Advertisement putting management platform and management terminal thereof

Abstract

The invention relates to the field of advertisement putting, and provides an advertisement putting management platform and a management terminal thereof, and the platform comprises an account management module which provides account registration and login for a user; the advertisement putting management module is used for providing advertisement putting setting for an advertiser; the advertisement creativity management module provides advertisement creativity for an advertiser and supports the user to upload or create the advertisement creativity; the audience orientation and marketing module collects and arranges user data, constructs a user portrait, and helps an advertiser to carry out audience orientation and accurately put an advertisement to a target user group according to the user portrait; and the data monitoring and effect evaluation module is used for collecting advertisement putting data for the put advertisement, creating a data report according to the advertisement putting data, and analyzing and evaluating the advertisement data. According to the technical scheme, the problem that the advertisement effect is difficult to improve due to the fact that an advertisement putting management platform cannot accurately position audiences in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of advertising delivery, and in particular to an advertising delivery management platform and a management terminal thereof. Background Art

[0002] With the rapid development of internet technology, ad management platforms play a vital role in the advertising industry. Built on computer and network communication technologies, ad management platforms provide advertisers, advertising agencies, and related operations personnel with a centralized, intelligent operating environment for planning, executing, monitoring, and optimizing advertising campaigns.

[0003] In the current advertising landscape, most platforms rely on basic user information, such as demographic characteristics like age, gender, and region, as well as simple behavioral data, such as browsing history and purchase history, to determine audience targeting. This targeting approach has the following shortcomings:

[0004] On the one hand, relying solely on demographic characteristics and limited behavioral data cannot fully and deeply understand users' true needs and interests. Users are complex and multifaceted individuals, and their consumption decisions are often influenced by multiple underlying factors that may be overlooked in traditional audience analysis. For example, two users of the same age and gender with similar browsing history may react very differently to the same advertisement due to different emotional states, life situations, or values.

[0005] On the other hand, with intensified market competition and the explosive growth of user information, advertising information overload is becoming increasingly serious. Without precise targeting of audiences truly interested in the ad content, ads can easily be lost in the flood of information. This extensive delivery method results in ads being shown to a large number of irrelevant or low-relevance users, not only wasting advertising budgets but also making it difficult to effectively improve key advertising performance metrics like click-through rates and conversion rates. Summary of the Invention

[0006] The present invention provides an advertisement delivery management platform and a management terminal thereof, which improve the advertisement delivery effect by using a more accurate audience positioning method.

[0007] The technical solutions of the present invention are as follows:

[0008] An advertising delivery management platform, comprising:

[0009] The account management module provides users with account registration, login, and management permissions for different user roles;

[0010] The advertising delivery management module provides advertisers with advertising delivery settings, allowing advertisers to set advertising delivery goals and formulate detailed delivery strategies based on the goals;

[0011] Ad creative management module, which provides advertising creatives to advertisers and supports users to upload or create advertising creatives;

[0012] The audience targeting and marketing module collects and organizes user data, builds user portraits, and helps advertisers target audiences based on user portraits, accurately delivering advertisements to target user groups.

[0013] The data monitoring and effect evaluation module collects advertising data for advertisements that have been delivered, creates data reports based on the advertising data, and analyzes and evaluates the advertising data.

[0014] Furthermore, the advertisement delivery management module includes:

[0015] The target selection and analysis unit provides advertisers with advertising delivery targets and provides target audience analysis based on the user data and user sentiment data provided by the audience targeting and marketing module. Through sentiment semantic network and multimodal sentiment natural language analysis, it deeply understands the emotional changes of users throughout the process of engaging with the product or brand, and dynamically adjusts the target audience analysis based on the changes in user sentiment data.

[0016] The delivery channel selection unit provides advertisers with advertising delivery channels, analyzes the target audiences of different channels based on the user data and user sentiment data of different channels provided by the audience targeting and marketing module, and recommends target channels to advertisers based on the target audience analysis of different channels;

[0017] The budget allocation unit develops budget allocation plans based on the advertiser's budget and advertising goals. This includes delivering advertising content tailored to the user's emotional journey stage and monitoring advertising effectiveness. This includes using multimodal sentiment analysis technology to monitor users' emotional reactions to ads in real time during the advertising process and dynamically adjusting channel budgets.

[0018] The delivery time control unit analyzes the target audience's behavioral habits and advertising acceptance at different times to determine the optimal advertising delivery time.

[0019] Furthermore, the advertising creative management module includes:

[0020] The creative material upload management unit provides users with a channel for uploading creative advertising materials and categorizes the creative advertising materials uploaded by users;

[0021] The creative editing management unit provides users with online creative editing and records different versions of each creative so that advertisers can choose different versions of the creative;

[0022] The creative review and approval unit uses a combination of manual and automated review to verify whether user-provided creative advertising materials contain any illegal content. It also establishes an approval process to conduct a step-by-step review of the innovation, accuracy, and legal risks of advertising copy and images.

[0023] The dynamic creative generation unit automatically generates advertising creative materials that match the current user emotional data based on the user data and user emotional data provided by the audience targeting and marketing module, and adjusts the generated advertising creative materials in real time according to changes in user emotional data;

[0024] The personalized creative matching unit uses reinforcement learning algorithms to establish a matching model between users and advertising creatives. Based on the various data of target users, it gives a score for the expected effect of the corresponding creatives on the target users. At the same time, it collects users' actual feedback on the advertisements to strengthen the matching model.

[0025] Furthermore, the reinforcement learning algorithm includes a strategy and a value function, wherein the value function is constructed and updated by a Q-learning algorithm. The Q-learning algorithm is based on reward feedback of state-action pairs and uses an update formula to continuously adjust the values ​​in the Q-table to approximate the optimal value function. The strategy selects actions based on the updated Q-table, thereby realizing the reinforcement learning process.

[0026] The Q-learning algorithm maintains a Q-table Q(s,a) that records the estimated value of taking action a in state s. State s is constructed based on user feature information, including user demographic information, browsing history, purchasing behavior, and emotional state. Action a is selecting and delivering an advertising creative.

[0027] The Q-table is updated according to the following formula:

[0028] Q(s t , a t )←Q(s t , a t )+α[r t+1 +γmax a Q(s t+1 ,a)-Q(s t , a t )],

[0029] Among them, s t is the state at time t, a t is the action taken at time t, r t+1 Taking actiont The immediate reward r obtained after α is the learning rate ranging from 0 to 1, γ is the discount factor ranging from 0 to 1, and max a Q(s t+1 ,a) indicates that in the next state s t+1 Take the maximum Q-value among all possible actions;

[0030] The instant reward r is based on the user's response to the advertisement. If the user clicks on the advertisement, r=1; if the user completes the purchase under the guidance of the advertisement, r=10; if the user has no response, r=0.

[0031] Furthermore, the strategy adopts an ε-greedy strategy, which randomly selects an action a with probability ε and selects the action with the maximum Q-value in the current Q-table with probability 1-ε, where ε is a parameter used to balance exploring new actions and utilizing existing experience.

[0032] Furthermore, the audience targeting and marketing module includes:

[0033] The user basic data collection unit collects and analyzes user data from various channels, integrates data from different channels, and establishes a unified identity for each user;

[0034] The user sentiment data collection unit, based on sentiment semantic networks and multimodal sentiment natural language analysis, collects multimodal information from various channels, including user text comments, social media expressions, and attitudes towards different advertisements, extracts sentiment features, and quantifies the sentiment features into sentiment polarity and sentiment intensity;

[0035] User profile building unit, which includes basic information such as age, gender, region, income level, and education level. Based on user behavior data across various channels, it explores their interests and hobbies and analyzes their behavior patterns, including purchasing behavior, browsing behavior, and search behavior.

[0036] The audience segmentation and targeting unit formulates rules based on advertisers' needs and divides users into different segments according to the rules. Based on user similarities, it automatically divides them into different groups through clustering algorithms. Each group has similar characteristics and behavior patterns.

[0037] The audience emotion segmentation and targeting unit dynamically targets users based on their emotional characteristics, segmenting them into highly positive emotion users, moderately positive emotion users, neutral emotion users, low negative emotion users, and highly negative emotion users, and targeting users with different emotional states;

[0038] The emotional dynamic monitoring unit uses multimodal emotional analysis technology to track the user's emotional changes, dynamically modify the user's emotional state, and synchronously modify the user's orientation.

[0039] Furthermore, the sentiment semantic network includes:

[0040] Data collection and preprocessing: collecting text data such as user comments and social media posts, processing the text data to remove noise, and performing lexical analysis on the processed text data to segment sentences into words or phrases;

[0041] Construct a vocabulary-emotion association matrix. Use words or characters with emotions to construct vocabulary-emotion associations. For each word, count its frequency of occurrence under different emotional polarities and construct a vocabulary-emotion association matrix. The vocabulary set is V = {v1, v2, ..., v n}, the sentiment polarity set is E = {e1, e2, e3}, and the vocabulary-sentiment association matrix M is an n×3 matrix, where M ij Represents vocabulary v i In emotional polarity j The probability of occurrence, v i Taken from the vocabulary set, e j Taken from the sentiment polarity set, in which e1, e2, and e3 represent positive, negative, and neutral, respectively;

[0042] Construct a semantic network. Based on the semantic relationship and emotional association between words, construct an emotional semantic network. Use a graph data structure to represent the emotional semantic network. Graph G = (V, E), where V is the node set of the words, E is the edge set of the semantic and emotional association between words, and the edge weights w between nodes ij By formula Calculate, where f ij is the probability that word i and word j appear together in the same emotional context, is the sum of the probabilities of word i and all other words co-occurring in the same emotional context, and the edge weight w ij represents the association strength of word j relative to word i in the sentiment semantic network;

[0043] Sentiment feature extraction, when analyzing new text, maps the words in the text to the sentiment semantic network, and proposes sentiment features by calculating the sentiment scores of the words in the text. For text T = {t1, t2, ..., t m}, its positive sentiment score S p By formula Calculate, where M ij is vocabulary v i Frequency under positive sentiment polarity, w ij is vocabulary v i The association weight with other positive sentiment words in the sentiment semantic network, similarly, calculate the negative sentiment score S nand neutral sentiment score S neu .

[0044] Furthermore, the multimodal emotional natural language includes:

[0045] Text modality analysis uses the sentiment semantic network method to perform sentiment analysis and obtain the sentiment characteristics of the text;

[0046] Visual modality analysis uses a camera or other device to capture facial image sequences of users as they interact with advertisements or products. Computer vision techniques are used to extract facial expression features, which are then fed into an expression classification model. The model maps facial expression features to corresponding emotion categories, including positive, negative, and neutral. The emotion intensity I is quantified based on the degree of expression d. The emotion intensity I is calculated using the formula I = kd + b, where k is the slope parameter, representing the effect of changes in the emotion degree d on the emotion intensity I, and b is the intercept parameter, representing the estimated value of the emotion intensity when the expression degree d = 0.

[0047] Audio modal analysis collects voice data from users when they interact with advertisements or products, and extracts acoustic features of the audio through signal processing technology. Acoustic features include fundamental frequency, Mel-frequency cepstral coefficients, volume, speaking rate, etc. The extracted acoustic features are input into the emotion classification model, and different emotional states are identified based on the acoustic features. The emotional intensity is quantified based on the numerical changes of the acoustic features. Let the acoustic feature vector be A = {a1, a2, ..., a n}, emotional intensity I is calculated through the multiple linear regression model I = β0+β1a1+β2a2+…+β n a n Calculate, where β i is the regression coefficient obtained by fitting the training data, representing the acoustic feature a i The degree of influence on emotional intensity I;

[0048] Multimodal fusion and sentiment feature extraction, using weighted fusion method, assuming the text sentiment feature vector is The visual emotion feature vector is The audio emotion feature vector is The fused emotional feature vector S is obtained by the formula S = ω T S T +ω V S V +ω A S A Calculate, where ω T 、ω V 、ω Ais the weight of each modality, and finally the comprehensive emotional features are extracted from the fused emotional feature vector S. The comprehensive emotional features include emotional polarity and emotional intensity.

[0049] Furthermore, the data monitoring and effect evaluation module includes:

[0050] Data dimensions and indicator units: for advertising data, statistics include ad exposure, click volume and click-through rate, conversion volume and conversion rate, and cost data; for user behavior data, statistics include user ad browsing behavior, interactive behavior, and purchasing behavior;

[0051] The emotional journey data integration unit collects emotional data from all aspects of user interaction with ads, including user expressions, user voice, product reviews, product prices, and product experience feedback at the time of ad exposure, after clicking on ads, during the purchase decision process, and after purchase. It deeply correlates this emotional data with traditional user behavior data to record changes in user data throughout the entire emotional journey.

[0052] The emotional journey stage division and indicator setting unit divides the emotional journey into the cognitive stage, the consideration stage, the decision stage, and the post-purchase stage. It analyzes the user's emotional and behavioral indicators for each stage and uses machine learning algorithms to build a comprehensive evaluation model based on the emotional and behavioral indicators at each stage of the emotional journey. This comprehensive evaluation model then infers the user's overall performance score for the entire emotional journey and the performance score for each stage.

[0053] The effectiveness evaluation unit conducts comparative analysis on the exposure, click-through rate, and conversion rate of different versions of advertisements to determine the version that is more popular with users. It analyzes factors such as user quality and cost-effectiveness of each channel for different delivery channels to determine the optimal channel combination. It analyzes the delivery effect of advertisements in different time periods to determine the optimal delivery time. It analyzes the role of advertisements in the entire user purchase journey through multi-touch attribution models such as linear attribution and time decay attribution, and provides more reasonable allocation of advertising budgets and optimization of delivery strategies based on the comprehensive effect score of users in their emotional journey.

[0054] An advertising delivery management terminal is provided, wherein the terminal is equipped with and runs any one of the above-mentioned advertising delivery management platforms.

[0055] The working principle and beneficial effects of the present invention are:

[0056] The present invention achieves a deeper understanding of users by comprehensively collecting and integrating multi-dimensional data of users (browsing history, purchasing behavior, etc.), and then constructs a detailed user portrait, accurately locates the target audience, and accurately delivers advertisements to the target user group that is truly interested by performing intelligent audience targeting based on the user portrait, thereby improving the accuracy and effectiveness of advertising delivery. By formulating targeted delivery strategies based on the goals set by advertisers (such as brand promotion, product sales, etc.), and selecting delivery channels and times in combination with the analysis of the behavioral habits of the target audience, the advertisements are more accurately delivered to the audience that may be interested, and by supporting the generation of diversified advertising creatives, advertisers can design content according to the characteristics of the target audience, and optimize the creativity to match it with the interests and needs of the audience, thereby attracting the attention of the target audience and improving the pertinence of the advertisement. By collecting multi-dimensional data of advertising delivery (exposure, click volume, etc.) in real time and analyzing and evaluating the effects, if a problem is found, it can be further analyzed whether it is a problem of creativity, strategy or positioning, thereby providing a basis for optimization, continuously adjusting the delivery strategy and audience targeting, thereby improving the ability of advertisements to accurately locate the audience and the advertising effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Figure 1 This is a schematic diagram of the structure of the advertising delivery management platform in the present invention;

[0059] Figure 2 This is a flow chart of the advertising delivery management platform in the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0061] like Figures 1 and 2 As shown, this embodiment proposes an advertising delivery management platform, including:

[0062] S1. Account management module: provides users with account registration and login, manages the permissions of different user roles (such as advertisers, advertising operators, financial personnel, etc.), and ensures that each user can only access and operate the functions and data for which they are authorized;

[0063] S2. Advertisement delivery management module provides advertisers with advertising delivery settings, allowing them to set advertising delivery goals, such as brand promotion, product sales, event publicity, etc., and formulate detailed delivery strategies based on the goals. According to the set strategies, advertisements are delivered to target channels, and the delivery process is monitored and managed to ensure that advertisements can be displayed normally as planned.

[0064] S3, the ad creative management module, provides advertisers with ad creatives and supports users to upload or create ad creatives. It provides functions for uploading, editing, and storing ad creatives, and manages ad creative versions, allowing advertisers to easily review historical creatives and compare the effectiveness of different creatives. It also supports the review and approval process for creatives to ensure that ad content complies with platform regulations and brand requirements.

[0065] S4, Audience Targeting and Marketing Module, collects and organizes user data, including demographic data (age, gender, region, etc.), behavioral data (browsing history, purchasing behavior, etc.), and hobby data, to build user profiles. Based on these profiles, advertisers can target their audiences and accurately deliver ads to user groups that meet the target audience characteristics, thereby improving advertising effectiveness and conversion rates.

[0066] S5. Data monitoring and effect evaluation module collects advertising data for the ads that have been delivered. It collects advertising data in real time, including exposure, clicks, click-through rate, conversion rate, cost data (such as cost per click CPC, cost per thousand impressions CPM, etc.), etc. It creates data reports based on the advertising data, presents them to advertisers through data reports and visualization tools, analyzes and evaluates the advertising data, and provides analysis and comparison of key indicators to help advertisers understand the effectiveness of advertising, identify problems and opportunities, and adjust delivery strategies in a timely manner.

[0067] In this embodiment, the advertisement delivery management module includes:

[0068] S201. The Target Selection and Analysis Unit provides advertisers with advertising objectives. Advertisers can select from a variety of objective types, such as increasing brand awareness, increasing product sales, promoting user registrations, and promoting events. Different objectives will guide subsequent strategy development. For example, if the goal is to increase brand awareness, the focus may be on ad exposure and reach; if the goal is to increase product sales, the focus may be on conversion rates and purchase guidance.

[0069] Based on the user data and user sentiment data provided by the audience targeting and marketing module, including but not limited to demographic characteristics (age, gender, region, income level, etc.), psychological characteristics (interests, hobbies, lifestyle, values, etc.), and behavioral characteristics (purchase history, browsing habits, search behavior, etc.), we conduct in-depth analysis of the target audience, provide target audience analysis, and build a detailed target audience portrait to more accurately target advertising targets.

[0070] Through sentiment semantic network and multimodal sentiment natural language analysis, we gain a deep understanding of the emotional changes in users throughout their interaction with a product or brand, while dynamically adjusting target audience analysis based on shifts in user sentiment data. From the first time a user hears about a product (awareness phase), through developing interest, comparative evaluation, and purchasing decisions, to post-purchase user experience and word-of-mouth communication, we create a detailed sentiment map, noting the positive, negative, or neutral emotional states that users may experience at each stage, as well as the factors influencing them. Based on the sentiment map, we identify the key factors that can effectively trigger positive user emotions or alter negative ones at each stage. These factors include, but are not limited to, specific advertising content, promotional information, user reviews, product demonstrations, and more.

[0071] S202: The channel selection unit continuously evaluates various advertising channels (such as search engines, social media platforms, video sharing platforms, industry vertical websites, email, etc.), including each channel's user traffic, user quality, user activity, advertising format (such as search ads, information flow ads, splash screen ads, etc.), and cost structure (such as CPC, CPM, CPA, etc.). It provides advertising channels to advertisers and analyzes the target audiences of different channels based on the user data and user sentiment data of different channels provided by the audience targeting and marketing module. Based on the target audience analysis of different channels, it recommends target channels to advertisers; at the same time, it ensures that advertising delivery across different channels can be coordinated with each other to form a coherent advertising communication system.

[0072] S203. The budget allocation unit develops a budget allocation plan based on the advertiser's budget and advertising objectives. This includes delivering advertising content tailored to the user's stage in the emotional journey. During the awareness stage, ads are run to pique users' curiosity and interest, such as creative and visually impactful short videos or engaging copy. During the purchase decision stage, ads are run to address user concerns by highlighting product advantages, value for money, and after-sales service, such as screenshots of user reviews and product quality assurance statements. This also includes budget allocation across different delivery channels, stages, and formats.

[0073] Monitor the effectiveness of advertising across all channels and ad formats. This includes leveraging multimodal sentiment analysis to monitor users' emotional reactions to ads in real time during the delivery process. Dynamically adjust channel budgets based on key metrics (such as click-through rate, conversion rate, and ROI) and real-time sentiment monitoring results. If a channel's advertising performance exceeds expectations, the budget can be increased appropriately. Conversely, if the performance is poor, the budget can be reduced or the delivery strategy adjusted. Furthermore, if users express annoyance or confusion when viewing an ad, the advertising strategy can be adjusted promptly.

[0074] S204: The delivery time control unit analyzes the target audience's behavior and advertising receptivity at different times of the day to determine the optimal time for advertising delivery. This includes different times of the day (e.g., office workers may have more time to browse ads in the evenings and on weekends), different days of the week (e.g., e-commerce platforms have higher traffic on weekends and promotional days), and different seasons of the year (e.g., travel products are in high demand before holidays). Based on these analyses, ads are scheduled to appear at the most favorable times to increase their exposure and click-through rate.

[0075] Control the cadence of advertising based on advertising objectives and the product lifecycle stage. For new product launches, intensive advertising can be used to quickly expand the market; for mature products, periodic pulsed advertising can be used to maintain brand engagement with users. At the same time, be mindful of avoiding excessive advertising that may lead to user fatigue or aversion, and insufficient advertising that may fail to achieve the desired results.

[0076] In this embodiment, the advertising creative management module includes:

[0077] S301, a creative material upload management unit provides users with a channel for uploading creative advertising materials and classifies the creative advertising materials uploaded by users according to multiple dimensions such as advertising campaign, product type, and delivery channel;

[0078] S302, the creative editing management unit, provides users with online creative editing, allowing them to perform a certain degree of editing on their material without the need for external software. For image material, this allows cropping, color adjustment, and text addition; for video material, this allows editing, subtitle addition, and special effects processing; and for text material, this allows formatting and grammar checking.

[0079] Recording different versions of each creative allows advertisers to choose from different creatives, compare changes between versions, and understand the direction of improvement and effectiveness of the creative. This helps track down issues when advertising performance is poor, or quickly find the right version when certain creative elements need to be reused.

[0080] S303, the Creative Review and Approval Unit, uses a combination of manual and automated review to verify whether user-submitted creative advertising materials contain any illegal content. Automated review uses keyword filtering and image recognition to check whether creatives contain any illegal content (e.g., pornography, violence, copyright infringement, etc.). Manual review, conducted by a professional team, examines the advertising's values, information authenticity, and compliance with platform regulations.

[0081] Establish an approval process to review the innovation, accuracy, and legal risks of advertising copy, advertising images, and other content step by step, ensuring that advertising creativity meets all requirements before it is released;

[0082] S304: The dynamic creative generation unit automatically generates advertising creative materials that match the current user's emotional data based on the user data and user emotion data provided by the audience targeting and marketing module, including the user's browsing behavior (such as pages viewed and dwell time), purchase history, geographic location, current device information, and emotional state (obtained through emotional semantic network and multimodal sentiment analysis). Based on different advertising objectives and product characteristics, the unit automatically generates advertising creative materials that match the current user's emotional data and adjusts the generated advertising creative materials in real time based on changes in the user's emotional data.

[0083] S305: The personalized creative matching unit uses a reinforcement learning algorithm to establish a matching model between users and ad creatives. Based on the target user's data, it generates a score for the expected performance of the corresponding creative within the target user. The model inputs the user's various feature data and outputs a score for the expected performance of each creative for that user. During ad delivery, the unit selects and delivers ad creatives to users based on the matching model's scores. It also collects actual user feedback (e.g., click-through rate, dwell time, conversion rate, etc.). This feedback serves as a reward signal for reinforcement learning, strengthening the matching model and adjusting its parameters to continuously optimize the model's prediction of user-creative matches.

[0084] In this example, reinforcement learning is a field within machine learning that focuses on how an agent takes a series of actions within an environment to maximize cumulative rewards. In the context of ad creative matching, the agent is the ad delivery system, the environment is the various conditions of users and the ad market, the action is selecting and delivering a specific ad creative to users, and the reward is the user's positive feedback on the ad (such as clicks, conversions, etc.).

[0085] Reinforcement learning algorithms include strategies and value functions. The value function is constructed and updated using the Q-learning algorithm. The Q-learning algorithm is based on reward feedback for state-action pairs and uses an update formula to continuously adjust the values ​​in the Q-table to approximate the optimal value function. The strategy selects actions based on the updated Q-table, thus achieving the reinforcement learning process.

[0086] The Q-learning algorithm maintains a Q-table Q(s,a) to record the estimated value of taking action a in state s. State s is constructed based on user characteristics, including user demographic information, browsing history, purchasing behavior, and emotional state. Action a is to select and launch an advertising creative.

[0087] The Q-table is updated according to the following formula:

[0088] Q(s t , a t )←Q(s t , a t )+α[r t+1 +γmax a Q(s t+1 ,a)-Q(s t , a t )],

[0089] Among them, s t is the state at time t, a t is the action taken at time t, r t+1 Taking action t The immediate reward r obtained after the reward is obtained, α is the learning rate between 0 and 1, which is used to control the speed of learning, γ is the discount factor between 0 and 1, which is used to measure the importance of future rewards, and max a Q(s t+1 ,a) indicates that in the next state s t+1 Take the maximum Q-value among all possible actions;

[0090] The immediate reward r is based on the user's response to the ad. If the user clicks on the ad, r = 1; if the user completes the purchase under the guidance of the ad, r = 10; if the user has no response, r = 0.

[0091] In this embodiment, the strategy adopts the ε-greedy strategy. The ε-greedy strategy is used to balance the exploration of new actions and the use of existing experience during the training process. The ε-greedy strategy randomly selects an action a with probability ε and selects the action with the largest Q-value in the current Q-table with probability 1-ε, where ε is a parameter used to balance the exploration of new actions and the use of existing experience.

[0092] In this embodiment, the audience targeting and marketing module includes:

[0093] S401. User Basic Data Collection Unit collects and analyzes user data from various channels, including but not limited to website browsing history (through website analytics tools such as Google Analytics), mobile app usage data (such as usage duration, frequency, and feature usage), social media activity (such as likes, comments, shared content, followed accounts and topics), offline purchase data (obtained through partnerships with retailers or loyalty programs), and customer service interaction records (such as inquiries and complaints). Data from different channels is integrated to create a unified identity for each user.

[0094] S402, user emotion data collection unit, based on the emotion semantic network and multimodal emotion natural language analysis, collects users' text comments, social media expressions, attitudes towards different advertisements and other multimodal information from various channels to extract emotion features, and quantifies the emotion features into emotion polarity (positive, negative, neutral) and emotion intensity (such as a score of 0-10). Users are divided into different emotion levels according to emotion polarity and intensity. For example, it can be divided into a highly positive emotion layer (with high emotion intensity and positive), a moderately positive emotion layer, a neutral emotion layer, a low negative emotion layer and a highly negative emotion layer. Users at each emotion level have different acceptance and reaction patterns to advertisements and products.

[0095] S403, user portrait construction unit, user portrait includes basic information such as age, gender, region, income level, education level, etc. Based on the user's behavioral data in various channels, their interests and hobbies are explored, and user behavior patterns are analyzed, including purchasing behavior (such as purchase frequency, purchase time, purchase product category, etc.), browsing behavior (such as browsed page type, browsing depth, dwell time, etc.), and search behavior (such as search keywords, search frequency, etc.).

[0096] S404, audience segmentation and targeting unit, formulates rules based on advertisers' needs, and divides users into different segment groups according to the rules. Based on the similarity of users, it automatically divides them into different groups through clustering algorithms, and each group has similar characteristics and behavior patterns.

[0097] S405, the audience emotion segmentation and targeting unit dynamically targets users based on their emotional characteristics, segmenting them into highly positive emotion users, moderately positive emotion users, neutral emotion users, low negative emotion users, and highly negative emotion users, and targeting users with different emotional states.

[0098] For highly positive emotional users, the focus should be on strengthening brand loyalty and promoting word-of-mouth communication. This should include delivering brand-value-added content (such as exclusive member discounts, brand stories, and cultural promotions), information on premium product upgrades, or inviting them to participate in brand community activities. Advertising creative should emphasize brand value, the emotional connection between users and the brand, and the unique experience.

[0099] For users with moderately positive emotions, the goal is to further deepen their positive feelings toward the brand and encourage them to move toward highly positive emotions. Ads should feature content such as introductions to new product features, user case studies, and styling suggestions. Ads should highlight the product's strengths and unique selling points, incorporating interesting creative elements like stylish outfit demonstration videos and detailed product images to entice users to learn more and purchase the product.

[0100] Neutral users: These users are still unsure about their attitude towards a brand or product and need to spark their interest. This requires offering attractive promotions, product trials, and comparative analysis of advantages compared to competitors. Advertising creative should focus more on capturing users' attention and sparking their curiosity.

[0101] For users with low negative emotions, it's important to alleviate their negative emotions before attempting to change their attitudes. Advertising should include explanations of the brand's improvement measures, the results of addressing user feedback, and targeted problem-solving solutions. Ads should demonstrate the brand's commitment to user feedback and its proactive approach to improvement.

[0102] Highly negative users require careful handling to avoid excessive intrusion. Use surveys and customer care to understand their specific issues and attempt to provide compensation or improvement measures. Advertising creative should be extremely cautious, emphasizing the brand's commitment to improvement and respect for users.

[0103] S406, the emotional dynamic monitoring unit, uses multimodal emotional analysis technology to track the user's emotional changes during the advertising delivery process. For example, by analyzing the user's immediate emotional reaction after seeing the advertisement (such as changes in expression, voice tone) and subsequent behavioral feedback (such as whether to click, whether to express new emotions in the comments), the user's emotional level information is updated.

[0104] Dynamically modify user emotional states and simultaneously modify user targeting. Simultaneously, timely adjust ad delivery strategies and creative content based on dynamic changes in user emotions. If a user's emotional state shifts from a low-negative to a neutral level, appropriately increase ad frequency and adjust ad content, transitioning from addressing issues to showcasing product strengths. This emotion-driven, layered audience targeting approach can more accurately meet the needs of users with varying emotional states, improving advertising effectiveness and user satisfaction.

[0105] In this embodiment, the sentiment semantic network includes:

[0106] Data collection and preprocessing: First, collect a large amount of text data, such as user comments, social media posts, and other text data, and process the text data, including removing noise (such as HTML tags, special symbols, stop words, etc.). Perform lexical analysis on the processed text data to segment sentences into words or phrases. Word segmentation can use tools such as Jieba (Chinese) or NLTK (English) to segment sentences into words or phrases. Part-of-speech tagging can determine the part of speech of each word, such as noun, verb, adjective, etc., which helps with subsequent sentiment analysis because adjectives tend to better reflect emotional tendencies.

[0107] Construct a vocabulary-emotion association matrix. Use words or characters with emotions to construct vocabulary-emotion associations. For each word, count its frequency of occurrence under different emotional polarities and construct a vocabulary-emotion association matrix. The vocabulary set is V = {v1, v2, ..., v n}, the sentiment polarity set is E = {e1, e2, e3}, and the vocabulary-sentiment association matrix M is an n×3 matrix, where M ij Represents vocabulary v i In emotional polarity j The probability of occurrence, v i Taken from the vocabulary set, e j Taken from the sentiment polarity set, in which e1, e2, and e3 represent positive, negative, and neutral, respectively.

[0108] Construct a semantic network. Based on the semantic relationships between words (such as synonyms, antonyms, hyponyms, etc.) and emotional associations, construct an emotional semantic network. Use a graph data structure to represent the emotional semantic network. Graph G = (V, E), where V is the node set of the vocabulary (i.e., vocabulary), E is the edge set of the semantic and emotional associations between words (i.e., the semantic and emotional associations between words), and the edge weights w between nodes. ij By formula Calculate, where f ij is the probability that word i and word j appear together in the same emotional context, is the sum of the probabilities of word i and all other words co-occurring in the same emotional context, and the edge weight w ij Represents the association strength of word j relative to word i in the sentiment semantic network.

[0109] Sentiment feature extraction, when analyzing new text, maps the words in the text to the sentiment semantic network, and proposes sentiment features by calculating the sentiment scores of the words in the text. For text T = {t1, t2, ..., t m}(t iis a word in the text), and its positive sentiment score S p By formula Calculate, where M ij is vocabulary v i Frequency of words with positive sentiment polarity (obtained from the word-sentiment association matrix), w ij is vocabulary v i The association weight with other positive sentiment words in the sentiment semantic network, similarly, calculate the negative sentiment score S n and neutral sentiment score S neu .

[0110] In this embodiment, the multimodal emotional natural language includes:

[0111] Text modality analysis: For text parts (such as user comments and social media copywriting), sentiment analysis is performed using the sentiment semantic network method to obtain the emotional characteristics of the text;

[0112] Visual modality analysis uses a camera or other device to capture facial image sequences of users interacting with advertisements or products. Computer vision techniques, such as convolutional neural networks (CNNs), are then used to extract facial expression features. For example, trained face recognition models (such as those in OpenCV or deep learning-based models like FaceNet) are used to locate the position and shape changes of key facial features (such as the eyes, mouth, and eyebrows). These extracted facial expression features are then fed into an expression classification model, which can be based on support vector machines (SVMs) or deep learning models (such as deep belief networks (DBNs) or recurrent neural networks (RNNs). This model maps facial expression features to corresponding emotion categories, including positive, negative, and neutral. The emotion intensity, I, is quantified based on the degree of expression, d. The emotion intensity, I, is calculated using the formula I = kd + b, where k is the slope parameter, representing the degree to which changes in emotion degree, d, affect the emotion intensity, I. This k is obtained by fitting a large dataset of labeled emotion intensity and expression degree. By using linear regression and other methods, we can calculate the ratio of the covariance between the expression degree and the emotion intensity to the variance of the expression degree to obtain k. Assuming there are n sample data points (d1, I1), (d2, I2), ..., (d n , I n ), the calculation formula of k is in is the sample mean of expression level d, is the sample mean of the emotion intensity I. The meaning of this formula is that k measures the closeness and direction of the linear relationship between the degree of expression and the emotion intensity. It represents the average change in emotion intensity caused by each unit change in the degree of expression. b is the intercept parameter, which expresses the estimated value of emotion intensity when the degree of expression d = 0. In the training data, b is also obtained by fitting. After calculating k, it can be calculated by the formula b is calculated to shift the overall data distribution. It takes into account the presence of underlying emotional intensity even when the expression level is zero, due to other factors (such as the individual's emotional baseline). For example, some users may have a slightly positive or negative emotional tendency even without a noticeable change in expression (d = 0). This tendency is reflected in b.

[0113] Audio modal analysis collects voice data from users when they interact with advertisements or products, such as voice comments and voice customer service records. The acoustic features of the audio are extracted through signal processing technology. Acoustic features include fundamental frequency, Mel-frequency cepstral coefficients (MFCC), volume, speaking rate, etc. The extracted acoustic features are input into the sentiment classification model. This model can be based on Gaussian mixture models (GMM) and deep learning models (such as long short-term memory networks (LSTMs) and convolutional neural networks (CNNs)). Different emotional states, such as excitement, calmness, and boredom, can be identified based on acoustic features. The intensity of emotion is quantified based on the numerical changes in the acoustic features. Let the acoustic feature vector be A = {a1, a2, ..., a n}, emotional intensity I is calculated through the multiple linear regression model I = β0+β1a1+β2a2+…+β n a n Calculate, where β i is the regression coefficient obtained by fitting the training data, representing the acoustic feature a i The degree of influence on emotional intensity I;

[0114] Multimodal fusion and sentiment feature extraction, using weighted fusion method, assuming the text sentiment feature vector is (positive, negative, and neutral sentiment scores respectively), the visual sentiment feature vector is The audio emotion feature vector is The fused emotional feature vector S is obtained by the formula S = ω T S T +ω V S V +ω A S A Calculate, where ω T 、ω V 、ω Ais the weight of each modality, and finally the comprehensive emotional features are extracted from the fused emotional feature vector S. The comprehensive emotional features include emotional polarity (determined by comparing the positive and negative scores in S) and emotional intensity (determined by calculating the difference between the positive and negative scores in S, etc.).

[0115] In this embodiment, the data monitoring and effect evaluation module includes:

[0116] S501. Data Dimensions and Indicators Unit: Statistics are collected for ad placement data, including ad impressions, clicks and click-through rates, conversions and conversion rates, and cost data. Impressions: Counts the number of times an ad is displayed across various delivery channels. By comparing impression data across different channels and time periods, we can understand the breadth of ad placement and the reach of each channel. Clicks and CTR: Clicks refer to the number of times users click on an ad, while CTR is the ratio of clicks to impressions, reflecting the extent to which an ad captures user attention and sparks user interest. Conversions and Conversion Rate: Conversion rate refers to the proportion of users who complete a specific target action (such as purchasing a product, registering an account, downloading an app, etc.) after viewing an ad. Conversions refer to the number of users who actually complete a conversion. These metrics are key indicators for measuring ad effectiveness and are directly related to advertisers' marketing goals. Cost data: This includes cost per click (CPC), cost per thousand impressions (CPM), and cost per action (CPA). CPC is the fee an advertiser pays each time a user clicks on an ad, CPM is the cost per 1,000 ad impressions, and CPA is the cost per conversion.

[0117] User behavior data is used to collect statistics on user ad browsing, interactive behavior, and purchasing behavior. Ad browsing behavior: Monitors the user's browsing path after seeing the ad, including the pages viewed, dwell time, scrolling depth, etc. This can help understand the user's attention to the ad content and whether they further explore related information. Interactive behavior: Such as users' likes, comments, and shares of ads. These interactive behaviors reflect the user's emotional inclination and engagement with the ad. Purchasing behavior: Records the user's conversion volume. At the same time, it records information such as the type, amount, and time of purchase of products purchased by the user. This helps analyze the user's purchasing preferences and consumption patterns, and provides a basis for subsequent advertising strategy adjustments and product recommendations.

[0118] S502, emotional journey data integration unit, collects emotional data from all aspects of user interaction with advertisements, including when the advertisement is exposed (multimodal emotional reactions such as user's facial expression, voice tone, etc. when seeing the advertisement), after clicking the advertisement (emotional evaluation of the advertisement content, such as inferred through changes in user's facial expression and interactive behavior during the period of staying on the page), the purchase decision process (emotional tendencies towards product evaluation, price, etc.), and emotional data such as user's facial expression, user voice, product evaluation, product price, product experience feedback, etc. after purchase. The emotional data is deeply associated with traditional user behavior data to record the data changes of users throughout the emotional journey.

[0119] S503, emotional journey stage division and indicator setting unit, divides the emotional journey into the cognitive stage, the consideration stage, the decision stage, and the post-purchase stage.

[0120] The emotional indicators in the cognitive stage are the interest aroused by the advertisement (the proportion of positive emotions such as curiosity and excitement that users have when seeing the advertisement is determined through facial expression analysis) and the brand impression (based on the user's emotional response to brand elements in the advertisement). The behavioral indicators are the duration of the advertisement and whether there is any active behavior to understand brand or product information (such as clicking on the brand logo or product introduction link in the advertisement).

[0121] The emotional indicators in the consideration stage are the degree of understanding and emotional inclination of product functions and features (by analyzing the user's attention and emotional response to the product introduction part in the advertising content), and the emotional advantage compared with competing products (if there is competitive product comparison content in the advertisement); behavioral indicators are the time spent on the product details page and the number of interactions with customer service or online consultation functions (reflecting the user's questions about the product and the need for further understanding).

[0122] The emotional indicators in the decision-making stage are the degree of acceptance of price and emotional response (such as satisfaction, hesitation, etc.), and the intensity of purchase intention (by analyzing the user's expression and behavior near the purchase button); the behavioral indicators are whether the product is added to the shopping cart and the completion of the purchase behavior.

[0123] The emotional indicators in the post-purchase stage are satisfaction with the product experience (through users' emotional expressions in product reviews and social media sharing) and the degree of brand loyalty improvement (whether there is a willingness to purchase again and the possibility of recommending to others); behavioral indicators are whether to purchase again, product sharing and recommendation behavior (such as shared links and the number of recommendations to friends).

[0124] We analyze users' emotional and behavioral indicators at each stage of the emotional journey. Using machine learning algorithms (using a multi-layer perceptron within a neural network to construct a comprehensive evaluation model), we build a comprehensive evaluation model based on these indicators at each stage of the emotional journey. This model then infers the overall effectiveness score for the user's entire emotional journey and the effectiveness score for each stage. During model training and optimization, we utilize extensive historical user data, adjusting model parameters to ensure that the model accurately reflects the relationship between emotional and behavioral data and advertising effectiveness.

[0125] S504: The effectiveness evaluation unit conducts A / B testing or multivariate testing on different creative versions of the same advertisement (e.g., different copy, images, and video styles). By comparing and analyzing the impressions, click-through rates, and conversion rates of different advertisement versions, the more popular version is determined.

[0126] Compare the effectiveness of advertising across different channels (e.g., search engines, social media, video platforms, etc.). Analyze factors such as user quality and cost-effectiveness of each channel to determine the optimal channel combination.

[0127] Analyze the effectiveness of advertising in different time periods (such as weekdays and weekends, daytime and nighttime, different seasons, etc.) Analyze the effectiveness of advertising in different time periods and determine the best time to advertise.

[0128] Analyze the role of advertising in the entire user purchase journey through multi-touchpoint attribution models such as linear attribution and time decay attribution, and combine the comprehensive effect score of users in the emotional journey to provide more reasonable allocation of advertising budgets and optimize delivery strategies.

[0129] An advertising delivery management terminal is provided, wherein the terminal is equipped with and runs any one of the above-mentioned advertising delivery management platforms.

[0130] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An advertising delivery management platform, characterized in that: include: The account management module provides users with account registration, login, and management permissions for different user roles; The advertising delivery management module provides advertisers with advertising delivery settings, allowing advertisers to set advertising delivery goals and formulate detailed delivery strategies based on the goals; Ad creative management module, which provides advertising creatives to advertisers and supports users to upload or create advertising creatives; The audience targeting and marketing module collects and organizes user data, builds user portraits, and helps advertisers target audiences based on user portraits, accurately delivering advertisements to target user groups. The data monitoring and effect evaluation module collects advertising data for advertisements that have been delivered, creates data reports based on the advertising data, and analyzes and evaluates the advertising data.

2. The advertising delivery management platform according to claim 1, characterized in that: The advertising delivery management module includes: The target selection and analysis unit provides advertisers with advertising delivery targets and provides target audience analysis based on the user data and user sentiment data provided by the audience targeting and marketing module. Through sentiment semantic network and multimodal sentiment natural language analysis, it deeply understands the emotional changes of users throughout the process of engaging with the product or brand, and dynamically adjusts the target audience analysis based on the changes in user sentiment data. The delivery channel selection unit provides advertisers with advertising delivery channels, analyzes the target audiences of different channels based on the user data and user sentiment data of different channels provided by the audience targeting and marketing module, and recommends target channels to advertisers based on the target audience analysis of different channels; The budget allocation unit develops budget allocation plans based on the advertiser's budget and advertising goals. This includes delivering advertising content tailored to the user's emotional journey stage and monitoring advertising effectiveness. This includes using multimodal sentiment analysis technology to monitor users' emotional reactions to ads in real time during the advertising process and dynamically adjusting channel budgets. The delivery time control unit analyzes the target audience's behavioral habits and advertising acceptance at different times to determine the optimal advertising delivery time.

3. The advertising delivery management platform according to claim 1, characterized in that: The advertising creative management module includes: The creative material upload management unit provides users with a channel for uploading creative advertising materials and categorizes the creative advertising materials uploaded by users; The creative editing management unit provides users with online creative editing and records different versions of each creative so that advertisers can choose different versions of the creative; The creative review and approval unit uses a combination of manual and automated review to verify whether user-provided creative advertising materials contain any illegal content. It also establishes an approval process to conduct a step-by-step review of the innovation, accuracy, and legal risks of advertising copy and images. The dynamic creative generation unit automatically generates advertising creative materials that match the current user emotional data based on the user data and user emotional data provided by the audience targeting and marketing module, and adjusts the generated advertising creative materials in real time according to changes in user emotional data; The personalized creative matching unit uses reinforcement learning algorithms to establish a matching model between users and advertising creatives. Based on the various data of target users, it gives a score for the expected effect of the corresponding creatives on the target users. At the same time, it collects users' actual feedback on the advertisements to strengthen the matching model.

4. The advertising delivery management platform according to claim 3, characterized in that: The reinforcement learning algorithm includes a strategy and a value function, wherein the value function is constructed and updated by a Q-learning algorithm. The Q-learning algorithm is based on reward feedback of state-action pairs and uses an update formula to continuously adjust the values ​​in the Q-table to approximate the optimal value function. The strategy selects actions based on the updated Q-table, thereby realizing the reinforcement learning process; The Q-learning algorithm maintains a Q-table Q(s,a) that records the estimated value of taking action a in state s. State s is constructed based on user feature information, including user demographic information, browsing history, purchasing behavior, and emotional state. Action a is selecting and delivering an advertising creative. The Q-table is updated according to the following formula: Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γmax a Q(s t+1 ,a)-Q(s t ,a t )], Among them, s t is the state at time t, a t is the action taken at time t, r t+1 Taking action t The immediate reward r obtained after α is the learning rate ranging from 0 to 1, γ is the discount factor ranging from 0 to 1, and max a Q(s t+1 ,a) indicates that in the next state s t+1 Take the maximum Q-value among all possible actions; The instant reward r is based on the user's response to the advertisement. If the user clicks on the advertisement, r=1; if the user completes the purchase under the guidance of the advertisement, r=10; if the user has no response, r=0.

5. An advertising delivery management platform according to claim 4, characterized in that: The strategy adopts an ε-greedy strategy, which randomly selects an action a with probability ε and selects the action with the maximum Q-value in the current Q-table with probability 1-ε, where ε is a parameter used to balance exploring new actions and utilizing existing experience.

6. The advertising delivery management platform according to claim 1, characterized in that: The audience targeting and marketing module includes: The user basic data collection unit collects and analyzes user data from various channels, integrates data from different channels, and establishes a unified identity for each user; The user sentiment data collection unit, based on sentiment semantic networks and multimodal sentiment natural language analysis, collects multimodal information from various channels, including user text comments, social media expressions, and attitudes towards different advertisements, extracts sentiment features, and quantifies the sentiment features into sentiment polarity and sentiment intensity; User profile building unit, which includes basic information such as age, gender, region, income level, and education level. Based on user behavior data across various channels, it explores their interests and hobbies and analyzes their behavior patterns, including purchasing behavior, browsing behavior, and search behavior. The audience segmentation and targeting unit formulates rules based on advertisers' needs and divides users into different segments according to the rules. Based on user similarities, it automatically divides them into different groups through clustering algorithms. Each group has similar characteristics and behavior patterns. The audience emotion segmentation and targeting unit dynamically targets users based on their emotional characteristics, segmenting them into highly positive emotion users, moderately positive emotion users, neutral emotion users, low negative emotion users, and highly negative emotion users, and targeting users with different emotional states; The emotional dynamic monitoring unit uses multimodal emotional analysis technology to track the user's emotional changes, dynamically modify the user's emotional state, and synchronously modify the user's orientation.

7. The advertising delivery management platform according to claim 6, characterized in that: The sentiment semantic network includes: Data collection and preprocessing: collecting text data such as user comments and social media posts, processing the text data to remove noise, and performing lexical analysis on the processed text data to segment sentences into words or phrases; Construct a vocabulary-emotion association matrix. Use words or characters with emotions to construct vocabulary-emotion associations. For each word, count its frequency of occurrence under different emotional polarities and construct a vocabulary-emotion association matrix. The vocabulary set is V = {v1, v2, ..., v n }, the sentiment polarity set is E = {e1, e2, e3}, and the vocabulary-sentiment association matrix M is an n×3 matrix, where M ij Represents vocabulary v i In emotional polarity j The probability of occurrence, v i Taken from the vocabulary set, e j Taken from the sentiment polarity set, in which e1, e2, and e3 represent positive, negative, and neutral, respectively; Construct a semantic network. Based on the semantic relationship and emotional association between words, construct an emotional semantic network. Use a graph data structure to represent the emotional semantic network. Graph G = (V, E), where V is the node set of the words, E is the edge set of the semantic and emotional association between words, and the edge weights w between nodes ij By formula Calculate, where f ij is the probability that word i and word j appear together in the same emotional context, is the sum of the probabilities of word i and all other words co-occurring in the same emotional context, and the edge weight w ij represents the association strength of word j relative to word i in the sentiment semantic network; Sentiment feature extraction, when analyzing new text, maps the words in the text to the sentiment semantic network, and proposes sentiment features by calculating the sentiment scores of the words in the text. For text T = {t1, t2, ..., t m }, its positive sentiment score S p By formula Calculate, where M ij is vocabulary v i Frequency under positive sentiment polarity, w ij is vocabulary v i The association weight with other positive sentiment words in the sentiment semantic network, similarly, calculate the negative sentiment score S n and neutral sentiment score S neu .

8. The advertising delivery management platform according to claim 6, characterized in that: The multimodal emotional natural language includes: Text modality analysis uses the sentiment semantic network method to perform sentiment analysis and obtain the sentiment characteristics of the text; Visual modality analysis uses a camera or other device to capture facial image sequences of users as they interact with advertisements or products. Computer vision techniques are used to extract facial expression features, which are then fed into an expression classification model. The model maps facial expression features to corresponding emotion categories, including positive, negative, and neutral. The emotion intensity I is quantified based on the degree of expression d. The emotion intensity I is calculated using the formula I = kd + b, where k is the slope parameter, representing the effect of changes in the emotion degree d on the emotion intensity I, and b is the intercept parameter, representing the estimated value of the emotion intensity when the expression degree d = 0. Audio modal analysis collects voice data from users when they interact with advertisements or products, and extracts acoustic features of the audio through signal processing technology. Acoustic features include fundamental frequency, Mel-frequency cepstral coefficients, volume, speaking rate, etc. The extracted acoustic features are input into the emotion classification model, and different emotional states are identified based on the acoustic features. The emotional intensity is quantified based on the numerical changes of the acoustic features. Let the acoustic feature vector be A = {a1, a2, ..., a n }, emotional intensity I is calculated through the multiple linear regression model I = β0+β1a1+β2a2+…+β n a n Calculate, where β i is the regression coefficient obtained by fitting the training data, representing the acoustic feature a i The degree of influence on emotional intensity I; Multimodal fusion and sentiment feature extraction, using weighted fusion method, assuming the text sentiment feature vector is The visual emotion feature vector is The audio emotion feature vector is The fused emotional feature vector S is obtained by the formula S = ω T S T +ω V S V +ω A S A Calculate, where ω T 、ω V 、ω A is the weight of each modality, and finally the comprehensive emotional features are extracted from the fused emotional feature vector S. The comprehensive emotional features include emotional polarity and emotional intensity.

9. The advertising delivery management platform according to claim 1, characterized in that: The data monitoring and effect evaluation module includes: Data dimensions and indicator units: for advertising data, statistics include ad exposure, click volume and click-through rate, conversion volume and conversion rate, and cost data; for user behavior data, statistics include user ad browsing behavior, interactive behavior, and purchasing behavior; The emotional journey data integration unit collects emotional data from all aspects of user interaction with ads, including user expressions, user voice, product reviews, product prices, and product experience feedback at the time of ad exposure, after clicking on ads, during the purchase decision process, and after purchase. It deeply correlates this emotional data with traditional user behavior data to record changes in user data throughout the entire emotional journey. The emotional journey stage division and indicator setting unit divides the emotional journey into the cognitive stage, the consideration stage, the decision stage, and the post-purchase stage. It analyzes the user's emotional and behavioral indicators for each stage and uses machine learning algorithms to build a comprehensive evaluation model based on the emotional and behavioral indicators at each stage of the emotional journey. This comprehensive evaluation model then infers the user's overall performance score for the entire emotional journey and the performance score for each stage. The effectiveness evaluation unit conducts comparative analysis on the exposure, click-through rate, and conversion rate of different versions of advertisements to determine the version that is more popular with users. It analyzes factors such as user quality and cost-effectiveness of each channel for different delivery channels to determine the optimal channel combination. It analyzes the delivery effect of advertisements in different time periods to determine the optimal delivery time. It analyzes the role of advertisements in the entire user purchase journey through multi-touch attribution models such as linear attribution and time decay attribution, and provides more reasonable allocation of advertising budgets and optimization of delivery strategies based on the comprehensive effect score of users in their emotional journey.

10. An advertisement delivery management terminal, characterized in that: The terminal is equipped with and runs any one of the advertising delivery management platforms in claims 1-9.

Images