Advertisement putting optimization system and method based on natural language processing

Abstract

The invention provides an advertisement putting optimization system and method based on natural language processing. The system comprises at least one advertisement main end and an optimization server. And the optimization server receives an input data file of the advertiser terminal, performs natural language analysis, processes the input data file through a built-in advertisement putting optimization model to obtain an optimization result, and feeds back the optimization result to the advertiser terminal. And the advertiser modifies the input data file according to the optimization result to generate an output data file. And the optimization server further evaluates the optimization effect and performs self-learning to update the model. According to the method, the advertisement content can be optimized, the advertisement effect can be predicted, the advertisement putting efficiency and effect are improved, and continuous learning and improvement of an advertisement putting optimization model are realized.

Description

Technical Field

[0001] The present invention relates to the field of digital marketing, and more specifically, to an advertisement delivery optimization system and method based on natural language processing. Background Art

[0002] In the field of digital marketing, advertising is an important means to attract potential customers and enhance brand awareness. With the development of Internet technology, advertisers can place advertisements through various online platforms in order to reach a wider audience. However, existing advertising delivery systems often lack the ability to deeply understand and optimize advertising content, resulting in unsatisfactory advertising results. In addition, the formulation of advertising delivery strategies usually relies on the experience and intuition of advertisers, and lacks data-driven scientific methods to guide the creation of advertising content and the adjustment of delivery strategies.

[0003] The optimization of advertising content usually needs to consider multiple factors such as keyword selection, content attractiveness, and relevance to the target audience. However, existing systems are often unable to automatically identify problems in advertising content, such as incorrect keyword selection, missing content, or content conflicts, which directly affect the click-through rate and conversion rate of the advertisement. At the same time, the prediction of advertising effectiveness is also a complex issue that requires comprehensive consideration of multiple factors such as advertising content, audience characteristics, and market trends. Traditional prediction methods often rely on simple statistical models and cannot accurately capture the interactions between these complex factors.

[0004] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: the existing advertising delivery system lacks the ability to deeply analyze and optimize the advertising content, resulting in the difficulty of achieving the expected advertising delivery effect; at the same time, the prediction of advertising effect lacks accuracy and reliability, making it difficult for advertisers to formulate the best advertising delivery strategy. In addition, the existing system also lacks the ability of self-learning and self-improvement, and cannot continuously improve the effect of advertising delivery as time goes by and data accumulates. Summary of the invention

[0005] The present invention provides an advertisement delivery optimization system and method based on natural language processing.

[0006] In a first aspect of the present invention, a system for optimizing advertisement delivery based on natural language processing is provided, comprising: In a second aspect of the present invention, a method for optimizing advertisement delivery based on natural language processing is provided, comprising: The above-mentioned embodiments of the present invention have at least the following beneficial effects: by adopting natural language processing technology, the advertising delivery optimization system of the present invention can deeply analyze the advertising content, identify and optimize problems such as incorrect keyword selection, missing content and content conflicts, thereby improving the quality and relevance of the advertising content. The advertising delivery optimization model built into the system can process the parsed data file, output the optimization results, and guide advertisers to modify the content to further improve the attractiveness and click-through rate of the advertisement. In addition, the system continuously learns from the advertising delivery optimization record data through a self-learning mechanism, and automatically updates the optimization model parameters, so that the advertising delivery strategy can adapt to market changes and continuously improve the advertising effect.

[0007] In addition, the advertising delivery optimization system of the present invention also includes an advertising effect prediction model, which can predict the subsequent effect of advertising delivery based on advertising content data, and provide data-driven decision support for advertisers. The evaluation module of the system can evaluate the advertising delivery optimization results and guide the subsequent actions of advertisers through quantified positive or negative feedback values. This data-based evaluation mechanism can ensure the continuous optimization of advertising delivery strategies and help advertisers achieve more efficient advertising delivery and better returns on investment. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which: Figure 1 A schematic diagram of the structure of an advertisement delivery optimization system based on natural language processing provided by an embodiment of the present invention; Figure 2 A flowchart of an advertisement delivery optimization method based on natural language processing provided by an embodiment of the present invention; Figure 3 The schematic diagram schematically shows the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0010] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0011] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0012] Reference below Figure 1 , Figure 1 The following is a schematic diagram of the structure of an advertisement delivery optimization system based on natural language processing provided by an embodiment of the present invention. Figure 1 As shown, an advertisement delivery optimization system 100 based on natural language processing includes: At least one advertiser terminal 101 and an optimization server 102; The optimization server is connected to each advertiser terminal for data interaction; The optimization server receives an input data file including advertisement content data generated by an advertiser through a certain advertiser terminal during the advertisement delivery process; after natural language analysis of the input data file, the optimization result is obtained by processing it through a built-in advertisement delivery optimization model, and the optimization result is fed back to the corresponding advertiser terminal; the advertiser modifies and / or supplements the input data file according to the optimization result through the advertiser terminal to obtain an output data file; The optimization server receives the output data file sent by the advertiser, evaluates the optimization result of this advertisement delivery according to the corresponding input data file, output data file and optimization result, and updates the advertisement delivery optimization model through self-learning based on the evaluation result.

[0013] It should be noted that the present invention describes an advertising delivery optimization system that optimizes advertising content and improves the effectiveness of advertising delivery through natural language processing technology. Here, natural language parsing refers to the process of using computer algorithms to understand and analyze human natural language text. This process includes but is not limited to steps such as word segmentation, part-of-speech tagging, syntactic analysis, and semantic understanding, with the purpose of extracting useful information from the text, such as keywords, entities, relationships, etc., for further processing.

[0014] Specifically, the natural language parsing module is the core component of the optimization server, which is responsible for converting the advertising content data into a machine-understandable form. This process may involve the use of pre-trained language models such as BERT or GPT, which can be trained on large amounts of text data to capture the complex patterns and semantics of language.

[0015] More specifically, the specific parameter settings for natural language parsing may include selecting a specific pre-trained model, adjusting the model's hyperparameters (such as learning rate, batch size, etc.), and customized training for specific industries or fields. In addition, natural language parsing can also include sentiment analysis to identify the emotional tendencies in advertising content, thereby better adjusting advertising strategies.

[0016] Preferably, the process of natural language parsing can be further refined into the following steps: first, the advertising text is segmented to divide the continuous text into meaningful words or phrases; then, part-of-speech tagging is performed to specify the grammatical role of each word, such as noun, verb, etc.; then, syntactic analysis is performed to construct the grammatical structure of the sentence and identify components such as the subject, predicate and object; finally, semantic understanding is performed to extract key information and concepts in the text, as well as the relationship between them.

[0017] Furthermore, machine learning algorithms can be used to optimize the accuracy of natural language parsing, for example, by adjusting model parameters through supervised learning to better adapt them to specific advertising content and industry terms. In some cases, multimodal natural language processing technology can also be considered to process advertising content containing images and videos, thereby achieving more comprehensive advertising content analysis and optimization.

[0018] In some embodiments, the optimization server includes a natural language parsing module, an evaluation module, an advertisement delivery optimization model, and a self-learning training database; The natural language parsing module is used to parse the input data file or the output data file using a natural language processing method to obtain the parsed input data file Inp or the output data file Out; The advertising delivery optimization model is used to process the parsed input data file Inp to obtain an optimization result Opt; The evaluation module is used to evaluate the optimization result of each advertisement delivery according to the parsed input data file Inp, output data file Out and optimization result Opt corresponding to each advertisement delivery process to obtain an evaluation result Eval; The self-learning training database is used to store the parsed input data file Inp, output data file Out, optimization result Opt and evaluation result Eval corresponding to each advertisement delivery optimization as a set of advertisement delivery optimization record data; The advertising delivery optimization model is a self-learning machine learning model, which performs self-learning training by retrieving multiple groups of advertising delivery optimization record data from a self-learning training database to update model parameters.

[0019] It should be noted that this embodiment further describes in detail the composition and functions of the optimization server. Here, the optimization server refers to the background system responsible for processing and optimizing the advertising content, which integrates multiple modules to achieve natural language analysis of the advertising content, effect evaluation and self-learning training of the optimization model.

[0020] Specifically, the optimization server includes a natural language analysis module, an evaluation module, an advertisement delivery optimization model, and a self-learning training database. The natural language analysis module is responsible for language processing of input data files and extracting key information.

[0021] More specifically, the evaluation module evaluates the advertising effect based on the optimization results and the output data file. The advertising optimization model is based on machine learning technology, which can process the parsed data file and output the optimization result.

[0022] Furthermore, the self-learning training database is used to store the record data of each advertising delivery optimization for the model to perform self-learning training and update model parameters. The specific parameter settings of these modules include selecting a specific machine learning algorithm, adjusting the model's hyperparameters, and customized training for specific industries or fields.

[0023] Preferably, the natural language parsing module can adopt natural language processing technology based on deep learning, such as recurrent neural network (RNN) or long short-term memory network (LSTM), to improve the ability to understand the advertising content.

[0024] Specifically, the evaluation module can be designed as a multi-dimensional evaluation system that not only considers the click-through rate and conversion rate of the advertisement, but also includes indicators such as user engagement and advertisement coverage. The advertisement delivery optimization model can be an integrated model that combines multiple sub-models to handle different types of advertisement content and delivery strategies.

[0025] More specifically, the self-learning training database can adopt a distributed storage system to handle large-scale data sets and ensure the security and accessibility of the data. In addition, it is also possible to consider using reinforcement learning technology to optimize the self-learning process so that the model can adapt more flexibly to market changes and user behavior.

[0026] In some embodiments, the advertisement delivery optimization model includes a first advertisement delivery optimization model and / or a second advertisement delivery optimization model; The first advertising delivery optimization model is an advertising content optimization model, which is used to identify problems including incorrect keyword selection, missing content, and content conflict in the parsed input data file Inp including advertising content data, and output optimization results; after the optimization results are fed back to the advertiser side, the advertiser modifies the input data file at the advertiser side with reference to the optimization results to obtain an output data file; The second advertising delivery optimization model is an advertising effect prediction model, which is used to generate prediction results for subsequent advertising delivery based on the parsed input data file Inp including advertising content data; after the prediction results are fed back to the advertiser side, the advertiser supplements the advertising delivery strategy of the input data file with reference to the prediction results on the advertiser side to obtain an output data file.

[0027] It should be noted that in the advertising delivery optimization system, special attention is paid to the composition of the advertising delivery optimization model, which includes the first advertising delivery optimization model and / or the second advertising delivery optimization model. Here, the first advertising delivery optimization model refers to a model specifically used to optimize advertising content, and the second advertising delivery optimization model refers to a model used to predict advertising effects.

[0028] Specifically, the first advertising delivery optimization model is an advertising content optimization model, and its task is to identify problems such as incorrect keyword selection, missing content, and content conflicts in the advertising content data, and output corresponding optimization results. The identification of these problems can be achieved through natural language processing technology, such as using semantic analysis to identify the accuracy of keywords and using machine learning algorithms to detect the completeness and consistency of content.

[0029] More specifically, after the optimization results are fed back to the advertiser, the advertiser can modify the input data file based on these results to improve the quality of the advertising content. The second advertising delivery optimization model generates prediction results for subsequent advertising delivery based on the advertising content data, which may involve using historical data and market trend analysis to predict the potential effect of advertising.

[0030] Preferably, the first advertising delivery optimization model can use a rule-based approach combined with machine learning technology to improve the accuracy of identifying problems. For example, a series of predefined rules can be set to identify common content errors, while using a machine learning model to identify more complex or less obvious problems.

[0031] Furthermore, the second advertising delivery optimization model can use time series analysis or prediction market response models to predict advertising effects, and these models can adjust their predictions based on the historical performance of advertisements and market dynamics. In addition, a user feedback mechanism can be introduced to further refine the prediction results, for example, by analyzing the user's click, comment and share behavior on the advertisement to adjust the model's prediction.

[0032] In some embodiments, the first advertising delivery optimization model or the second advertising delivery optimization model includes: The current policy model and target policy model of the machine learning model; the current value model and target value model of the machine learning model; The current strategy model takes the input data file after natural language parsing as input and outputs the optimization result of the input data file; after each self-learning of the advertising delivery optimization model, the model parameters of the current strategy model are updated; The target strategy model, whose model parameters are the model parameters of the delayed synchronized current strategy model, is used to generate an optimization result Opt+1 for the output data file according to the output data file Out after the input natural language parsing in the self-learning process of the advertising delivery optimization model; The current value model is used to predict the Q value of the current self-learning process according to the input data file Inp after natural language parsing and the corresponding optimization result Opt during the self-learning process of the advertising delivery optimization model; after each self-learning of the advertising delivery optimization model, the model parameters of the current strategy model are updated; The target value model is used to predict the Q' value of the current self-learning process according to the input output data file Out and the optimization result Opt+1 output in the target strategy model with the output data file Out as input during the self-learning process of the advertising delivery optimization model; its model parameters are the current value model parameters of delay synchronization; The self-learning learning process includes the following steps: Randomly sample N groups of advertising delivery optimization record data {Inp, j, Opt, j, Eval, j, Out, j} from the self-learning training database; j=1,2,3…N; Inp, j is the input data file after natural language parsing in the j-th group of data, Opt, j is the optimization result in the j-th group of data; Eval, j is the evaluation result in the j-th group of data; Out, j is the output data file after natural language parsing in the j-th group of data; Use Inp,j and Opt,j to make predictions in the current value model, and calculate Inp,j as the predicted value Q of the input; Use Out,j to calculate Opt+1,j in the target strategy model; Combine Out,j and Opt+1,j to make predictions in the target value model, and calculate the predicted value Q' of the target with Out,j as input; Combine Q' and Eval,j to calculate the target value with Inp,j as input; Update the parameters of the current value model according to the loss gradient of N learnings; the loss gradient is determined according to the predicted value and target value of the input Inp,j; Use the maximum prediction value of N learnings to update the parameters of the current policy model.

[0033] It should be noted that this embodiment describes in detail the structure and function of the first and second advertising delivery optimization models. Here, the current strategy model of the machine learning model refers to a real-time working model used to generate advertising optimization results, while the target strategy model is an auxiliary model used to generate optimization results in the self-learning process. The current value model and the target value model are used to predict the Q value and Q' value in the self-learning process, respectively, which are key indicators for evaluating the optimization effect.

[0034] Specifically, the first advertising delivery optimization model or the second advertising delivery optimization model includes a current strategy model and a target strategy model of a machine learning model, as well as a current value model and a target value model. The current strategy model takes an input data file after natural language parsing as input and outputs an optimization result.

[0035] More specifically, during the self-learning process, the parameters of the model are updated based on the learned knowledge. The parameters of the target strategy model are delayed synchronized versions of the parameters of the current strategy model, which are used to generate optimization results for the output data file during the self-learning process. The current value model and the target value model predict the Q value and Q' value during the self-learning process, respectively. These values ​​are calculated based on the input data and the optimization results, and are used to evaluate and guide the self-learning process.

[0036] Preferably, the learning process of self-learning may include the following steps: first, randomly sample N groups of advertising delivery optimization record data from the self-learning training database. Then, use these data to make predictions in the current value model to calculate the predicted value Q of the input data file. Next, use the output data file to calculate the optimization result Opt+1 in the target strategy model, and combine these results to make predictions in the target value model to calculate the predicted value Q' of the target.

[0037] Furthermore, the target value is calculated by combining Q' and the evaluation result Eval, and the parameters of the current value model and the current policy model are updated according to these values. This process can be refined or provide alternatives by adjusting the number of samples N, the choice of optimization algorithm (such as gradient descent, stochastic gradient descent, etc.), and parameters such as the learning rate.

[0038] In some embodiments, the loss is calculated using the mean square error, and the parameters of the current value model are updated according to the gradient of the loss. The loss function is:

[0039] in, are the model parameters of the current value model, are the model parameters of the target value model; Use the maximum predicted Q value to update the parameters of the current policy model. The loss function is

[0040] in, are the model parameters of the current policy model.

[0041] It should be noted that this embodiment describes in detail how to use mean square error to calculate the loss and update the parameters of the current value model according to the gradient of the loss. Here, mean square error is a commonly used loss function that measures the difference between the model's predicted value and the actual value. By minimizing this error, the performance of the model can be optimized.

[0042] Specifically, the process of calculating the loss using mean square error involves comparing the prediction of the current value model with the prediction of the target value model. The loss function is defined as the average of the sum of the squares of the differences between the predicted value and the actual value, and the formula is

[0043] in are the model parameters of the current value model, are the model parameters of the target value model, is the number of samples. By calculating the gradient of this loss function, the parameters of the current value model can be updated to make it closer to the parameters of the target value model.

[0044] Preferably, when updating the parameters of the current policy model, the method of maximizing the predicted Q value can be adopted. This means that during the self-learning process, the system will try to find the policy that can produce the highest Q value. The loss function can be defined as ,in It's the state. is the action. By maximizing this predicted Q value, the parameters of the current strategy model can be updated to guide the advertising delivery strategy more effectively.

[0045] Furthermore, different optimization algorithms, such as Adam or RMSprop, can be used to speed up the parameter update process. The learning process can also be refined by adjusting hyperparameters such as learning rate and discount factor, or regularization techniques can be used to prevent model overfitting.

[0046] In some embodiments, the optimization result output by the current strategy model of the first advertisement delivery optimization model is an advertisement content optimization result, including a problematic text position in the input data file and optimized text content; The optimization result output by the current strategy model of the second advertising delivery optimization model is an advertising effect prediction result, including the text position for effect prediction in the input data file and the predicted text content.

[0047] It should be noted that this embodiment describes in detail the specific output contents of the first advertising delivery optimization model and the second advertising delivery optimization model. Here, the advertising content optimization result refers to the improvement suggestions made by the system after analyzing the advertising content, and the advertising effect prediction result refers to the system's prediction of the advertising delivery effect based on the advertising content data.

[0048] Specifically, the optimization results output by the current strategy model of the first advertising delivery optimization model include the problematic text positions in the input data file and the optimized text content. This means that the model not only identifies the problem, but also provides specific modification suggestions. For example, if the model identifies that the keyword selection is wrong, it may recommend replacing it with a more attractive or relevant keyword.

[0049] Furthermore, the optimization results output by the current strategy model of the second advertising delivery optimization model include the text positions for effect prediction and the predicted text contents in the input data file, which involve the prediction of possible user reactions and market performance after the advertising delivery.

[0050] Preferably, the first advertisement delivery optimization model may use natural language generation technology to generate optimized text content, ensuring that the modified advertisement content not only solves the problem but also maintains the original style and tone.

[0051] Furthermore, the second advertising delivery optimization model can combine user behavior data and market trend analysis to improve the accuracy of prediction. For example, the user's historical click data can be used to predict the attractiveness of new advertising content.

[0052] Preferably, a multi-objective optimization strategy can be introduced to simultaneously consider the click-through rate and conversion rate of the advertisement to achieve a more comprehensive prediction of the advertisement effect.

[0053] In some embodiments, in the evaluation module, a text comparison is performed on the input data file and the corresponding output data file to find out the different text positions and text contents in the two files; then the text positions and text contents involved in the optimization results are compared; if the modified position and / or content in the output data file is the same as that in the optimization result, a corresponding quantized positive feedback value is assigned; otherwise, a corresponding quantized negative feedback value is assigned; and all positive feedback values ​​and negative feedback values ​​are added together and output as the evaluation result.

[0054] It should be noted that this embodiment describes in detail the working principle of the evaluation module. Here, the evaluation module refers to a component in the system used to evaluate the optimization effect of advertising delivery. It quantifies the optimization effect of advertising content by comparing and analyzing the input data file and the output data file. Text comparison refers to comparing two text files word by word or sentence by sentence to identify the differences between them.

[0055] Specifically, the evaluation module compares the input data file with the corresponding output data file to find out the different text positions and text contents in the two files. This process can be achieved through automated text comparison tools that can identify operations such as insertion, deletion and modification in the text.

[0056] Furthermore, the evaluation module compares these differences with the text positions and text contents involved in the optimization results to determine whether the advertiser has made corresponding modifications according to the optimization suggestions of the system.

[0057] Preferably, the evaluation process of the evaluation module can be further refined into the following steps: first, use a text comparison algorithm, such as the longest common subsequence algorithm or the difference algorithm, to identify the differences between the input and output files; then, match these differences with the problem locations and suggested modifications indicated in the optimization results; then, assign positive or negative quantitative feedback values ​​based on the matching results, for example, if the advertiser's modification is consistent with the optimization suggestion, a positive feedback value is assigned, otherwise a negative feedback value is assigned; finally, add all positive feedback values ​​and negative feedback values ​​to obtain the final evaluation result.

[0058] Furthermore, a weighted scoring mechanism can be introduced to assign different weights to different types of text modifications to more accurately reflect the potential impact of the modifications on advertising effectiveness.

[0059] In some embodiments, the optimization server further includes a first pre-training database and / or a second pre-training database; The first pre-training database is used to pre-train the current strategy model of the first advertising delivery optimization model to obtain model parameters of the current strategy model when the first advertising delivery optimization model is cold started; The second pre-training database is used to pre-train the current strategy model of the second advertising delivery optimization model to obtain model parameters of the current strategy model when the second advertising delivery optimization model is cold started.

[0060] The sample data in the first pre-training database or the second pre-training database is a historical advertisement content file and the modification result and / or supplementation result are obtained after analyzing the historical advertisement content file.

[0061] It should be noted that this embodiment elaborates on the role and content of the pre-training database in the optimization server. Here, the pre-training database refers to a repository containing pre-prepared data sets, which are used to train the model before it is actually applied so that the model can better understand and process advertising content. Cold start refers to the process of starting a system or model without any previous data or experience.

[0062] Specifically, the optimization server includes a first pre-training database and / or a second pre-training database, which are used to pre-train the current strategy model of the model when the first advertising delivery optimization model and the second advertising delivery optimization model are cold started.

[0063] More specifically, the sample data in the pre-training database are historical advertising content files and the modified and / or supplemented results obtained after analyzing these historical advertising content files. This means that before the model starts processing new advertising content, it will use these historical data to learn common patterns and effective optimization strategies for advertising content.

[0064] Preferably, the pre-training database can use various types of data to enrich its content, including but not limited to historical advertising copy, user feedback data, market analysis reports, etc. These data can be collected from the Internet through automated crawler technology, or obtained from partners.

[0065] Furthermore, during the pre-training process, specific parameters such as training rounds, batch size, learning rate, etc. can be set to optimize the learning effect of the model. In addition, transfer learning techniques can be used to apply knowledge learned in one field to another field, for example, applying strategies learned from consumer product advertising to car advertising. This can accelerate the cold start process of the model and improve its performance in new fields. It is also possible to consider using reinforcement learning techniques to further train and optimize the model by simulating different advertising scenarios.

[0066] In some embodiments, it also includes a data acquisition terminal; The advertiser terminal is connected to a database of the advertising platform; The advertiser terminal stores the input data files and output data files generated by the advertiser during the advertisement delivery process in a database of the advertisement platform; The data collection end is respectively connected to the database of the advertising platform and the optimization server for communication, and is used for extracting input data files and output data files from the advertising platform database and outputting them to the optimization server.

[0067] It should be noted that the present embodiment describes in detail the functions and effects of the data collection terminal. Here, the data collection terminal refers to a special component or service that is responsible for extracting input data files and output data files from the database of the advertising platform and transmitting them to the optimization server.

[0068] Specifically, the data collection end is connected to the database of the advertising platform and the optimization server respectively, and is used to extract input data files and output data files from the advertising platform database and output them to the optimization server. This means that the data collection end acts as a bridge for data transmission, ensuring that the optimization server can obtain the latest advertising data for analysis and optimization. The data collection end can interact with the database of the advertising platform through API calls, database queries, or file transfers.

[0069] Preferably, the data collection end can use a scheduled task to regularly pull the latest data files from the advertising platform database. For example, a data synchronization task can be set to be executed every morning to ensure that the optimization server always processes the latest data.

[0070] Furthermore, the data collection end can implement data cleaning and preprocessing operations, such as removing invalid or duplicate data records, filling missing values, or converting data into a format that can be processed by the optimization server. In order to improve the security of data transmission, encrypted transmission protocols such as SSL / TLS can be used to ensure that data is not intercepted or tampered with during transmission. You can also consider implementing data compression technology to reduce bandwidth consumption during transmission.

[0071] The above-mentioned embodiments of the present invention have the following beneficial effects: the advertising delivery optimization system of the present invention can efficiently process the input data files generated by the advertiser by integrating the natural language parsing module, the evaluation module, the advertising delivery optimization model and the self-learning training database. The system can conduct in-depth analysis of the advertising content, identify and optimize the problems in the advertising content, and provide optimization results for advertisers to refer to, thereby improving the quality of the advertising content. At the same time, the evaluation module of the system can evaluate the advertising delivery optimization results, and update the advertising delivery optimization model through self-learning based on the evaluation results, so that the model parameters are continuously optimized and the advertising delivery effect is continuously improved.

[0072] In addition, the first advertising delivery optimization model and the second advertising delivery optimization model included in the system can provide professional support for advertising content optimization and advertising effect prediction respectively. The first model focuses on identifying problems in advertising content and providing optimization suggestions, while the second model predicts advertising delivery effects based on advertising content data, assisting advertisers in formulating more effective delivery strategies. This dual-model design can more comprehensively cover all aspects of advertising delivery, from content creation to effect evaluation, and provide advertisers with a full range of optimization services. Through this design, the system can not only improve the quality of advertising content, but also predict and optimize advertising effects to maximize the benefits of advertising delivery.

[0073] like Figure 2 As shown, in some embodiments, a method 200 for optimizing advertisement delivery based on natural language processing includes: Step S1: cold start the optimization server, train the advertisement delivery optimization model in the optimization server through pre-training data, and initialize the advertisement delivery optimization model parameters in the optimization server; Step S2: The optimization server receives an input data file including advertisement content data generated by an advertiser through an advertiser terminal during an advertisement delivery process; Step S3: After natural language analysis of the input data file, the optimization result Opt is obtained by processing it through the built-in advertising delivery optimization model, and then fed back to the corresponding advertiser end; Step S4: The advertiser modifies and / or supplements the input data file according to the optimization result Opt through the advertiser terminal to obtain an output data file; Step S5, the optimization server receives the output data file sent by the advertiser, and evaluates the optimization result of this advertisement delivery according to the corresponding parsed input data file Inp, output data file Out and optimization result Opt to obtain an evaluation result Eval; Step S6: storing the parsed input data file Inp, output data file Out, optimization result Opt and evaluation result Eval corresponding to each advertisement delivery optimization as a set of advertisement delivery optimization record data; Step S7: retrieve N groups of advertising delivery optimization record data to perform self-learning training on the advertising delivery optimization model and update the model parameters.

[0074] It is understandable that the steps described in the natural language processing-based advertising delivery optimization method 200 are similar to those described in the reference Figure 1Therefore, the modules, features and beneficial effects described above for the advertising delivery optimization system based on natural language processing are also applicable to the advertising delivery optimization method 200 based on natural language processing and the operations contained therein, and will not be repeated here.

[0075] Reference below Figure 3 , which shows a schematic diagram of a structure 300 of an electronic device suitable for implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0076] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 to a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0077] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0078] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0079] The above descriptions are only some preferred embodiments of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. An advertisement delivery optimization system based on natural language processing, characterized in that: It includes at least one advertiser terminal and an optimization server; the optimization server is connected to each advertiser terminal for data exchange; The optimization server receives an input data file including advertisement content data generated by an advertiser through a certain advertiser terminal during the advertisement delivery process; after natural language analysis of the input data file, the optimization result is obtained by processing it through a built-in advertisement delivery optimization model, and the optimization result is fed back to the corresponding advertiser terminal; The advertiser modifies and / or supplements the input data file according to the optimization result through the advertiser terminal to obtain an output data file; The optimization server receives the output data file sent by the advertiser, evaluates the optimization result of this advertisement delivery according to the corresponding input data file, output data file and optimization result, and updates the advertisement delivery optimization model through self-learning based on the evaluation result.

2. The advertisement delivery optimization system according to claim 1, characterized in that: The optimization server includes a natural language parsing module, an evaluation module, an advertisement delivery optimization model and a self-learning training database; The natural language parsing module is used to parse the input data file or the output data file using a natural language processing method to obtain the parsed input data file Inp or the output data file Out; The advertising delivery optimization model is used to process the parsed input data file Inp to obtain an optimization result Opt; The evaluation module is used to evaluate the optimization result of each advertisement delivery according to the parsed input data file Inp, output data file Out and optimization result Opt corresponding to each advertisement delivery process to obtain an evaluation result Eval; The self-learning training database is used to store the parsed input data file Inp, output data file Out, optimization result Opt and evaluation result Eval corresponding to each advertisement delivery optimization as a set of advertisement delivery optimization record data; The advertising delivery optimization model is a self-learning machine learning model, which performs self-learning training by retrieving multiple groups of advertising delivery optimization record data from a self-learning training database to update model parameters.

3. The advertisement delivery optimization system according to claim 2, characterized in that: The advertising delivery optimization model includes a first advertising delivery optimization model and / or a second advertising delivery optimization model; The first advertising delivery optimization model is an advertising content optimization model, which is used to identify problems including incorrect keyword selection, missing content, and content conflicts in the parsed input data file Inp including advertising content data and output optimization results; After the optimization result is fed back to the advertiser side, the advertiser modifies the input data file on the advertiser side with reference to the optimization result to obtain an output data file; The second advertising delivery optimization model is an advertising effect prediction model, which is used to generate a prediction result of subsequent advertising delivery based on the parsed input data file Inp including the advertising content data; After the prediction result is fed back to the advertiser side, the advertiser supplements the input data file with the advertisement delivery strategy with reference to the prediction result at the advertiser side to obtain the output data file.

4. The advertisement delivery optimization system according to claim 1, characterized in that: The first advertising delivery optimization model or the second advertising delivery optimization model both includes: The current policy model and target policy model of the machine learning model; the current value model and target value model of the machine learning model; in, The current strategy model takes the input data file after natural language parsing as input and outputs the optimization result of the input data file; after each self-learning of the advertising delivery optimization model, the model parameters of the current strategy model are updated; The target strategy model, whose model parameters are the model parameters of the delayed synchronized current strategy model, is used to generate an optimization result Opt+1 for the output data file according to the output data file Out after the input natural language parsing in the self-learning process of the advertising delivery optimization model; The current value model is used to predict the Q value of the current self-learning process according to the input data file Inp after natural language parsing and the corresponding optimization result Opt during the self-learning process of the advertising delivery optimization model; after each self-learning of the advertising delivery optimization model, the model parameters of the current strategy model are updated; The target value model is used to predict the Q' value of the current self-learning process according to the input output data file Out and the optimization result Opt+1 output in the target strategy model with the output data file Out as input during the self-learning process of the advertising delivery optimization model; its model parameters are the current value model parameters of delay synchronization; The self-learning learning process includes the following steps: Randomly sample N groups of advertising delivery optimization record data {Inp, j, Opt, j, Eval, j, Out, j} from the self-learning training database; j=1,2,3…N; Inp, j is the input data file after natural language parsing in the j-th group of data, Opt, j is the optimization result in the j-th group of data; Eval, j is the evaluation result in the j-th group of data; Out, j is the output data file after natural language parsing in the j-th group of data; Use Inp,j and Opt,j to make predictions in the current value model, and calculate Inp,j as the predicted value Q of the input; Use Out,j to calculate Opt+1,j in the target strategy model; Combine Out,j and Opt+1,j to make predictions in the target value model, and calculate the predicted value Q' of the target with Out,j as input; Combine Q' and Eval,j to calculate the target value with Inp,j as input; Update the parameters of the current value model according to the loss gradient of N learnings; the loss gradient is determined according to the predicted value and target value of the input Inp,j; Use the maximized prediction value of N learnings to update the parameters of the current policy model.

5. The advertisement delivery optimization system according to claim 4, characterized in that: The loss is calculated using the mean square error, and the parameters of the current value model are updated according to the gradient of the loss. The loss function is: in, are the model parameters of the current value model, are the model parameters of the target value model; Use the maximum predicted Q value to update the parameters of the current policy model. The loss function is in, are the model parameters of the current policy model.

6. The advertisement delivery optimization system according to claim 4, characterized in that: The optimization result output by the current strategy model of the first advertisement delivery optimization model is an advertisement content optimization result, including a text position with problems in the input data file and optimized text content; The optimization result output by the current strategy model of the second advertising delivery optimization model is an advertising effect prediction result, including the text position for effect prediction in the input data file and the predicted text content.

7. The advertisement delivery optimization system according to claim 6, characterized in that: In the evaluation module, a text comparison is performed on the input data file and the corresponding output data file to find out the different text positions and text contents in the two files; then the text positions and text contents involved in the optimization result are compared; if the modified position and / or content in the output data file is the same as that in the optimization result, a corresponding quantized positive feedback value is assigned; otherwise, a corresponding quantized negative feedback value is assigned; and all positive feedback values ​​and negative feedback values ​​are added together and output as the evaluation result.

8. The advertisement delivery optimization system according to claim 4, characterized in that: The optimization server also includes a first pre-training database and / or a second pre-training database; The first pre-training database is used to pre-train the current strategy model of the first advertising delivery optimization model to obtain model parameters of the current strategy model when the first advertising delivery optimization model is cold started; The second pre-training database is used to pre-train the current strategy model of the second advertising delivery optimization model to obtain model parameters of the current strategy model when the second advertising delivery optimization model is cold started. The sample data in the first pre-training database or the second pre-training database is a historical advertisement content file and the modification result and / or supplementation result are obtained after analyzing the historical advertisement content file.

9. The advertisement delivery optimization system according to any one of claims 1 to 8, characterized in that: It also includes a data collection terminal; The advertiser terminal is connected to a database of the advertising platform; The advertiser terminal stores the input data files and output data files generated by the advertiser during the advertisement delivery process in a database of the advertisement platform; The data collection end is respectively connected to the database of the advertising platform and the optimization server for communication, and is used for extracting input data files and output data files from the advertising platform database and outputting them to the optimization server.

10. An advertisement delivery optimization method based on natural language processing, characterized in that: The following steps are involved: Step S1: cold start the optimization server, train the advertisement delivery optimization model in the optimization server through pre-training data, and initialize the advertisement delivery optimization model parameters in the optimization server; Step S2: The optimization server receives an input data file including advertisement content data generated by an advertiser through an advertiser terminal during an advertisement delivery process; Step S3: After natural language analysis of the input data file, the optimization result Opt is obtained by processing it through the built-in advertising delivery optimization model, and then fed back to the corresponding advertiser end; Step S4: The advertiser modifies and / or supplements the input data file according to the optimization result Opt through the advertiser terminal to obtain an output data file; Step S5, the optimization server receives the output data file sent by the advertiser, and evaluates the optimization result of this advertisement delivery according to the corresponding parsed input data file Inp, output data file Out and optimization result Opt to obtain an evaluation result Eval; Step S6: storing the parsed input data file Inp, output data file Out, optimization result Opt and evaluation result Eval corresponding to each advertisement delivery optimization as a set of advertisement delivery optimization record data; Step S7: retrieve N groups of advertising delivery optimization record data to perform self-learning training on the advertising delivery optimization model and update the model parameters.

Images