used to determine whether the detection result contains any warning items;

[0017] The early warning module is used to input the matrix to be processed and the early warning items in the detection results into the action recommendation model if any early warning item exists in the detection results. The action recommendation model then performs trend analysis and action recommendation on the matrix to be processed and all the early warning items to obtain the early warning result.

[0018] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described intelligent early warning method.

[0019] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described intelligent early warning method.

[0020] The intelligent early warning method, device, computer equipment, and storage medium provided by this invention acquire an early warning configuration file corresponding to a user identifier and all deployment field data; perform indicator calculations on all deployment field data to obtain multiple indicator data; generate a matrix from all deployment field data and all indicator data to obtain a matrix to be processed; input the early warning configuration file and the matrix to be processed into an early warning management model, and perform early warning condition detection on the matrix to be processed through the early warning management model to obtain a detection result; determine whether the detection result contains any early warning items; if the detection result contains any early warning items, input the matrix to be processed and the early warning items in the detection result into a deployment action recommendation model, and perform trend analysis and action recommendation on the matrix to be processed and all early warning items through the deployment action recommendation model to obtain an early warning result. Therefore, it realizes the automatic generation of a matrix to be processed corresponding to the current user identifier through indicator calculation and matrix generation, and automatically detects whether there are any early warning items that need to be warned through the early warning management model, achieving intelligent early warning. When there are early warning items, automatic trend analysis and action recommendation can automatically recommend deployment actions, enabling timely early warning and action, avoiding the expansion of losses, and saving deployment costs. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the application environment of the intelligent early warning method in one embodiment of the present invention;

[0023] Figure 2 This is a flowchart of an intelligent early warning method according to an embodiment of the present invention;

[0024] Figure 3 This is a flowchart of step S20 of the intelligent early warning method in one embodiment of the present invention;

[0025] Figure 4 This is a schematic block diagram of an intelligent early warning device according to an embodiment of the present invention;

[0026] Figure 5 This is a schematic block diagram of the detection module of an intelligent early warning device in one embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The intelligent early warning method provided by this invention can be applied to, for example... Figure 1 In this application environment, the client (computer device or terminal) communicates with the server via a network. The client (computer device or terminal) includes, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0030] In one embodiment, such as Figure 2 As shown, an intelligent early warning method is provided, the technical solution of which mainly includes the following steps S10-S60:

[0031] S10: Obtain the alert configuration file and all delivery field data associated with the ad identifier.

[0032] Understandably, the advertising identifier is to assign a unique identifier to different advertising objects. An advertising object can perform platform operations on the advertising platform through the advertising identifier, such as monitoring advertising data, adjusting advertising delivery plans, etc. An advertising identifier corresponds to an early warning configuration file. The early warning configuration file is a file containing a set of early warning parameters configured for the delivery strategy of the advertising object corresponding to the advertising identifier. The delivery field data is the data of relevant fields involved in the delivery process of the advertising object corresponding to the advertising identifier.

[0033] In one embodiment, prior to step S10, i.e., before obtaining the alert configuration file associated with the ad identifier and all delivery field data, the following steps are included:

[0034] Obtain the advertising type and warning requirements corresponding to the advertising identifier.

[0035] Understandably, the ad type refers to the platform type for ad delivery, such as a broadcast platform, and the warning requirement refers to a warning requirement document set for the ad object corresponding to the ad identifier.

[0036] The warning requirement is analyzed by identifying warning attribute keywords, and the warning attribute results are obtained.

[0037] Understandably, the process of identifying warning attribute keywords for the warning requirement can be implemented using a trained keyword recognition model. The network structure of the keyword recognition model can be a BERT-based network structure. The warning requirement is segmented into words by the keyword recognition model. This segmentation involves dividing the text in the warning requirement into the smallest unit of characters or words. Then, the keyword recognition model extracts warning attribute features from each of the smallest units of characters or words. Based on the extracted warning attribute features, they are matched with each warning attribute to determine the probability of each warning attribute. The smallest units of characters or words that are greater than or equal to a threshold are recorded as the identified keywords. In this way, all keywords related to the warning attribute in the warning requirement can be identified, and the identified keywords are recorded as the warning attribute result.

[0038] The warning attribute features are features related to the project or field being warned, such as click-through rate, daily consumption, conversion rate, and other related or similar noun features.

[0039] Based on the warning attribute results, the warning requirement is subjected to contextual semantic recognition to obtain the attribute configuration results corresponding to the warning attribute results.

[0040] Understandably, the contextual semantic recognition can employ the Bi-LSTM algorithm, also known as the bidirectional long short-term memory network algorithm, which uses co-encoding in both forward and reverse directions to perform embedded word vector conversion verification. This ensures that unit characters or words are converted into the most semantically consistent codes, extracting the true entities from the codes of each unit character or word, removing non-entity words such as function words and auxiliary words, and determining the extracted entities as the unit characters or words to be processed. The warning attribute results are then removed from these extracted characters or words, leaving only the unit characters or words that do not contain the warning attribute results. Each remaining unit character or word is then matched with the keywords in each of the warning attribute results to obtain the attribute configuration results corresponding to each keyword in the warning attribute results. The matching process involves matching each keyword with each remaining unit character or word. The process involves converting the word vectors of keywords into word vectors, then calculating the distance between the word vectors of keywords and unit characters or words within their conceptual domains. This involves calculating the Euclidean distance between the higher-level conceptual domains of the keyword's word vector and the higher-level conceptual domains of the unit character or word's word vector, resulting in the first association distance. Next, the dictionary distance between the keyword's word vector and the unit character or word's word vector is calculated, which is the Euclidean distance between the two word vectors in the dictionary, resulting in the second association distance. Finally, the text distance between the keyword and the unit character or word is calculated, which is the word count distance between characters in the warning requirement, resulting in the third association distance. Finally, the association value is obtained by weighted summing the first, second, and third association distances. Keywords and unit characters or words with association values ​​less than or equal to a preset value are recorded as the unit characters or words corresponding to the keyword in the attribute configuration results.

[0041] Based on the ad type and the attribute configuration result, the warning configuration file corresponding to the ad identifier is generated.

[0042] Understandably, the template library stores configuration templates corresponding to all ad types. One ad type corresponds to one configuration template. The configuration template corresponding to the ad type is retrieved from the template library, and the attribute configuration results are automatically filled into the positions of the corresponding keywords according to their respective positions, thereby generating the warning configuration file corresponding to the ad identifier.

[0043] This invention achieves the following: First, it obtains the advertising type and alert requirement corresponding to the user identifier. Then, it identifies the alert attribute keywords for the alert requirement to obtain the alert attribute result. Based on the alert attribute result, it performs contextual semantic recognition on the alert requirement to obtain the attribute configuration result corresponding to the alert attribute result. Finally, it generates the alert configuration file corresponding to the user identifier based on the advertising type and the attribute configuration result. This allows for the automatic identification of the account's alert configuration file based on the advertising type and alert requirement of the advertising identifier, eliminating the need for manual configuration, reducing costs, and improving efficiency.

[0044] S20, calculate the metrics for all the data in the delivery fields to obtain multiple metric data.

[0045] Understandably, the indicator data refers to the data of relevant indicators calculated according to preset indicator rules based on the relevant advertising field data. For example, the indicator data is the daily consumption, which is the difference between the daily consumption cost and the daily revenue cost in the advertising field data; the indicator data is the conversion rate, which is the ratio of clicks to impressions in the advertising field data, and so on.

[0046] In one embodiment, such as Figure 3 As shown, in step S20, i.e., calculating the metrics for all the data in the delivery fields to obtain multiple metric data, including:

[0047] S201, based on preset indicator rules, filter indicator attribute data from all the data in the delivery fields.

[0048] Understandably, the preset indicator rules are rules set according to requirements, and the indicator attribute data are the delivery field data included in the preset indicator rules. The indicator attribute data is selected from all the delivery field data.

[0049] S202, the indicator attribute data is analyzed by each indicator model in the preset indicator rules to obtain the indicator data.

[0050] Understandably, the indicator model is a model established to output indicator data according to preset indicator rules, and the indicator analysis is the process of analyzing indicator data according to preset indicator rules. The indicator data includes daily consumption and conversion rate. The indicator attribute data, namely daily consumption cost and daily revenue cost, are input into the daily consumption analysis model, and the daily consumption indicator data is output through the daily consumption analysis model. The indicator attribute data, namely clicks and impressions, are input into the conversion rate analysis model, and the conversion rate indicator data is output through the conversion rate analysis model. For example, daily consumption is the difference between daily consumption cost and daily revenue cost in the campaign field data; the conversion rate is the ratio of clicks to impressions in the campaign field data.

[0051] This invention enables the filtering of indicator attribute data from all the data in the delivery fields based on preset indicator rules; and the analysis of the indicator attribute data using the indicator models in the preset indicator rules to obtain the indicator data. In this way, the indicator rules that need to be focused on can be preset automatically, and the corresponding indicator data can be calculated automatically, thereby improving the diversity and configurability of the indicator data.

[0052] In one embodiment, the indicator attribute data includes delivery volume attribute values, click attribute values, download attribute values, cost attribute values, and consumption attribute values.

[0053] Understandably, the delivery volume attribute is the value of the delivery quantity, the click attribute is the number of times the ad is clicked, the download attribute is the number of times the ad is downloaded, the cost attribute is the number of downloads, the cost attribute is the total cost of the ad delivery, and the consumption attribute is the cost consumed.

[0054] In one embodiment, step S202, namely, performing indicator analysis on the indicator attribute data using the indicator models in the preset indicator rules to obtain the indicator data, includes:

[0055] Click-through rate analysis is performed on the delivery volume attribute value and the click attribute value using the click metric model in the preset metric rules to obtain click metric data.

[0056] Understandably, the click metric model is a model that outputs click metric data through the delivery volume attribute value and the click attribute value, and the click-through rate analysis is an analysis process of the ratio of the click attribute value to the delivery volume attribute.

[0057] By using the conversion indicator model in the preset indicator rules, the conversion rate of the delivery volume attribute value, the click attribute value, and the download attribute value is analyzed to obtain conversion indicator data.

[0058] Understandably, the conversion metric model is a model that outputs conversion metric data through the delivery volume attribute value, the click attribute value, and the download attribute value, and the conversion rate analysis is the analysis process of the ratio of the sum of the click attribute value and the download attribute value to the delivery volume attribute value.

[0059] By using the consumption indicator model in the preset indicator rules, the consumption of the cost attribute value and the consumption attribute value is analyzed to obtain consumption indicator data.

[0060] Understandably, the consumption index model is a model that outputs consumption index data through the cost attribute value and the consumption attribute value, and the consumption analysis is the process of analyzing the difference between the cost attribute value and the consumption attribute value.

[0061] The click metric data, the conversion metric data, and the consumption metric data are recorded as the metric data.

[0062] Understandably, the click metric data, the conversion metric data, and the consumption metric data are defined as the metric data.

[0063] In this way, by analyzing click metric models, conversion metric models, and consumption metric models, click metric data, conversion metric data, and consumption metric data can be obtained respectively, and the metric data related to click-through rate, conversion rate, and consumption can be automatically output.

[0064] S30, generate a matrix from all the aforementioned field data and all the aforementioned indicator data to obtain the matrix to be processed.

[0065] Understandably, the matrix generation can be achieved by concatenating all the delivery field data and all the indicator data into a matrix to obtain the matrix to be processed. Alternatively, the matrix generation can be achieved by taking all the delivery field data as a multidimensional matrix of a specific size, concatenating all the indicator data into the multidimensional matrix, and filling the gaps with zeros until the matrix to be processed reaches a preset size. The preset size provides the matrix size basis for subsequent input into the early warning management model.

[0066] S40, input the early warning configuration file and the matrix to be processed into the early warning management model, and use the early warning management model to perform early warning condition detection on the matrix to be processed to obtain the detection result.

[0067] Understandably, the early warning management model is a trained neural network model used to predict whether the input matrix to be processed and the early warning configuration file meet the early warning conditions. The early warning condition detection process is as follows: First, the early warning configuration file is interpreted with thresholds to extract each early warning parameter, and the interpreted early warning parameters are summarized into a threshold parameter array; second, the matrix to be processed is used to predict the indicators to obtain the prediction results; finally, based on the threshold parameter array, the prediction results are used to detect the early warning conditions to obtain the detection results. The detection results reflect whether there are any early warning items in all the currently deployed field data.

[0068] Among them, the

[0069] In one embodiment, step S40, namely, performing early warning condition detection on the matrix to be processed through the early warning hosting model to obtain the detection result, includes:

[0070] The thresholds in the warning configuration file are interpreted to obtain a threshold parameter array.

[0071] Understandably, the threshold interpretation is the process of interpreting the configured warning parameters from the warning configuration file, that is, obtaining the corresponding warning parameters from all warning attributes in the warning configuration file, thereby obtaining the threshold parameter array.

[0072] The prediction results are obtained by using the early warning management model to predict the indicators of the matrix to be processed.

[0073] Understandably, the indicator prediction is as follows: the early warning hosting model is used to convolve the matrix to be processed to extract data trend features, and the extracted data trend features are compared with the data trend features of each historical data matrix. The historical data matrix with the highest similarity to the extracted data trend features is selected, and the historical prediction results associated with the historical data matrix are obtained. Based on the linear relationship between the historical prediction results and the data of each delivery field and each indicator in the historical data matrix, the data is mapped to the matrix to be processed. The prediction result of the matrix to be processed is predicted by combining the linear relationships. The prediction result reflects the result of predicting the trend of each indicator data of the current matrix to be processed.

[0074] The data trend features refer to the trend changes caused by the mutual influence between the data in the delivery field and the indicator data. In the process of extracting data trend features, the delivery field data and the indicator data associated with the delivery field data are extracted together. That is, while convolving the surrounding data, the convolution between the delivery field data and the associated indicator data is also combined, so as to extract more data trend features and improve the accuracy of the prediction results.

[0075] Based on the threshold parameter array, the prediction results are subjected to early warning condition detection to obtain the detection results.

[0076] Understandably, the threshold parameter array also includes conditional rules for each warning parameter. These conditional rules are logical conditions for selecting the warning parameters, such as: greater than, less than, equal to, greater than or equal to, between, etc. The warning condition detection involves comparing each warning parameter in the threshold parameter array with the indicator data in the prediction result to obtain a comparison result. Then, the comparison result is matched with the conditional rules in the threshold parameter array to determine whether each item in the comparison result meets the corresponding conditional rule. If it meets the corresponding conditional rule, the item is determined to be a warning item; if it does not meet the corresponding conditional rule, the item is determined not to be a warning item. After traversing all items in the comparison result, all warning items obtained are recorded as the detection result. The detection result reflects whether there is a warning item in the matrix to be processed.

[0077] This invention achieves the following: by interpreting the threshold of the early warning configuration file to obtain a threshold parameter array; by using the early warning management model to predict the indicators of the matrix to be processed to obtain a prediction result; and by using the threshold parameter array to detect early warning conditions based on the prediction result to obtain the detection result. In this way, early warning items can be automatically detected through threshold interpretation and early warning management model without manual identification, achieving intelligent management and timely early warning.

[0078] S50, determine whether there is a warning item in the detection result.

[0079] S60, if any warning item exists in the detection results, input the matrix to be processed and the warning item in the detection results into the delivery action recommendation model, and perform trend analysis and action recommendation on the matrix to be processed and all the warning items through the delivery action recommendation model to obtain the warning result.

[0080] Understandably, if any warning item exists in the detection results, the matrix to be processed and all the warning items in the detection results are input into the action recommendation model. The action recommendation model performs trend analysis and action recommendation on the matrix to be processed and all the warning items. The trend analysis process involves concatenating the matrix to be processed and all the warning items to obtain the current data matrix, filling any missing positions with zeros during the concatenation process. The historical user data matrix is ​​then analyzed by extracting action features. The action recommendation process involves predicting actions based on the extracted action features, predicting the action category and action result, and using the predicted action category and action result as the warning result. The warning result reflects the warning content predicted by the current warning configuration file and all delivery field data. Through the warning result, future actions and their consequences can be recommended to users or relevant staff, enabling timely measures to be taken and greatly reducing the time spent by users or relevant staff.

[0081] This invention achieves intelligent early warning by acquiring the early warning configuration file and all delivery field data corresponding to the user identifier; calculating indicators for all delivery field data to obtain multiple indicator data; generating a matrix from all delivery field data and all indicator data to obtain a matrix to be processed; inputting the early warning configuration file and the matrix to be processed into an early warning management model, and using the early warning management model to detect early warning conditions in the matrix to be processed to obtain detection results; determining whether the detection results contain any early warning items; if the detection results contain any early warning items, inputting the matrix to be processed and the early warning items in the detection results into a delivery action recommendation model, and using the delivery action recommendation model to perform trend analysis and action recommendation on the matrix to be processed and all early warning items to obtain early warning results. Therefore, it realizes the automatic generation of the matrix to be processed corresponding to the current user identifier through indicator calculation and matrix generation, and the automatic detection of whether there are early warning items that need to be warned through the early warning management model, achieving intelligent early warning. When early warning items exist, automatic trend analysis and action recommendation can automatically recommend delivery actions, enabling timely early warning and action, avoiding the expansion of losses, and saving delivery costs.

[0082] In one embodiment, before step S60, that is, before inputting the matrix to be processed and the warning items in the detection results into the action recommendation model, the following steps are included:

[0083] Obtain a historical user dataset; the historical user dataset includes a historical user data matrix, as well as action labels and action execution results associated with the historical user data matrix.

[0084] Understandably, the historical user dataset is a collection of all the historical user data matrices. The historical user data matrix is ​​obtained by generating a matrix from historical delivery field data and indicator data obtained after indicator calculation, and then combining it with historical early warning items. The historical user data matrix is ​​a data matrix composed of historical delivery field data after the occurrence of historical early warning items. Each historical user data matrix is ​​associated with an action label and an action execution result. The action label is the category of the action executed in the past, and the action execution result is the set of results obtained after the historical user data matrix executes the associated action label.

[0085] The historical user data matrix is ​​input into a deep learning model containing initial parameters.

[0086] Understandably, the deep learning model includes initial parameters that change during iterative training, and the deep learning model is a neural network model based on deep learning.

[0087] The deep learning model is used to extract action features from the historical user data matrix, and action prediction is performed based on the extracted action features to obtain the predicted action category and the predicted action result.

[0088] Understandably, the action features are features associated with the predicted action data. The extraction of the action features is a process of convolving the historical user data matrix and extracting data-associated features from the historical user data matrix during the convolution process. Action prediction is performed based on the extracted action features. The action prediction process involves classifying the extracted action features to obtain the action category with the highest probability, recording this action category as the predicted action category, and then performing result prediction on the historical user data matrix after executing the predicted action category. That is, after executing the predicted action category, the process of predicting the next value of each delivery field data and indicator data in the historical user data matrix is ​​performed to obtain the corresponding predicted action result.

[0089] Based on the cross-entropy loss algorithm, a first loss value is obtained according to the predicted action category and the action label associated with the historical user data matrix, and a second loss value is obtained according to the predicted action result and the action execution result associated with the historical user data matrix.

[0090] Understandably, a first loss value and a second loss value are calculated using the cross-entropy loss algorithm. The first loss value reflects the difference distance between the predicted action category and the action label associated with the historical user data matrix, and the second loss value reflects the difference distance between the predicted action result and the action execution result associated with the historical user data matrix.

[0091] The first loss value and the second loss value are weighted to obtain the total loss value.

[0092] If the total loss value does not reach the convergence condition, the initial parameters in the deep learning model are iteratively updated, and the step of extracting action features from the historical user data matrix through the deep learning model is executed until the total loss value reaches the convergence condition. The converged deep learning model is then recorded as the action recommendation model.

[0093] Understandably, when the total loss value does not reach the preset convergence condition, the initial parameters in the deep learning model are iteratively updated to continuously train and learn, and the step of extracting action features from the historical user data matrix through the deep learning model is returned to be executed. This process is repeated until the convergence condition is reached. The convergence condition can be that the total loss value is very small after 20,000 calculations and will not decrease further. That is, when the total loss value is very small after 20,000 calculations and will not decrease further, training stops, and the converged deep learning model is recorded as the action recommendation model. Alternatively, the convergence condition can be that the total loss value is less than a set threshold. That is, when the total loss value is less than a set threshold, training stops, and the converged deep learning model is recorded as the action recommendation model.

[0094] Thus, by using action feature extraction and deep learning training methods, an action recommendation model can be automatically trained. By gradually narrowing the gap between predicted action categories and action labels, as well as between predicted results and action execution results, the action recommendation model can be converged and trained, thereby improving the accuracy and reliability of the early warning output.

[0095] In one embodiment, after step S50, i.e. after determining whether the detection result contains a warning item, the following steps are included:

[0096] If the detection results do not contain any warning items, the system periodically scans all the delivery field data corresponding to the advertising identifier.

[0097] In this way, even when there are no warning projects, the system can periodically scan the corresponding delivery field data of the advertising identifier to promptly detect whether any warnings will appear in the delivery field data, thus avoiding manual monitoring.

[0098] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0099] In one embodiment, an intelligent early warning device is provided, which corresponds one-to-one with the intelligent early warning method described in the above embodiments. For example... Figure 4 As shown, the intelligent early warning device includes an acquisition module 11, a calculation module 12, a generation module 13, a detection module 14, a judgment module 15, and an early warning module 16. Detailed descriptions of each functional module are as follows:

[0100] Module 11 is used to obtain the early warning configuration file and all field data corresponding to the user identifier;

[0101] Calculation module 12 is used to perform indicator calculations on all the data in the delivery fields to obtain multiple indicator data;

[0102] The generation module 13 is used to generate a matrix from all the delivery field data and all the indicator data to obtain a matrix to be processed;

[0103] The detection module 14 is used to input the early warning configuration file and the matrix to be processed into the early warning management model, and to perform early warning condition detection on the matrix to be processed through the early warning management model to obtain the detection result;

[0104] The judgment module 15 is used to determine whether the detection result contains any warning items;

[0105] The early warning module 16 is used to input the matrix to be processed and the early warning items in the detection results into the delivery action recommendation model if any early warning item exists in the detection results. The delivery action recommendation model performs trend analysis and action recommendation on the matrix to be processed and all the early warning items to obtain the early warning result.

[0106] In one embodiment, such as Figure 5 As shown, the detection module 14 includes:

[0107] Interpretation unit 41 is used to interpret the warning configuration file for thresholds to obtain a threshold parameter array;

[0108] Prediction unit 42 is used to predict indicators of the matrix to be processed through the early warning management model to obtain prediction results;

[0109] The detection unit 43 is used to perform early warning condition detection on the prediction result based on the threshold parameter array, and obtain the detection result.

[0110] Specific limitations regarding intelligent early warning devices can be found in the limitations of intelligent early warning methods described above, and will not be repeated here. Each module in the aforementioned intelligent early warning device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0111] In one embodiment, a computer device is provided, which may be a client or a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a readable storage medium and internal memory. The readable storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the readable storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an intelligent early warning method.

[0112] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent early warning method described in the above embodiment.

[0113] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the intelligent early warning method described above.

[0114] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0115] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0116] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An intelligent early warning method, characterized in that, include: Obtain the alert configuration file and all delivery field data corresponding to the user identifier; Calculate metrics for all the aforementioned delivery fields to obtain multiple metric data; A matrix is ​​generated from all the aforementioned delivery field data and all the aforementioned indicator data to obtain the matrix to be processed; The warning configuration file and the matrix to be processed are input into the warning management model. The warning management model is used to detect the warning conditions of the matrix to be processed to obtain the detection results. Determine whether any warning items are present in the test results; If any warning item exists in the detection results, the matrix to be processed and the warning item in the detection results are input into the action recommendation model. The action recommendation model is used to perform trend analysis and action recommendation on the matrix to be processed and all the warning items to obtain the warning result. Before obtaining the warning configuration file corresponding to the user identifier, the following steps are included: Obtain the advertising type and alert requirements corresponding to the user identifier; The warning requirements are analyzed using warning attribute keywords to identify the warning attribute results. Based on the warning attribute results, the warning requirement is subjected to contextual semantic recognition to obtain the attribute configuration results corresponding to the warning attribute results; Based on the advertisement type and the attribute configuration result, generate the warning configuration file corresponding to the user identifier; The step of performing early warning condition detection on the matrix to be processed through the early warning hosting model to obtain detection results includes: The thresholds in the warning configuration file are interpreted to obtain a threshold parameter array; The prediction results are obtained by using the early warning management model to predict the indicators of the matrix to be processed. Based on the threshold parameter array, the prediction result is subjected to early warning condition detection to obtain the detection result; Specifically, each early warning parameter in the threshold parameter array is compared with the indicator data in the prediction result to obtain the comparison result. Then, it is determined whether each item in the comparison result meets the corresponding condition rule. If the item meets the corresponding condition rule, it is determined as an early warning item.

2. The intelligent early warning method as described in claim 1, characterized in that, The step involves calculating metrics for all the data in the delivery fields to obtain multiple metric data, including: Based on preset indicator rules, indicator attribute data is filtered out from all the data in the delivery fields; The indicator data is obtained by analyzing each indicator model in the preset indicator rules to obtain the indicator attribute data.

3. The intelligent early warning method as described in claim 2, characterized in that, The indicator attribute data includes delivery volume attribute values, click attribute values, download attribute values, cost attribute values, and consumption attribute values; the indicator data is obtained by analyzing the indicator attribute data through the indicator models in the preset indicator rules, including: Click-through rate analysis is performed on the delivery volume attribute value and the click attribute value using the click metric model in the preset metric rules to obtain click metric data. By using the conversion indicator model in the preset indicator rules, the conversion rate of the delivery volume attribute value, the click attribute value, and the download attribute value is analyzed to obtain conversion indicator data; By using the consumption indicator model in the preset indicator rules, the consumption of the cost attribute value and the consumption attribute value is analyzed to obtain consumption indicator data. The click metric data, the conversion metric data, and the consumption metric data are recorded as the metric data.

4. The intelligent early warning method as described in claim 1, characterized in that, Before inputting the matrix to be processed and the warning items in the detection results into the action recommendation model, the following steps are included: Obtain a historical user dataset; the historical user dataset includes a historical user data matrix, as well as action labels and action execution results associated with the historical user data matrix; Input the historical user data matrix into a deep learning model containing initial parameters; The deep learning model is used to extract action features from the historical user data matrix, and action prediction is performed based on the extracted action features to obtain the predicted action category and the predicted action result. Based on the cross-entropy loss algorithm, a first loss value is obtained according to the predicted action category and the action label associated with the historical user data matrix, and a second loss value is obtained according to the predicted action result and the action execution result associated with the historical user data matrix. The first loss value and the second loss value are weighted to obtain the total loss value; If the total loss value does not reach the convergence condition, the initial parameters in the deep learning model are iteratively updated, and the step of extracting action features from the historical user data matrix through the deep learning model is executed until the total loss value reaches the convergence condition. The converged deep learning model is then recorded as the action recommendation model.

5. An intelligent early warning device, characterized in that, include: The acquisition module is used to acquire the early warning configuration file and all field data corresponding to the user identifier; The calculation module is used to perform indicator calculations on all the data in the delivery fields to obtain multiple indicator data. The generation module is used to generate a matrix from all the aforementioned delivery field data and all the aforementioned indicator data to obtain a matrix to be processed. The detection module is used to input the early warning configuration file and the matrix to be processed into the early warning management model, and to perform early warning condition detection on the matrix to be processed through the early warning management model to obtain the detection result; The judgment module is used to determine whether the detection result contains any warning items; The early warning module is used to input the matrix to be processed and the early warning items in the detection results into the delivery action recommendation model if any early warning item exists in the detection results. The delivery action recommendation model performs trend analysis and action recommendation on the matrix to be processed and all the early warning items to obtain the early warning result. Before obtaining the warning configuration file corresponding to the user identifier, the following steps are included: Obtain the advertising type and alert requirements corresponding to the user identifier; The warning requirements are analyzed using warning attribute keywords to identify the warning attribute results. Based on the warning attribute results, the warning requirement is subjected to contextual semantic recognition to obtain the attribute configuration results corresponding to the warning attribute results; Based on the advertisement type and the attribute configuration result, generate the warning configuration file corresponding to the user identifier; The detection module includes: The interpretation unit is used to interpret the warning configuration file for thresholds to obtain a threshold parameter array; The prediction unit is used to predict the indicators of the matrix to be processed through the early warning management model and obtain the prediction results. A detection unit is used to perform early warning condition detection on the prediction result based on the threshold parameter array, and obtain the detection result; Specifically, each early warning parameter in the threshold parameter array is compared with the indicator data in the prediction result to obtain the comparison result. Then, it is determined whether each item in the comparison result meets the corresponding condition rule. If the item meets the corresponding condition rule, it is determined as an early warning item.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent early warning method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent early warning method as described in any one of claims 1 to 4.

Images