Data-driven advertising effectiveness evaluation and analysis system and method

Abstract

This invention belongs to the field of advertising analysis and provides a data-driven advertising effectiveness evaluation system and method. Based on multi-source behavioral data before and after user interaction with an advertisement, a user's cognitive state vector is constructed. The changes in the user's cognitive state vector caused by the advertisement are modeled to generate a cognitive intervention vector, which represents the change in the user's cognitive state after exposure to the advertisement. Based on the user's verbal feedback, visual behavior, and emotional trajectory, the consistency between the user's behavioral response and the direction of the cognitive intervention vector is evaluated, and a multimodal behavioral consistency score is output. The user's behavioral path from ad click to final conversion is tracked, forming a conversion chain feature, identifying path length, path node type, and path response delay. Finally, the cognitive intervention vector, multimodal behavioral consistency score, and conversion chain feature are integrated to calculate a comprehensive cognitive effect score for advertising effectiveness evaluation.

Description

Technical Field

[0001] This invention belongs to the field of advertising analysis, and specifically relates to a data analysis-based advertising effectiveness evaluation and analysis system and method. Background Technology

[0002] As digital advertising technology continues to evolve, advertisers' evaluation of ad performance has gradually shifted from traditional coarse-grained metrics such as impressions and click-through rates to higher-level needs such as user intent recognition, behavior prediction, and precise attribution. Currently, most mainstream ad evaluation methods rely on explicit metrics such as user click behavior, page dwell time, or conversion behavior. While these can reflect a certain degree of user responsiveness, they struggle to depict the intrinsic impact of advertising on users' cognitive states.

[0003] Cognitive psychology research shows that users' responses to advertising are not impulsive but involve a complex psychological process involving information perception, interest arousal, cognitive processing, intention formation, and ultimately, behavioral execution. Using clicks or conversion results alone as the basis for evaluating advertising effectiveness often fails to reveal the true impact of advertising on users' cognitive systems. This is especially true in the promotion of high-value products or services, where users may experience cognitive shifts after their initial ad exposure but not immediately make a purchase. Such potential cognitive intervention effects are often overlooked in traditional evaluation systems.

[0004] Furthermore, current advertising evaluation systems generally lack systematic integration of user multimodal behavioral data (such as voice, facial expressions, and interaction patterns), making it impossible to achieve comprehensive modeling of the entire process from psychology to behavior to conversion, thus limiting the optimization space for advertising placement strategies. Summary of the Invention

[0005] To address the problems in existing technologies, this invention provides a data analysis-based advertising effectiveness evaluation and analysis system, comprising:

[0006] The user cognitive state acquisition module is used to construct a user cognitive state vector based on multi-source behavioral data before and after user contact with the advertisement.

[0007] The advertising intervention modeling module is used to model the changes in the user's cognitive state vector caused by advertising, and generate the cognitive intervention vector of the advertisement, which represents the amount of change in the user's cognitive state after exposure to the advertisement.

[0008] The multimodal consistency analysis module is used to evaluate the consistency between the user's behavioral response and the direction of the cognitive intervention vector based on the user's language feedback, visual behavior and emotional trajectory, and output a multimodal behavioral consistency score.

[0009] The conversion chain tracking module is used to track the user's behavioral path from ad click to final conversion, form conversion chain characteristics, and identify path length, path node type and path response latency.

[0010] The performance score generation module is used to integrate the cognitive intervention vector, multimodal behavioral consistency score and conversion chain features to calculate the comprehensive cognitive effect score of the advertising campaign for performance evaluation.

[0011] Furthermore, the user cognitive state acquisition module includes the following sub-modules:

[0012] The multi-source data collection submodule is used to collect multi-source behavioral data of users before and after contacting advertisements;

[0013] The multimodal feature extraction submodule is used to convert the collected multimodal data into feature vectors;

[0014] The cognitive state vector construction submodule is used to fuse multimodal features to construct the user's cognitive state vector.

[0015] Furthermore, the multimodal feature extraction submodule is used to extract features from webpage behavior data, language feedback data, and visual emotion data respectively, and convert the extracted features into a high-dimensional embedding representation in the same vector space for use by the cognitive state vector construction submodule.

[0016] Furthermore, the advertising intervention modeling module includes the following sub-modules:

[0017] The original state change extraction submodule is used to calculate the amount of original state change based on the cognitive state vector;

[0018] The intervention attribution modeling submodule is used to model state changes with advertising and user characteristics to generate intervention vectors;

[0019] The intervention vector regularization and label annotation submodule is used to perform regularization on intervention vectors and add semantic labels.

[0020] Furthermore, the intervention vector regularization and label annotation submodule is used to perform scale unification processing on the advertising intervention vector and bind corresponding cognitive semantic labels to each vector dimension. The labels include attention activation, emotional resonance, interest enhancement, and intent enhancement, which are used to enhance the interpretability and traceability of the intervention vector.

[0021] Furthermore, the multimodal consistency analysis module includes the following sub-modules:

[0022] The multimodal behavioral feature extraction submodule is used to extract behavioral features from language, visual, and emotion data;

[0023] The behavior vector unified projection submodule is used to uniformly project multimodal features onto the behavior response vector;

[0024] The consistency calculation submodule is used to calculate the consistency score between the behavioral response vector and the cognitive intervention vector.

[0025] Furthermore, the conversion chain tracking module includes the following sub-modules:

[0026] The Behavior Sequence Capture and Node Structuring submodule is used to record and structure the user's behavior node sequences;

[0027] The transformation chain feature extraction submodule is used to extract path length, node type and response latency features from the behavior sequence;

[0028] The conversion path classification and tagging submodule is used to classify conversion paths and label them.

[0029] The transformation chain structure encoding and output submodule is used to output structured transformation path feature vectors.

[0030] Furthermore, the conversion path classification and tagging submodule is used to classify user paths into fast response paths, deep engagement paths, hesitant paths, and bounce paths based on the behavioral node type, node order, response time distribution, and path density in the conversion path, and to assign tags representing the behavioral intent level of each type of path, in order to evaluate the structural performance of the advertising conversion chain.

[0031] Furthermore, the effect rating generation module includes the following sub-modules:

[0032] The feature standardization module is used to normalize the intervention intensity, consistency score, and conversion features;

[0033] The feature weighted fusion module is used to weight and combine various normalized features to generate a comprehensive score;

[0034] The threshold grading and interpretation module is used to grade the comprehensive score and output the score results and impact factors.

[0035] The present invention also provides a data analysis-based method for evaluating and analyzing the effectiveness of advertising campaigns, using any of the aforementioned data analysis-based advertising campaign effectiveness evaluation and analysis systems to perform the evaluation and analysis of advertising campaign effectiveness.

[0036] This invention provides a data-driven advertising effectiveness evaluation and analysis system and method. Focusing on the psychological and cognitive changes of users during ad exposure, and combining user behavior data and multimodal response signals, it constructs an organically collaborative framework for advertising intervention modeling, multimodal consistency scoring, and conversion path analysis, achieving the following beneficial effects:

[0037] By constructing a user's cognitive state vector before and after advertising and establishing an advertising intervention model, the specific impact of advertising on psychological dimensions such as user attention, interest, emotional arousal, and intent formation can be quantified, thereby breaking through the traditional click-centric evaluation method.

[0038] The system incorporates multimodal data sources such as user language feedback, visual behavior, and emotional trajectories. It uses a behavioral consistency scoring mechanism to verify whether cognitive interventions have been accepted by users and translated into actual responses, thereby enhancing the reliability and interpretability of the assessment model.

[0039] Through the conversion chain tracking module, the system can fully record and analyze the user's path behavior from ad exposure to conversion, extracting structural features such as path length, node type, and response latency, providing a foundation for identifying high-potential users and optimizing the path.

[0040] The performance rating generation module integrates three key factors: cognitive intervention intensity, multimodal response consistency, and conversion path structure. It outputs a comprehensive cognitive effect rating for advertising attribution and strategy optimization, which has the advantages of strong quantification ability, high interpretability, and strong adaptability.

[0041] In summary, the system provided by this invention constructs a cognitive-driven advertising effectiveness analysis framework from the perspective of user cognition, filling the gap in the existing technology of lacking modeling of advertising intervention mechanisms and verification of multimodal responses, and has high practical value and promotion prospects. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0044] The invention will now be described in preferred form with reference to the accompanying drawings and specific embodiments.

[0045] This embodiment solves the above problems through the following steps:

[0046] In one embodiment, reference Figure 1This invention provides a data analysis-based advertising effectiveness evaluation system. Specifically, it is a comprehensive system that uses multi-source user behavior data analysis, combined with user cognitive state modeling and multimodal consistency calculation, to quantitatively evaluate the influence of advertising, conversion paths, and cognitive response levels. The system constructs a user's cognitive state vector before and after ad exposure, models the cognitive shift caused by ad intervention, and integrates multimodal data such as user language feedback, visual behavior, and emotional trajectories to form a cognitive behavior consistency score. Combined with the path chain characteristics from user click to conversion, it calculates the comprehensive cognitive effect score of the ad, thereby achieving interpretable evaluation and refined optimization of advertising effectiveness.

[0047] The user cognitive state acquisition module is used to construct a user cognitive state vector based on multi-source behavioral data before and after user interaction with advertisements.

[0048] The core idea of ​​the user cognitive state acquisition module is that the true role of advertising is not merely to prompt users to click, but to guide users to undergo a psychological and cognitive shift, thereby driving subsequent behavioral changes. Cognitive psychology posits that individuals process information through an internal mechanism of "acceptance—understanding—evaluation—response." The effectiveness of advertising is reflected in the gradual cognitive evolution of users from "indifference" to "interest" and even "generating a purchase intention." Therefore, to scientifically evaluate whether advertising is truly effective, one cannot rely solely on behavioral indicators such as click-through rate or dwell time, but should instead construct a cognitive state vector model that can characterize changes in user psychology.

[0049] By analyzing multi-source behavioral data (such as clickstream, voice feedback, eye movement, and emotion recognition) before and after ad exposure, a series of feature indicators can be extracted. These features collectively constitute an expression of the user's cognitive state. Mathematically, this expression can be modeled as a high-dimensional vector called the "cognitive state vector," which represents the user's overall performance across multiple cognitive dimensions (attention, interest, emotional activation, purchase inclination, etc.). Comparing the cognitive state vectors before and after ad exposure allows for the calculation of cognitive shifts, thereby determining whether the ad effectively intervened in the user's cognitive system.

[0050] The user cognitive status acquisition module includes the following four core sub-modules:

[0051] 1. Multi-source data acquisition submodule

[0052] Function: Collects user behavior and physiological performance from multiple data sources in real time or near real time for subsequent cognitive state modeling.

[0053] enter:

[0054] User page view data and click behavior data before and after ad exposure;

[0055] The user's text or voice interaction content;

[0056] Optional visual and physiological data (such as facial expressions captured by a camera, heart rate collected by wearable devices, etc.);

[0057] User historical profile data (such as gender, occupation, and interest tags).

[0058] Processing logic:

[0059] Call the website or app's event tracking system to obtain page access history;

[0060] Access the ASR / NLP system to process user language content;

[0061] Access the CV module (optional) to extract facial expression recognition features or eye-tracking data;

[0062] By integrating a unified timestamp, data from the "pre-visit window" and "post-visit window" of the advertisement can be extracted in segments.

[0063] Output:

[0064] The original multimodal data set D after time alignment

[0065] 2. Multimodal Feature Extraction Submodule

[0066] Function: Converts raw behavioral and perceptual data into standardized feature vectors for subsequent modeling.

[0067] enter:

[0068] Data set D includes user web browsing behavior, language content, visual trajectories, etc.

[0069] Processing logic:

[0070] Webpage behavior characteristics: extracting page dwell time, scroll depth, mouse hotspot, and click frequency;

[0071] Language / text features: Use pre-trained language models (such as BERT) to extract semantic embeddings and perform sentiment analysis;

[0072] Visual features: Facial muscle activation and micro-expression frequency are extracted using facial expression recognition models (such as FER and OpenFace);

[0073] All features are uniformly normalized and standardized.

[0074] Output:

[0075] Eigenvectors F of different modes behavior ,F text ,F visualOptionally, Early Fusion or Late Fusion can be performed to form a unified input vector F. all .

[0076] 3. Cognitive State Vector Construction Submodule

[0077] Function: Based on multimodal feature fusion modeling, the cognitive state vector of users is used as an expression of their psychological state before and after advertising intervention.

[0078] enter:

[0079] The fused feature vector F all ;

[0080] Feature label dictionary and cognitive dimension mapping rules (used to interpret the psychological meaning of each dimension).

[0081] Processing logic:

[0082] Use multimodal fusion models, such as:

[0083] Attention-based Transformer encoder; multi-channel CNN fusion model; graph neural network (considering user history behavior chain graph);

[0084] Output cognitive state vector C = [C1, C2, ..., C d Each dimension represents a psychological cognitive indicator, such as attention index, interest level, emotional activation level, and information comprehension depth.

[0085] Output:

[0086] Vector C before ad exposure pre ;

[0087] Vector C after ad exposure post .

[0088] Specific examples:

[0089] A real estate company placed an advertisement for its new property development on a short video platform. The system recorded the following data about a user before and after viewing the advertisement:

[0090] Before the ad: The average page view lasted 5 seconds, with no clicks, and the facial expression remained stable;

[0091] After the ad: The time spent on the page increases to 25 seconds; clicking "View Details" will display smiles and focused expressions.

[0092] The text feedback contained the keywords "feels good" and "hadn't considered it before".

[0093] The system generates the following after processing:

[0094] Cpre = [0.3, 0.1, 0.2, 0.1] (Attention, Emotional Activation, Interest Matching, Purchase Intent)

[0095] C post =[0.7,0.6,0.8,0.5].

[0096] The advertising intervention modeling module is used to model the changes in user cognitive state vectors caused by advertising, and generate a cognitive intervention vector for advertising, which represents the amount of change in a user's cognitive state after exposure to an advertisement.

[0097] The design concept of the advertising intervention modeling module is based on the stimulus-response mechanism and causal modeling ideas in cognitive psychology. In the field of advertising evaluation, traditional methods focus on statistical metrics such as click-through rate (CTR) and conversion rate (CVR). However, these behavioral indicators cannot reveal how advertising changes users' attitudes and decisions at the psychological level. To understand the true value of advertising, it is necessary to quantify its direct intervention effect on users' cognitive states—that is, whether advertising triggers increased user attention, generated interest, activated emotions, or enhanced purchase intention.

[0098] Therefore, the role of the advertising intervention modeling module is to treat advertising as an input perturbation to the cognitive system, and to model and analyze the differences in users' cognitive states before and after advertising. By outputting a cognitive intervention vector, the system can capture the psychological intensity, direction, and multidimensional components of the advertising effect. This vector can be used for subsequent advertising attribution analysis and can also provide quantitative basis for material optimization and strategy adjustment.

[0099] This module is implemented using a modular structure and contains the following three sub-modules:

[0100] Submodule 1: Original State Change Extraction Module

[0101] Function: Extract the difference in users' original cognitive state before and after exposure to an advertisement, as the basis for subsequent modeling.

[0102] Implementation steps:

[0103] Receive vector data C provided by the front-end module pre C post ;

[0104] Calculate the original change vector: ΔC raw =C post -C pre ;

[0105] This vector initially reflects the direction and magnitude of changes in cognitive state, but causal attribution or noise removal is not performed.

[0106] Output: Original change vector ΔCraw .

[0107] Submodule 2: Intervention Attribution Modeling Module

[0108] Function: By incorporating the attributes and characteristics of the advertisement itself, user profile information, and external context, we can perform causal attribution analysis on changes in cognitive state and extract the dominant influence of the advertisement.

[0109] Implementation steps:

[0110] enter:

[0111] Original change vector ΔC raw ;

[0112] Advertising feature vector A∈R k (such as ad type, distribution channels, copywriting style, and emotional appeal of images);

[0113] User profile vector U∈R m (such as age, interests, and historical behavior);

[0114] Contextual features X∈R n (such as time, geographical location, concurrent activities, etc.)

[0115] Constructing intervention modeling functions:

[0116] I ad =f intv (ΔC raw ,A,U,X)

[0117] In the context, the function f intv It can be constructed based on any of the following methods:

[0118] Multi-task neural network modeling;

[0119] Heterogeneous graph neural network modeling (user-advertisement-context triples);

[0120] Interpretable causal modeling based on structural equation modeling (SEM);

[0121] Modeling by combining Bayesian networks with attention mechanisms.

[0122] Output cognitive intervention vector:

[0123] I ad ∈R d Each dimension represents the intensity of the advertisement's intervention on a certain cognitive indicator (such as attention, interest, or emotion);

[0124] Optionally, output the intervention intensity index S intensity =||I ad ||2.

[0125] Submodule 3: Intervention Vector Regularization and Label Annotation Module

[0126] Function: Normalizes intervention vectors and adds semantic labels to each dimension for subsequent interpretable output and decision support.

[0127] step:

[0128] Vector normalization improves comparability between different users;

[0129] Assign a cognitive dimension label to each component, such as:

[0130] Dimension 1 → Attention Activation

[0131] Dimension 2 → Degree of Emotional Resonance

[0132] 3rd Dimension → Intention Enhancement Factor

[0133] If the intervention model is an interpretable model (such as attention weights or regression coefficients), it can further output the importance scores of advertising factors in each dimension of intervention;

[0134] Output structured intervention vector results and annotation information for use by the visualization module and optimization engine.

[0135] Output: Final cognitive intervention vector I ad and semantic tag mapping table.

[0136] In a further implementation, the intervention modeling function is constructed as follows:

[0137] The model input is:

[0138] Original change vector ΔC raw ;

[0139] Advertising feature vector A∈R k ;

[0140] User profile vector U∈R m ;

[0141] Contextual features X∈R n .

[0142] Project each modal input onto a unified dimensional space:

[0143] h ΔC =W1·ΔC raw +b1

[0144] h A =W2·A+b2

[0145] h U =W3·U+b3

[0146] h X =W4·X+b4

[0147] in,

[0148] W i and b i These are the training parameters.

[0149] Construct a self-attention module to weight and fuse the importance of the four inputs:

[0150]

[0151] The Score function is a simple linear scoring method:

[0152] Score(h i ) = v T ·tanh(W a h i +b a )

[0153] Where v is a learnable parameter vector, h i Unified dimensional space projection, W a and b a These are the learnable parameters of the corresponding unified dimensional space projection.

[0154] The final fusion output is:

[0155]

[0156] The final intervention vector is output using a multilayer perceptron (MLP):

[0157]

[0158] Among them, I ad σ is the intervention vector; σ is the nonlinear activation function (such as ReLU or Swish); the rest are learnable parameters.

[0159] If the system already has conversion labels y∈{0,1} (e.g., whether to purchase), the following loss functions are jointly trained:

[0160] Intervention to predict loss:

[0161]

[0162] Conversion prediction loss:

[0163]

[0164] Where y is the sample label. For predicted labels.

[0165] Joint losses:

[0166]

[0167] λ1 and λ2 are preset weight parameters.

[0168] For example:

[0169] A user's cognitive state before and after being exposed to a coffee machine advertisement is as follows: C pre =[0.3,0.1,0.2,0.1](Attention, Emotional Activation, Interest, Purchase Intent)C post =[0.7,0.6,0.8,0.5]

[0170] Calculate the original state change vector:

[0171] ΔC raw =[0.4,0.5,0.6,0.4]

[0172] The advertising feature vector includes:

[0173] Copywriting style: Light and cheerful;

[0174] Main image color scheme: warm colors;

[0175] CTA: Emphasizing the emotional motivation of "rewarding yourself";

[0176] Launch time: 8:00 PM, during peak user activity hours.

[0177] User profile:

[0178] Age 35, prefers a high-quality lifestyle;

[0179] Historical browsing preferences: Small home appliances, lifestyle content;

[0180] Location: First-tier city.

[0181] After integrating the above information, the intervention modeling module outputs:

[0182] I ad =[0.35,0.47,0.55,0.33]

[0183] This vector indicates that advertising has a clear advantage in the two dimensions of "emotional activation" and "interest arousal".

[0184] The multimodal consistency analysis module is used to assess the consistency between the user's behavioral response and the direction of the cognitive intervention vector based on the user's language feedback, visual behavior and emotional trajectory, and output a multimodal behavioral consistency score.

[0185] The ultimate goal of advertising is to guide users to produce perceptible behavioral responses, such as clicks, purchases, and inquiries. However, the impact of advertising on users' cognitive states (i.e., the intervention vector) alone cannot guarantee that the impact is truly effective. User cognition may be shifted, but this may not translate into externally observable behavior, and there may even be spurious responses (such as habitual clicks). Therefore, it is necessary to introduce a verification mechanism to assess whether the user's actual behavior is consistent with the cognitive intervention direction predicted by the system.

[0186] The multimodal consistency analysis module is based on this concept. It extracts behavioral features from three modalities—language feedback, visual behavior, and emotional changes—and compares them with cognitive intervention vectors in terms of directionality and intensity to quantify their degree of consistency. The output of this module is a consistency score index, reflecting the extent to which the advertising intervention is actually received and responded to by users in reality, thereby improving the overall reliability and interpretability of the system.

[0187] In one specific implementation, the multimodal consistency analysis module consists of the following four modules:

[0188] Submodule 1: Multimodal Behavioral Feature Extraction Module

[0189] Function: Extract feature representations of three types of behavioral signals from data after users are exposed to advertisements.

[0190] enter:

[0191] Within 5 minutes of the user's ad exposure:

[0192] Language feedback data: such as customer service conversation texts, comments, and voice-to-text conversions;

[0193] Visual behavioral data: eye tracking, page scrolling behavior, mouse path;

[0194] Emotional trajectory data: facial expression recognition data, tone of voice emotion recognition results (such as cheerful, disgusted, etc.).

[0195] Output:

[0196] Behavioral feature vector B∈R d It is composed of the following concatenation of subvectors:

[0197] Language feedback embedding (e.g., via BERT encoding);

[0198] Visual behavioral characteristics (such as gaze duration and frequency of follow-up visits);

[0199] Characteristics of emotional changes (such as the amplitude of emotional fluctuations and the increase in positivity).

[0200] Submodule 2: Behavioral Vector Unified Projection Module Function: Maps multimodal behavioral features B to intervention vectors I ad Within the same cognitive space, define the transformation function g: The specific implementation method is as follows:

[0201] Use a separate linear projection for each mode:

[0202]

[0203] After splicing, a fully connected layer is used for unified fusion:

[0204]

[0205] in:

[0206] W1,W2,W3,W f ,b1,b2,b3,b f These are learnable parameters;

[0207] It is a behavioral response representation in the same space as the intervention vector.

[0208] Submodule 3: Consistency Calculation Module Function: Evaluate behavioral response vectors With intervention vector I ad The degree of consistency.

[0209] Using a fusion of directional consistency and amplitude consistency indicators:

[0210] Orientational consistency (cosine similarity):

[0211]

[0212] Consistency of amplitude ratio:

[0213]

[0214] Final consistency score (weighted fusion):

[0215] S consistency =α·S dir +(1-α)·S mag

[0216] Where α is a configurable weight parameter.

[0217] For example:

[0218] In the aforementioned scenario, the user's post-advertising cognitive intervention vector is:

[0219] I ad =[0.35,0.47,0.55,0.33]

[0220] The system collects user behavior data after advertising:

[0221] Verbal feedback: "I never thought about buying a coffee machine before, this feels like a great deal";

[0222] Browse the page for more than 30 seconds and repeatedly zoom in to view the images;

[0223] Facial expression recognition: from neutral to positive, with a smile lasting 6 seconds.

[0224] After feature extraction and projection, the behavior vector is:

[0225]

[0226] Calculate the consistency score:

[0227] S dir =0.996 (consistent in direction and height);

[0228] S mag ≈0.94 (behavioral response magnitude is close to intervention);

[0229] S consistency =0.7·0.996 + 0.3·0.94 = 0.978

[0230] System output:

[0231] Consistency score: 0.978;

[0232] Response tag: High consistency;

[0233] Dominant modalities: verbal feedback and visual behavior.

[0234] This indicates that the advertisement not only had a significant impact on the cognitive level, but also that users showed strong and genuine responses, representing a combination of high-quality creative materials and highly compatible users.

[0235] The conversion chain tracking module is used to track the user's behavioral path from ad click to final conversion, form conversion chain characteristics, and identify path length, path node type and path response latency.

[0236] In ad campaign analytics, whether a user ultimately converts is not determined by a single click or cognitive response, but rather by a sequence of multiple interactive behaviors. This sequence of behaviors often includes multiple "node events"—such as clicking an ad, browsing product details, adding to cart, contacting customer service, registering an account, and submitting an order—which together constitute the user's conversion chain.

[0237] The endpoint (whether or not a conversion occurs) alone cannot fully reflect the value of an advertisement because:

[0238] Some ads drive conversions over long paths (such as for high-decision-cost products like home purchases and cars);

[0239] Even if some ads do not lead to conversions, the behavioral paths show a strong intention (such as long dwell time and frequent inquiries);

[0240] Some paths appear active, but their behavior has no logical order, which may be noise.

[0241] Therefore, it is necessary to build a system mechanism to track and model the complete behavioral path of users from ad exposure to final conversion, and extract its key structural features (path length, node type, response latency, etc.) to support the structural evaluation, attribution analysis and strategy optimization of advertising effectiveness.

[0242] In one specific implementation, the conversion chain tracking module is implemented by the following four sub-modules:

[0243] Submodule 1: Behavior sequence capture and node structuring module

[0244] Function: Starting from an ad click, record all key user behavior events and structure them as "path nodes".

[0245] Implementation steps:

[0246] Define a standardized set of path event types: E = {Click on an ad (e1), Enter the details page (e2), Swipe the page (e3), Add to favorites (e4), Customer service conversation (e5), Submit form (e6), Order completed (e7), ...};

[0247] The system starts listening to all user interaction events from the ad click time t0, recording the timestamp and event type:

[0248] P u ={(e1,t1),(e2,t2),…,(e n ,t n )}

[0249] Each path is sorted by time to form a user conversion path chain with time attributes.

[0250] Submodule 2: Transformation chain feature extraction module

[0251] Function: Extract multi-dimensional behavioral path features based on structured path data.

[0252] Extraction dimensions:

[0253] Path length: L = |P u |=n represents the total number of nodes in the path.

[0254] Node type distribution vector (one-hot or frequency): V e =[f1,f2,…,f k ]

[0255] Among them, f i Indicates event type e i The number of times it appears in the path or whether it appears at all.

[0256] Average response time (average time interval between actions):

[0257] Total response time (from ad to final conversion): T total =t n -t1

[0258] Behavioral sparsity index (time density): This represents the average number of actions a user takes per unit of time.

[0259] Sub-module 3: Conversion Path Classification and Tagging Module

[0260] Function: Classify paths into typical conversion types based on their behavioral patterns, facilitating subsequent explanation and grouping.

[0261] Classification logic:

[0262] Rapid decision-making path: short path length, fast response time, and includes high-value nodes;

[0263] Deep engagement path: long path, long dwell time, includes high-intent nodes such as consultation / registration;

[0264] Hesitant path: Repeated clicks without conversion, commonly seen among price-sensitive users;

[0265] Jump-out path: Only 1-2 nodes, exits quickly or becomes unresponsive;

[0266] Abnormal path: The behavior nodes have no logical order, such as clicking to pay immediately after the details page but not completing the transaction.

[0267] In practical implementation, K-means or HMM (Hidden Markov Model) can be used to cluster path behavior patterns and automatically identify behavior patterns.

[0268] Submodule 4: Conversion chain structure encoding and output module

[0269] Function: Encodes the extracted structural path features into a unified vector representation for use by subsequent modules.

[0270] Vector structure example:

[0271]

[0272] The first four dimensions represent path structure indicators, and the last k dimensions represent event distribution.

[0273] It can be used for modules such as advertising attribution model input, user segmentation, and strategy optimization;

[0274] Supports comparative analysis (multiple paths of the same user or different user paths of the same advertisement).

[0275] For example:

[0276] After a user clicks on the coffee machine ad, their behavioral path is as follows:

[0277] (e1,t1): Click the ad (0 seconds);

[0278] (e2,t2): Enter the details page (+1 second);

[0279] (e3,t3): Swipe to browse the page (+4 seconds);

[0280] (e4,t4): Add item to favorites (+12 seconds);

[0281] (e5,t5): Open the comments section (+15 seconds);

[0282] (e6,t6): Enter the customer service window and send "Is this model durable?" (+24 seconds);

[0283] (e7,t7): Exit the page (+40 seconds);

[0284] System extracts path features:

[0285] Path length L = 7;

[0286] Node type distribution: including ad clicks, details, favorites, comments, customer service, etc.;

[0287]

[0288] T total =40s

[0289] S = 7 / 40 ≈ 0.175

[0290] The path is classified as a deep engagement path because it contains high-intent behaviors (favorites + customer service) but has not yet been converted. The system can mark it as a high-potential user and push coupons or recall ads in the future.

[0291] The performance score generation module is used to integrate the cognitive intervention vector, multimodal behavioral consistency score and conversion chain features to calculate the comprehensive cognitive effect score of the advertising campaign for performance evaluation.

[0292] In advertising evaluation tasks, a single click, cognitive response, or conversion event often fails to fully reflect the true effectiveness of an ad. For example, an ad may trigger a strong cognitive shift but fail to elicit a behavioral response; another ad may be clicked rapidly but have no subsequent conversions.

[0293] To address this issue of biased assessment, it is necessary to construct a comprehensive scoring model that integrates information from different dimensions, including:

[0294] Cognitive intervention intensity: Does the advertisement truly affect the user's cognitive state?

[0295] Multimodal behavioral consistency: Whether the user has performed behavior consistent with the change in cognition;

[0296] Conversion path structure: Whether the user's behavioral chain from click to conversion is reasonable and of high intent.

[0297] Therefore, the performance scoring generation module is dedicated to integrating the core outputs of these sub-modules and generating a comprehensive cognitive effect score through model integration. This score serves as a unified quantitative basis for evaluating advertising performance and is used for selecting advertising materials, allocating advertising budgets, and ranking user responses.

[0298] In one specific implementation, the effect rating generation module includes the following three sub-modules:

[0299] Submodule 1: Feature Standardization Module

[0300] Function: To uniformly normalize the scores and vector features from different modules, making them comparable on the same numerical scale.

[0301] enter:

[0302] Intervention Vector I ad ∈R d ;

[0303] Behavioral consistency score S consistency ∈[0,1];

[0304] Transformation chain feature vector F chain ∈R k .

[0305] Processing steps:

[0306] Normalization of intervention vector magnitude:

[0307]

[0308] Normalization of each indicator in the transformation chain features:

[0309]

[0310] in:

[0311] u i It is the mean of the i-th dimension, σ i It is the standard deviation;

[0312] After standardization Submodule 2: Feature Weighted Fusion Module Function: Constructs the final scoring function using a linear weighting method, which can use static weights or training optimization. CES Calculation:

[0313] CES=α·S intervene +β·S consistency +γ·S chain

[0314] Among them, S intervene This indicates the score for the intensity of advertising cognitive intervention;

[0315] S consistency Indicates the multimodal consistency score;

[0316] S chain Indicates the score for the conversion path structure;

[0317] The weighting coefficients α, β, and γ satisfy α + β + γ = 1.

[0318] Submodule 3: Threshold Grading and Interpretation Module Function: Maps the score values ​​to different effect level labels for business use.

[0319] Example Level:

[0320] CES ≥ 0.85 → Extremely strong cognitive effect

[0321] 0.7≤CES<0.85→Valid

[0322] 0.5≤CES<0.7→General

[0323] CES < 0.5 → Weak effect

[0324] Furthermore, it can also output simultaneously:

[0325] Weighting and contribution of each indicator (explanation of source);

[0326] Weaknesses are marked (e.g., excessively sparse paths or insufficient consistency).

[0327] It can be used for recall strategies or campaign optimization suggestions.

[0328] For example:

[0329] Continuing with the previous example

[0330] Intervention vector:

[0331] I ad =[0.35,0.47,0.55,0.33],

[0332] Behavioral consistency score: S consistency =0.978

[0333] Transformation path feature F chain =[7,6.67,40,0.175], normalized S chain =0.81

[0334] Assume weight α = 0.3, β = 0.4, γ = 0.3

[0335] Then we have:

[0336] CES=0.3·0.85+0.4·0.978+0.3·0.81=0.895

[0337] System output:

[0338] CES score: 0.895;

[0339] Rating: Extremely strong cognitive effect;

[0340] Dominant factor: Consistency score has the highest weight, while path structure is a secondary factor.

[0341] Strategy Recommendations:

[0342] We recommend continuing to run this ad campaign among high-awareness, high-consistency demographics.

[0343] It is recommended to iteratively fine-tune the emotional arousal dimension based on the current materials.

[0344] In another embodiment, the present invention also provides a data analysis-based method for evaluating and analyzing the effectiveness of advertising campaigns, using the foregoing embodiments to perform the evaluation and analysis of advertising campaign effectiveness.

[0345] It should be noted that the explanations and descriptions of the aforementioned advertising placement effect evaluation and analysis system embodiments based on data analysis also apply to the methods of the embodiments of this application, and will not be repeated here.

[0346] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0347] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0348] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0349] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. For some module structures not specifically defined in this invention, the content described in the prior art shall prevail. The prior art mentioned in the foregoing background and specific embodiments sections can be considered as part of this invention and used to understand the meaning of some technical features or parameters.

Claims

1. A data-driven advertising effectiveness evaluation and analysis system, characterized in that, The system includes the following modules: The user cognitive state acquisition module is used to construct a user cognitive state vector based on multi-source behavioral data before and after user interaction with the advertisement; the user cognitive state acquisition module includes the following sub-modules: The multi-source data collection submodule is used to collect multi-source behavioral data of users before and after contacting the advertisement; The multimodal feature extraction submodule is used to convert the collected multimodal data into feature vectors; The cognitive state vector construction submodule is used to fuse multimodal features to construct the user's cognitive state vector; The advertising intervention modeling module is used to model the changes in user cognitive state vectors caused by advertising, generating a cognitive intervention vector representing the change in a user's cognitive state after exposure to an advertisement. The advertising intervention modeling module includes the following sub-modules: an original state change extraction sub-module, used to calculate the original state change based on the cognitive state vector; an intervention attribution modeling sub-module, used to model the state change with advertising and user features to generate an intervention vector; and an intervention vector regularization and label annotation sub-module, used to perform regularization processing on the intervention vector and add semantic labels. The multimodal consistency analysis module is used to evaluate the consistency between the user's behavioral response and the direction of the cognitive intervention vector based on the user's language feedback, visual behavior, and emotional trajectory, and output a multimodal behavioral consistency score. The multimodal consistency analysis module includes the following sub-modules: a multimodal behavioral feature extraction sub-module, used to extract behavioral features from language, visual, and emotional data; a behavioral vector unified projection sub-module, used to uniformly project the multimodal features onto the behavioral response vector; and a consistency calculation sub-module, used to calculate the consistency score between the behavioral response vector and the cognitive intervention vector. The conversion chain tracking module is used to track the user's behavioral path from ad click to final conversion, forming conversion chain features and identifying path length, path node type, and path response latency. The conversion chain tracking module includes the following sub-modules: a behavior sequence capture and node structuring sub-module, used to record and structure the user's behavior node sequence; a conversion chain feature extraction sub-module, used to extract path length, node type, and response latency features from the behavior sequence; a conversion path classification and tagging sub-module, used to classify conversion paths and label them; and a conversion chain structure encoding and output sub-module, used to output a structured conversion path feature vector. The effect score generation module is used to integrate the cognitive intervention vector, multimodal behavioral consistency score, and conversion chain features to calculate the comprehensive cognitive effect score of the advertising campaign for evaluation of campaign effectiveness. The effect score generation module includes the following sub-modules: a feature standardization module, used to normalize the intervention intensity, consistency score, and conversion features; a feature weighted fusion module, used to weight and combine the normalized features to generate a comprehensive score; and a threshold classification and interpretation module, used to classify the comprehensive score and output the score results and influencing factors.

2. The advertising effectiveness evaluation and analysis system based on data analysis according to claim 1, characterized in that, The multimodal feature extraction submodule is used to extract features from webpage behavior data, language feedback data and visual emotion data respectively, and convert the extracted features into a high-dimensional embedding representation in the same vector space for use by the cognitive state vector construction submodule.

3. The advertising effectiveness evaluation and analysis system based on data analysis according to claim 1, characterized in that, The intervention vector regularization and label annotation submodule is used to perform scale unification processing on the advertising intervention vectors and bind corresponding cognitive semantic labels to each vector dimension. The labels include attention activation, emotional resonance, interest enhancement and intent enhancement, which are used to enhance the interpretability and traceability of the intervention vectors.

4. The advertising effectiveness evaluation and analysis system based on data analysis according to claim 1, characterized in that, The conversion path classification and tagging submodule is used to classify user paths into fast response paths, deep engagement paths, hesitant paths, and bounce paths based on the type of behavioral nodes, node order, response time distribution, and path density in the conversion path, and to assign tags to each type of path to indicate its behavioral intent level, in order to evaluate the structural performance of the advertising conversion chain.

5. A data-driven method for evaluating the effectiveness of advertising campaigns, characterized in that, The method uses the data analysis-based advertising performance evaluation and analysis system as described in any one of claims 1-4 to perform advertising performance evaluation and analysis.

Images