ASA advertisement keyword and material generation system based on AI

Abstract

According to the ASA advertisement keyword and material generation system based on the AI, keyword generation is achieved through the method that a knowledge graph is combined with an XLNet, and a multi-mode Transform encoder is designed to process advertisement materials. The system realizes end-to-end automatic processing from advertisement demand understanding to putting optimization by adopting effect estimation of multi-task learning and optimized putting based on a layered multi-arm machine.

Description

Technical Field

[0001] The present invention relates to the field of advertising technology, and in particular to an AI-based ASA advertising keyword and material generation system. Background Art

[0002] ASA (Apple Search Ads) advertising is one of the important channels for promoting mobile applications. Compared with other forms of mobile advertising, ASA ads are directly placed on the search results page of the Apple App Store. They are highly scene-relevant and targeted, and are a powerful tool for reaching high-intent users and improving promotion results. However, ASA ads have a wide keyword coverage and are difficult to monetize. How advertisers can quickly and accurately generate keywords and advertising creatives that are highly matched with products based on a massive library of material and product information, thereby improving advertising results, is a key technical challenge facing ASA advertising optimization.

[0003] Traditional ASA advertising optimization mainly relies on manual configuration of keywords and design of materials. Advertising optimizers need to have a deep understanding of products and industries, and summarize targeted keyword combinations and creative strategies to generate high-quality advertising content. This process is not only time-consuming and labor-intensive, but also easily affected by experience limitations and cognitive blind spots. The generated ads often lack generalization capabilities and are difficult to flexibly respond to changing market environments and user needs. In addition, keyword screening and creative design lack quantitative indicators and real-time feedback. Advertisers cannot accurately estimate the effects of newly added keywords and materials, resulting in low utilization of advertising budgets and difficulty in guaranteeing ROI.

[0004] In recent years, the development of artificial intelligence technology has brought new breakthroughs to ASA advertising optimization. Large-scale language models (LLMs) such as BERT and GPT in the field of NLP have made significant progress in understanding semantics and generating text. Multimodal machine learning models have also gradually matured and can process material information in different forms such as text, images, and videos, making it possible for computers to "understand" advertising content. At the same time, the application of artificial intelligence technologies such as knowledge graphs and reinforcement learning in actual scenarios has become more and more extensive. These cutting-edge technological advances provide new ideas and methods for ASA advertising generation and optimization.

[0005] The present invention conducts research on three core issues in the generation of ASA advertising keywords and materials: 1) How to automatically generate ASA delivery keywords that are highly matched with products based on product information and historical advertising delivery data to improve advertising effects; 2) How to accurately predict the delivery effects of new keywords (such as click-through rate, conversion rate, etc.) in combination with historical performance data of keywords; 3) How to automatically generate advertising materials (including copywriting, pictures, videos) that are consistent with the brand image, and optimize the materials in real time according to the data during the delivery process. The present invention innovatively proposes an AI-based ASA advertising keyword and material generation system. The system uses powerful natural language understanding capabilities and XLNet's advantages in text generation and multimodal information processing to efficiently analyze advertisers' advertising needs, accurately grasp product selling points and audience characteristics, and intelligently construct advertising keywords and materials that meet the brand tone. On this basis, the present invention designs a set of dynamic optimization algorithms to count advertising effect data in real time, and continuously optimize keyword combination strategies and creative presentation methods. Experimental results show that this method can significantly improve key performance indicators such as the click-through rate and conversion rate of ASA advertisements. While improving the delivery effect, it greatly improves the automation and efficiency of advertisement optimization, providing a new breakthrough for ASA advertising practice. Summary of the invention

[0006] In order to solve the above problems, the present invention provides an ASA advertising keyword and material generation system based on artificial intelligence, the structure diagram of which is as follows: Figure 1 As shown, the system includes the following modules:

[0007] ASA advertising demand understanding module, which completes the understanding of advertising demand according to the following steps:

[0008] Step 1.1 Use the deep learning model of BiLSTM-CRF architecture to analyze advertiser needs;

[0009] Step 1.2 introduces the multi-head attention mechanism MHA to enhance the understanding of advertising intent:

[0010] Q, K, V = H Q，K，V W Q，K，V ;

[0011]

[0012] MHA(Q,K,V)=concat(head1,...,head k )W O ;

[0013] Among them, H Q，K，V Represents the input feature matrix of advertising demand, W Q，K，Vrepresents the learnable weight matrix, d represents the dimension of the attention head, and head k represents the output of the kth attention head, concat represents the vector concatenation operation, and W O represents the output mapping matrix, Q, K, and V represent query, key, and value matrices, respectively.

[0014] ASA advertising keyword generation module, which generates advertising keywords according to the following steps:

[0015] Step 2.1 Precise advertising keyword mining based on knowledge graph:

[0016]

[0017] in, represents the knowledge subgraph of the advertising product p, (h, r, t) represents the head entity, relationship and tail entity triples in the subgraph, v h , v r , v t Represent their vector representations respectively, ||·||2 represents the L2 norm, and argmin represents the parameter value that minimizes the objective function.

[0018] Step 2.2: Ad keyword expansion based on XLNet:

[0019] P(w i |S \i )=XLNet(S|w i =[MASK]);

[0020]

[0021] Among them, S represents the advertising seed keyword sequence, w i represents the i-th word, h i Represents the last layer w of XLNet i The hidden state of the position, Q(w i ) represents the quality assessment score of the advertising keyword, b i and b j represents the bias term, w i and w j Represents the word vector, S \i represents the sequence except the i-th word, and [MASK] represents the mask tag.

[0022] ASA ad creative generation module, which generates ad creatives according to the following steps:

[0023] Step 3.1 Design a multimodal Transformer encoder to process advertising materials:

[0024] h (m)=MMTE(x (m) , p (m) ), m∈{t,i,v};

[0025] Among them, x (m) Represents the content feature vector of the ad creative in mode m, p (m) represents the position feature vector, m represents the modality type, t, i, v represent text, image, and video modalities respectively, and h (m) represents the modal features of the output, and MMTE stands for Multimodal Transformer Encoder.

[0026] Step 3.2 introduces cross-modal matching loss to optimize ad creatives:

[0027]

[0028] in, Represents a set of advertising samples in a batch. represents the feature representation of the advertisement sample x in mode m, x + represents the positive sample that matches the sample x, x′ represents other samples in the batch, and · represents the dot product operation.

[0029] Advertisement effect estimation module, which estimates the advertisement effect according to the following steps:

[0030] Step 4.1 Design the feature interaction layer to model advertising creatives and keywords:

[0031] z ad =MLP(e ad )⊙v ad ;

[0032] z kw =MLP(e kw )⊙v kw ;

[0033] z=[z ad , z kw , z ad ⊙z kw ];

[0034] Among them, e ad and e kw are the Embedding vectors of advertising creativity and keywords respectively, v ad and v kw are the additional feature vectors of advertising creativity and keywords respectively, ⊙ is the Hadamard product (element-wise product), [·,·] represents vector concatenation, and MLP stands for multi-layer perceptron.

[0035] Advertisement delivery optimization module, which performs advertisement delivery optimization according to the following steps:

[0036] Step 5.1 Use the Hierarchical Multi-Armed Bandit Strategy (HMAB) for ad delivery optimization:

[0037]

[0038] Among them, a represents the advertising strategy, EU PK (a) represents the expected utility calculated based on prior knowledge, n a represents the number of times strategy a is selected, n i represents the number of selections of the i-th strategy, α represents the exploration factor (used to balance exploration and exploitation), Represents the total number of selections for all strategies.

[0039] Step 5.2: Use evolutionary algorithms to optimize advertising bids:

[0040] Fitness(b)=ω1×CTR b +ω2×Rank b -ω3×CPC b +ω4×Quality b ;

[0041] Among them, b represents the advertising bid vector, CTR b Represents the estimated click rate, Rank b Represents the ad ranking score, CPC b stands for cost per click, Quality b represents the advertising quality score, ω1, ω2, ω3, and ω4 represent the weight coefficients of click-through rate, ranking, cost, and quality, respectively.

[0042] The present invention also provides a storage medium for generating computer-implemented instructions for ASA advertising keywords and materials based on AI, which can be read by a computer device, and the computer program instructions stored therein can realize the full-process intelligent processing of ASA advertising when executed by a processor. The storage medium contains a deep understanding module for advertiser needs, which performs demand analysis through a deep learning model, and uses a multi-head attention mechanism MHA to enhance advertising intent understanding, construct an input feature matrix of advertising needs, and perform feature processing through a learnable weight matrix to achieve accurate grasp and structured expression of advertising needs.

[0043] The core functional modules of the storage medium include an advertising keyword generation system based on knowledge graph and XLNet, a material processing engine based on multimodal Transformer, an effect estimation framework based on multi-task learning, and a layered multi-armed bandit delivery optimization strategy. Among them, the keyword generation system conducts precise keyword mining by constructing a product knowledge subgraph, and uses the XLNet model to achieve dynamic keyword expansion; the material processing engine can uniformly process multimodal content such as text, images, and videos, and optimize the generation quality of advertising creativity through cross-modal matching loss; the effect estimation framework uses a feature interaction layer to model the correlation between advertising creativity and keywords, and estimates multiple effect indicators through multi-task learning; the delivery optimization strategy uses a combination of HMAB algorithm and evolutionary algorithm to achieve multi-level intelligent optimization and bid adjustment from Campaign to Ad Group.

[0044] To ensure efficient operation of the system, the storage medium also contains a complete data management and system monitoring module. The data management module is responsible for establishing the index structure of the advertising material library, storing and maintaining the entity relationship data, model parameters and training data of the knowledge graph, and supporting incremental data updates and garbage collection mechanisms. The system monitoring module records the system operation status and performance indicators in real time, monitors the model prediction effect, adaptively adjusts the system parameter configuration, and realizes anomaly detection and fault recovery. The storage medium can be a computer-readable storage medium in various forms such as ROM, RAM, disk, CD, etc., which has a special area for storing intermediate calculation results, evaluation indicators and other data, as well as a cache area for temporary data storage during program execution to ensure the stability and efficiency of system operation.

[0045] This invention significantly improves the accuracy and coverage of keyword generation through the innovative combination of knowledge graph and XLNet. At the same time, it uses a multimodal Transformer architecture to achieve unified processing and generation of text, image and video materials, and ensures semantic consistency between materials of different forms through cross-modal matching loss. The effect prediction framework based on multi-task learning solves the correlation modeling problem between prediction tasks through feature interaction layers and uncertainty weighting mechanisms, and still maintains a high accuracy rate in data sparse scenarios.

[0046] At the system level, the Hierarchical Multi-Armed Bandit (HMAB) optimization strategy realizes multi-level intelligent optimization from Campaign to Ad Group, and combines the bid optimization of evolutionary algorithms to significantly improve the automation of advertising and the efficiency of budget utilization. The entire system greatly reduces the cost of manual intervention through end-to-end automated processing, while maintaining strong market adaptability and being able to quickly adjust strategies based on delivery data. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1This is the system structure diagram of this application. DETAILED DESCRIPTION

[0048] XLNet is a Transformer-like language model proposed by Google, which has achieved better performance than BERT in many NLP tasks. Unlike traditional language models, XLNet adopts the Permutation Language Modeling training paradigm, which can model the bidirectional dependency of text sequences and has stronger semantic understanding and context perception capabilities. Applying XLNet to the automatic generation of advertising copy can fully mine the semantic information in the material library, learn the brand's language style and industry terms, and then generate fluent, natural and contagious advertising terms.

[0049] Advertisement optimization requires comprehensive consideration of material information of different modalities such as text, images, and videos. XLNet adopts the Two-Stream Self-Attention mechanism, which can flexibly integrate feature representations of different modalities. This mechanism maps inputs of different modalities to the same semantic space by introducing identification vectors (ModalityIndicator) and position vectors (PositionEmbedding), thereby modeling the interaction and connection between modalities. On this basis, the present invention designs a multi-modal Transformer Encoder (MMTE) to achieve end-to-end representation learning of text, image, and video features:

[0050] h (m) =MMTE(x (m) , p (m) ), m∈{t,i,υ} (1);

[0051] Where x represents the input content feature vector, p represents the input position feature vector, and t, i, and v represent text, image, and video modalities, respectively. MMTE adaptively fuses information from different modalities through the Self-Attention mechanism, and finally obtains a unified multimodal feature representation h. Based on this representation, the present invention further designs a cross-modal matching (CM) loss function:

[0052]

[0053] in is a set of advertising samples in a batch. Formula (2) explicitly models the cross-modal consistency of advertising materials by maximizing the matching degree of different modal features in the same advertisement.

[0054] The intelligent ASA advertising optimization system proposed in the present invention mainly includes the following four modules: demand understanding module, keyword generation module, material generation module and effect prediction module. Among them, the demand understanding module and the keyword generation module realize the structured representation of advertising demand and the automatic generation of keywords; the material generation module comprehensively uses the text generation and multimodal feature fusion capabilities of XLNet to automatically generate high-quality graphics, texts and video materials; the effect prediction module predicts the key effect indicators of advertising creativity and keyword combinations through machine learning algorithms based on historical data. These four modules work together to ultimately output an optimized advertising placement configuration.

[0055] The demand understanding module aims to convert the advertising demand provided by advertisers in natural language into structured information representation. Traditional demand understanding mainly relies on manually designed keyword extraction rules and template matching, and has limited generalization ability. This paper combines advanced natural language processing technology to design an end-to-end advertising demand understanding model based on deep learning.

[0056] The backbone network of the model adopts the BiLSTM-CRF architecture. The BiLSTM layer models the contextual information of the input text sequence, and the CRF layer further considers the constraint relationship between slot labels to decode the optimal slot filling path. In addition, the present invention introduces a multi-head attention (MHA) layer before the BiLSTM layer:

[0057]

[0058] The MHA layer learns different semantic perspectives of the input sequence in parallel through multiple attention heads, improving the accuracy of intent recognition and slot filling.

[0059] The present invention pre-trains the model on large-scale advertising corpus and designs a set of ASA advertising-specific intent slot annotation systems based on business scenarios. For a given advertising demand description, the model can accurately parse out key information such as product type, audience characteristics, promotion goals, and output a standardized JSON format representation. This structured demand representation can effectively support subsequent keyword expansion, material generation, and other links.

[0060] High-quality keywords are the basis of ASA advertising. The present invention proposes a keyword generation method that integrates knowledge graph and deep learning. The method is divided into two main steps: targeted keyword mining and generalized keyword expansion.

[0061] First, the present invention uses a commercial large model to build a product knowledge graph based on the product information provided by advertisers. The knowledge graph takes the product as the central node and links related concepts through various semantic relationships (such as attributes, functions, usage scenarios, etc.). The present invention uses a graph embedding method based on TransE to learn the low-dimensional vector representation of product nodes:

[0062]

[0063] in is the knowledge subgraph of product p, (h, r, t) is the triple in the subgraph, representing the head entity, relation, and tail entity respectively. TransE assumes v h +v r ≈v t ,The embedding representation of entities and relations is learned by minimizing the translation error.,On this basis, the present paper designs a heuristic scoring mechanism that ,comprehensively considers factors such as the semantic relevance between keywords and products, industry popularity, and historical click-through rates, to mine a set of highly targeted ,seed keywords from the knowledge graph.

[0064] Seed keywords reflect the accurate grasp of product selling points, but their number is limited and it is difficult to meet the coverage requirements of advertising. Therefore, in the second step, the present invention generates more related words by expanding generalized keywords based on seed keywords. Inspired by the MaskLanguageModel and MLM paradigms, this method uses XLNet as a generator, takes seed keywords as input, and generates generalized keywords by iterative mask prediction. Specifically, let S = {w1, w2, ..., w K} is a seed keyword sequence, in which the present invention randomly masks a proportion of words and inputs them into XLNet to obtain the probability distribution of masked words:

[0065] P(w i |S \i )=XLNet(S|w i =[MASK]);

[0066] Based on this distribution, a new word w′ is sampled. i Replace the original mask word to get a new sequence S', and repeat the above process until a sufficient number of generalized keywords are generated. To ensure the quality of the generated words, the present invention sets constraints such as top-k sampling, and introduces a quality evaluator in the last layer of XLNet to dynamically adjust the weight of the generated words:

[0067]

[0068] h i is the last layer w of XLNet i The hidden state of the position, Q(w i) is the output of the quality evaluator w i Quality score. The resulting generalized keywords are highly semantically related to the seed keywords and are more diverse in form, which helps to attract a wide range of interests among potential users.

[0069] By combining targeted keywords and generalized keywords, this method improves keyword coverage while ensuring relevance. The application of a large number of real advertising cases shows that the keyword generation method based on knowledge graph + XLNet can significantly improve the display volume and click-through rate of ASA ads.

[0070] High-quality advertising creativity is the key to improving advertising effectiveness. Traditional material design mainly relies on experienced advertising creative personnel, which is costly and inefficient. The present invention proposes a multi-modal, multi-task end-to-end material generation method, which can automatically generate highly personalized graphics and video materials based on an advertising corpus.

[0071] This method uses XLNet as the basis of the generative model and incorporates a specific design for the ASA material generation task: 1) Introducing advertising industry corpus in the pre-training stage to enhance the model's understanding and generation capabilities of advertising copy; 2) Using the multimodal Transformer encoder (Formula (1)) as the underlying feature extractor to achieve unified modeling of image, text, and video features; 3) Designing multiple decoders to complete different material generation subtasks (such as copy, keyframes, video editing, etc.), and using cross-modal matching loss (Formula (2)) to achieve collaborative training between decoders; 4) Adding an external knowledge injection mechanism to the decoder to use prior information such as knowledge graphs to guide material generation.

[0072] During model training, the present invention constructs a large-scale advertising material corpus (including text, pictures, and videos) into a series of <keywords, advertising creativity> pairs. The keyword information is passed to each decoder through the attention mechanism to guide the generation of the corresponding material. For example, for the text generation task, the model's goal is to maximize the following conditional probability:

[0073]

[0074] Where y is the generated copy sequence, x is the input keyword sequence, is the knowledge graph corresponding to the keyword. Similarly, the key frame generation task is based on x and The key frame sequence is generated through autoregressive decoding as input. The video editing task further optimizes the shot combination through reinforcement learning based on the key frames, so that the video reaches the best in terms of information volume, aesthetics, rhythm, etc.

[0075] The present invention trains the multimodal material generation model on a massive amount of advertising material corpus, and designs a set of evaluation systems based on manual scoring to evaluate the generation quality. The results of offline experiments and online A / B tests show that the model can automatically design personalized advertising ideas that are highly attractive to users based on the input keywords, and the average click-through rate and conversion rate are significantly improved. At the same time, the model also has good generalization ability, and can also generate high-quality materials for new categories and new scenarios, saving advertisers a lot of creative design costs.

[0076] Advertising effect prediction is crucial for optimizing advertising delivery strategies. This paper proposes a deep effect prediction model based on Multi-task Learning (MTL), which can simultaneously predict key indicators such as click-through rate (CTR), conversion rate (CVR), and retention rate (RR) of advertising creativity and keyword combinations.

[0077] The model takes advertising creative (ad) and keywords (kw) as input, and first maps them into dense vectors through their respective Embedding layers. Due to the variety of creative and keyword types, the present invention further introduces the FeatureInteraction layer to model the high-order interaction features between the two:

[0078] zad=MLP(ead)⊙vad zkw=MLP(ekw)⊙vkw z=[zad, zkw, zad⊙zkw];

[0079] e ad and e kw are the embedding vectors of creativity and keywords respectively, and ⊙ is the Hadamard product. By introducing MLP transformation and second-order interaction term z ad ⊙z kw , this layer can flexibly learn the nonlinear relationship between features.

[0080] The output z of the FeatureInteraction layer is passed into multiple Task-specific MLP classifiers to predict different target indicators:

[0081]

[0082] These classifiers share the underlying feature z, so they can explicitly model the relationship between different indicators, while maintaining a certain degree of task differentiation through independent classification layers. The model training process adopts a joint learning paradigm, combining the losses of multiple tasks:

[0083] L=λ1L CTR +λ2L CVR +λ3L RR ;

[0084] λ i is the weight coefficient of each task. In addition, the present invention also introduces a Gate mechanism after each classifier to adaptively adjust the task weight:

[0085]

[0086] in is a learnable parameter vector. The Gate mechanism enables the model to dynamically assign task weights based on input ideas and keywords, improving the flexibility of estimation.

[0087] This MTL-based effect prediction model makes full use of the correlation between different indicators and can accurately predict advertising effects in sparse scenarios. Offline experiments show that its AUC value is 3%-5% higher than that of traditional independent prediction models.

[0088] The goal of advertising delivery optimization is to find keyword combinations and creative strategies to maximize the overall advertising revenue while meeting budget constraints. The present invention proposes a hierarchical delivery optimization framework to optimize the combination strategy of keywords and creatives, bidding and budget allocation from top to bottom.

[0089] The top layer uses a multi-armed bandit (MAB) model, treating each keyword or creative as a bandit arm. As the delivery process progresses, MAB explores the expected returns of different arms through ∈-greedy, and uses the UCB algorithm to balance exploration and utilization, gradually converging to the global optimal delivery combination.

[0090] Based on Multi-armed Bandit, MAB, the present invention considers the specific constraints and goals of the ASA scenario and designs a series of improvements: 1) Introducing a hierarchical bandit structure (Hierarchical MAB, HMAB), first optimizing budget allocation at the Campaign level, and then optimizing keywords and creative combinations at the Ad Group level to improve optimization efficiency; 2) Designing a reward function based on expected utility (EU), while considering multiple factors such as click-through rate, conversion rate, and cost, and reasonably weighing advertising revenue and cost; 3) Adding prior knowledge (PK) guidance to the bandit algorithm, making full use of information such as click-through rate provided by the effect estimation model, and accelerating strategy convergence. The modified HMAB reward can be expressed as:

[0091]

[0092] Where a is a combination strategy, EU PK (a) Expected utility calculated based on prior knowledge, n ais the number of times strategy a has been selected, n i is the number of selections for the i-th arm, and α is the exploration factor.

[0093] In Multi-armed Bandit, after MAB obtains the combination strategy, the lower layer uses evolutionary algorithm (EA) to further optimize the keyword bid. Different from the traditional manual price adjustment rules, EA starts from multiple randomly generated bidding schemes and iteratively searches for the optimal bid through operations such as selection, crossover, and mutation. Among them, the Fitness function comprehensively considers factors such as ad click-through rate, ranking, and click price:

[0094]

[0095] b is the bid vector, CTR b ,Rank b The CTR and ranking estimated by the effect estimation model, CPC b is the click price, Quality b is the creative quality score, ω i is the weight coefficient. After each round of iteration, the bid solutions with higher scores are retained and cross-mutated, and low-quality solutions are eliminated. This process continues for multiple rounds until the bid converges.

[0096] To verify the effectiveness of the proposed method, the present invention conducted a large-scale experimental comparison and analysis in a real ASA advertising scenario. The experimental data set covers the advertising data of multiple advertisers in multiple industries, with a time span of 3 months. The present invention focuses on evaluating the effect improvement of the three modules of keyword optimization, material generation, and effect estimation.

[0097] The present invention uses an A / B test framework to conduct experiments, and randomly divides the traffic of each advertiser into a control group (Control Group, CG) and an experimental group (Test Group, TG), with the traffic ratio of the two being 1:1. The control group adopts the advertiser's original delivery strategy, and the experimental group adopts the optimization method proposed by the present invention. Observe the improvement effect of the experimental group compared with the control group.

[0098] The evaluation indicators used in the experiment mainly include:

[0099]

[0100] The baseline models for experimental comparison include:

[0101] 1) Classic TF-IDF keyword extraction and manual configuration (TF-IDF+Manual)

[0102] 2) Material generation based on rule templates (Template Creation)

[0103] 3) DeepFFM CTR prediction model for independent tasks (DeepFFM)

[0104] The present invention compares the effects of the knowledge graph + XLNet (KG + XLNet) keyword generation method proposed in the present invention with the TF-IDF + Manual baseline. The results are shown in Table 1:

[0105] Table 1:

[0106]

[0107] It can be seen that the KG+XLNet method has achieved significant effect improvement in various industries, with the average CTR and CTR increased significantly. Among them, the e-commerce and game industries performed particularly well. The present invention believes that the main reasons are: on the one hand, the products in these two industries are updated and iterated quickly, the homogeneous competition is fierce, and advertisers have a strong demand for accurate reach; on the other hand, the knowledge graph can better mine the segmented attributes of the product, combined with the powerful generation ability of XLNet, to obtain both accurate and diverse keyword combinations. The improvement in the education industry is relatively small because the user search habits in this industry are relatively stable, and at the same time, advertisers have a strong subjective preference for keywords, which limits the space for algorithm optimization to a certain extent. In general, the KG+XLNet method can fully integrate industry knowledge and language priors, automatically generate customized keywords, and bring substantial effect improvement to advertisers.

[0108] This paper compares the effects of the proposed multimodal XLNet (MM-XLNet) material generation model with the TemplateCreation baseline. Since Template Creation relies on manual configuration and has a large workload, this paper selects three representative industries for comparison. The results are shown in Table 2:

[0109] Table 2:

[0110]

[0111] Experimental results show that the MM-XLNet model is superior to traditional template-based methods in generating both image and video materials. Further analysis shows that the materials generated by MM-XLNet are superior in visual appeal and information completeness. Thanks to the model's deep understanding of images and texts, the generated materials are more in line with user needs. In addition, through the fusion modeling of multimodal information, MM-XLNet can better grasp the inherent connection between different material forms, making the generated images and videos highly consistent with the advertiser's brand tone and product selling points.

[0112] The present invention also builds a multi-task learning (MTL) effect prediction model proposed in the present invention, and compares it with the traditional independent task DeepFFM model, and the results are shown in Table 3. Two learning paradigms, single task loss (STL Loss) and multi-task joint loss (MTL Loss), were used in the experiment. It can be seen that whether it is the CTR or CVR prediction task, the model using the MTL learning paradigm has achieved better results, indicating that joint learning helps to share and transfer knowledge between different tasks. In addition, the dynamic MTL Loss with the addition of the Gate mechanism is also further improved compared to the simple static weighted MTL Loss, which confirms the value of the Gate mechanism in modeling the importance differences of different tasks and improving the generalization of the model.

[0113] Table 3:

[0114]

[0115] To further verify the superiority of the MTL model, the present invention has statistically analyzed the overall effect of advertising after adopting different effect prediction models, as shown in Table 4. It can be seen that whether it is the HMAB optimization or the manually configured delivery strategy, the effect prediction based on the MTL model (Ours) makes the overall ROI and CPA better than the DeepFFM model, which once again confirms the effect improvement brought by multi-task learning by modeling the connection between different indicators and alleviating data sparsity. It is worth mentioning that the HMAB optimization framework also has obvious advantages in various indicators compared with manual rules, thanks to the ability of HMAB to continuously learn and dynamically adjust strategies, and make full use of the effect prediction results to guide the decision-making process of advertising delivery.

[0116] Table 4:

[0117]

[0118] In summary, the results of large-scale online experiments have confirmed the effectiveness of a series of optimization methods proposed in this invention. The knowledge graph combined with XLNet enhances the richness and accuracy of keyword expansion; the multimodal XLNet material generation model provides personalized and diversified high-quality advertising creativity; the MTL effect prediction model significantly improves the prediction effect of key indicators such as click-through rate and conversion rate; the hierarchical MAB optimization framework realizes the adaptive adjustment of strategies and maximizes the revenue of advertising. The comprehensive application of these methods has made all aspects of advertising delivery intelligent and automated, and the advertiser's input-output ratio (ROI) has been greatly improved overall.

[0119] This paper proposes an end-to-end solution based on knowledge graph and multimodal learning for the three core issues of keyword expansion, material generation, and effect estimation in ASA advertising optimization. This solution innovatively combines the natural language understanding ability, the structured expression of knowledge graph, and the advantages of XLNet in text generation and multimodal feature modeling, and has achieved significant effect improvement in massive advertising scenarios.

[0120] Specifically, the main contributions of the present invention are as follows:

[0121] 1) It is proposed to introduce knowledge graph into ASA advertising optimization, explicitly model prior knowledge such as product attributes and user characteristics based on knowledge graph, and learn the vectorized representation of products through graph neural network to realize automatic mining of targeted keywords.

[0122] 2) Improve the XLNet model and achieve end-to-end generation of multimodal materials such as advertising copy, pictures, and videos through iterative mask prediction, multimodal feature fusion, and cross-modal consistency constraints. The generated personalized creativity effectively matches the product selling points and user preferences, improving the attractiveness and conversion effect of the advertisement.

[0123] 3) A multi-task learning model is proposed to predict advertising results. The feature interaction layer is used to model the high-level interactions between advertising creatives and keywords. The gate mechanism is introduced to dynamically adjust the weights of different task objectives. This greatly improves the prediction results of key indicators such as click-through rate and conversion rate in sparse data environments.

[0124] 4) Designed a hierarchical delivery optimization strategy for HMAB, and realized intelligent advertising budget allocation and delivery combination optimization at different granularities of Campaign and Ad Group. This framework is highly adaptive and can adjust strategies in real time according to environmental changes, making full use of effect estimation information to guide delivery decisions and maximize advertisers' ROI.

Claims

1. AI-based ASA advertising keyword and material generation system, characterized by: The system includes the following modules: ASA advertising demand understanding module; ASA advertising keyword generation module; ASA creative generation module; Advertising effect estimation module; Advertisement delivery optimization module.

2. The AI-based ASA advertising keyword and material generation system according to claim 1 is characterized in that: The ASA advertising demand understanding module understands advertising demand according to the following steps: Step 1.1 Use the deep learning model of BiLSTM-CRF architecture to analyze advertiser needs; Step 1.2 introduces the multi-head attention mechanism MHA to enhance the understanding of advertising intent: Q,K,V=H Q，K，V W Q，K，V ; MHA(Q,K,V)=concat(head1,...,head k )W O ; Among them, H Q，K，V Represents the input feature matrix of advertising demand, W Q，K，V represents the learnable weight matrix, d represents the dimension of the attention head, and head k represents the output of the kth attention head, concat represents the vector concatenation operation, and W O represents the output mapping matrix, Q, K, and V represent query, key, and value matrices, respectively.

3. The AI-based ASA advertising keyword and material generation system according to claim 1, characterized in that: The ASA advertisement keyword generation module generates advertisement keywords according to the following steps: Step 2.1 Precise advertising keyword mining based on knowledge graph: in, represents the knowledge subgraph of the advertising product p, (h, r, t) represents the head entity, relationship and tail entity triples in the subgraph, v h , v r , v t They represent their vector representations respectively, ||·||2 represents the L2 norm, and arg min represents the parameter value that minimizes the objective function; Step 2.2: Ad keyword expansion based on XLNet: P(w i |S \i )=XLNet(S|w i =[MASK]); Among them, S represents the advertising seed keyword sequence, w i represents the i-th word, h i Represents the last layer w of XLNet i The hidden state of the position, Q(w i ) represents the quality assessment score of the advertising keyword, b i and b j represents the bias term, w i and w j Represents the word vector, S \i represents the sequence except the i-th word, and [MASK] represents the mask tag.

4. The AI-based ASA advertising keyword and material generation system according to claim 1, characterized in that: The ASA advertising material generation module generates advertising materials according to the following steps: Step 3.1 Design a multimodal Transformer encoder to process advertising materials: Step 3.1 Design a multimodal Transformer encoder to process advertising materials: h (m) =MMTE(x (m ),p (m) ),m∈{t,i,v}; Among them, x (m) Represents the content feature vector of the ad creative in mode m, p (m) represents the position eigenvector, m represents the mode type, t ， i, v represent text, image, and video modalities respectively, h(m) represents the output modality features, and MMTE represents the multimodal Transformer encoder; Step 3.2 introduces cross-modal matching loss to optimize ad creatives: in, Represents a set of advertising samples in a batch. represents the feature representation of the advertisement sample x in mode m, x + represents the positive sample that matches the sample x, x′ represents other samples in the batch, and · represents the dot product operation.

5. The AI-based ASA advertising keyword and material generation system according to claim 1, characterized in that: The advertising effect estimation module estimates the advertising effect according to the following steps: Step 4.1 Design the feature interaction layer to model advertising creatives and keywords: With ad =MLP(e ad )⊙v ad ; With kw =MLP(e kw )⊙v kw ; z=[z ad , z kw , z ad ⊙z kw ]; where e ad and e kw are the embedding vectors of advertising creativity and keywords respectively, v ad and v kw are the additional feature vectors of advertising creativity and keywords respectively, ⊙ is the Hadamard product (element-wise product), [·,·] represents vector concatenation, and MLP stands for multi-layer perceptron.

6. The AI-based ASA advertising keyword and material generation system according to claim 1, characterized in that: The advertisement delivery optimization module optimizes advertisement delivery according to the following steps: Step 5.1 Use the Hierarchical Multi-Armed Bandit Strategy (HMAB) for ad delivery optimization: Among them, a represents the advertising strategy, EU PK (a) represents the expected utility calculated based on prior knowledge, n a represents the number of times strategy a is selected, n i represents the number of selections of the i-th strategy, α represents the exploration factor used to balance exploration and exploitation, represents the total number of selections for all strategies; Step 5.2: Use evolutionary algorithms to optimize advertising bids: Fitness(b)=ω l ×CTR b +ω2×Rank b -ω3×CPC b +ω4×Quality b ; Among them, b represents the advertising bid vector, CTR b Represents the estimated click rate, Rank b Represents the ad ranking score, CPC b stands for cost per click, Quality b represents the advertising quality score, ω1, ω2, ω3, and ω4 represent the weight coefficients of click-through rate, ranking, cost, and quality, respectively.

Images