E-commerce advertisement putting strategy optimization method and system based on multi-source perception

Abstract

The invention provides an e-commerce advertisement putting strategy optimization method and system based on multi-source perception. The method comprises the following steps: firstly, extracting a user interest vector representing user interests and preferences and a scene vector representing current interaction scene semantics and a user instant state; a preference association evaluation quantity is generated based on the user interest vector and the candidate advertisement feature vector; under the guidance of an advertisement putting strategy template, on the basis of the scene vector and the candidate advertisement feature vector, generating a scene adaptation evaluation quantity representing the adaptation degree of the candidate advertisement to the current interaction scene; and performing fusion evaluation on the current advertisement putting decision through the preference association evaluation quantity and the scene adaptation evaluation quantity to obtain a strategy decision score of advertisement putting, and adjusting and executing an advertisement putting strategy for the target user in the current interaction scene according to the strategy decision score. By adopting the scheme of the invention, the robustness of the advertisement putting system can be improved based on the dynamic decision of collaborative analysis of the user interest preference and the real-time interaction scene.

Description

Technical Field

[0001] This application relates to the field of internet advertising technology, and more specifically, to a method and system for optimizing e-commerce advertising placement strategies based on multi-source perception. Background Technology

[0002] As internet advertising services evolve from extensive display to precise targeting, effectively reaching target users has become a core challenge. Current mainstream methods mainly rely on building interest profiles based on users' historical behavior or performing simple matching based on real-time context. While these methods improve ad relevance to some extent, they often have limitations such as rigid adaptation and large fluctuations in decision-making when dealing with the dynamic evolution of user interests and complex and ever-changing interaction scenarios, making it difficult to maintain stable advertising performance in changing environments.

[0003] Existing advertising strategies face significant bottlenecks in improving robustness. The root cause lies in the system's failure to achieve deep collaborative analysis and dynamic decision-making between long-term user interests and real-time interaction scenarios. Specifically, existing solutions either rely heavily on static historical user interest models, failing to adequately respond to users' immediate contextual intentions. This leads to a sharp drop in performance when user interests temporarily shift or scenarios change abruptly. Alternatively, they overemphasize real-time context matching, neglecting stable user interest baselines. This makes advertising decisions too susceptible to random interaction noise, resulting in unstable performance. Both of these fragmented analysis models lead to insufficient generalization and adaptability when facing real-world scenarios such as data sparsity, interest drift, and scene switching, resulting in poor overall decision robustness. Therefore, how to improve the robustness of advertising systems through dynamic decision-making based on collaborative analysis of user interests and real-time interaction scenarios has become a challenging problem for the industry. Summary of the Invention

[0004] This application provides a method and system for optimizing e-commerce advertising delivery strategies based on multi-source perception. It can make dynamic decisions based on collaborative analysis of user interests and preferences and real-time interaction scenarios, thereby improving the robustness of the advertising delivery system.

[0005] In a first aspect, this application provides a method for optimizing e-commerce advertising placement strategies based on multi-source perception, comprising the following steps: Acquire long-term behavioral data of target users and generate user interest vectors that represent their interests and preferences; Collect multi-dimensional contextual information of the current interaction scenario to generate a scenario vector that represents the semantics of the current interaction scenario and the real-time state of the target user; Based on the scenario vector, the optimal associated historical scenario is matched from the pre-built historical scenario strategy library, and the advertising placement strategy template corresponding to the optimal associated historical scenario is obtained. For candidate ads, a preference association evaluation metric is generated based on the user interest vector and the candidate ad feature vector to characterize the degree of association between the candidate ad and the target user's long-term interests. Guided by the advertising delivery strategy template, a scenario adaptation evaluation metric is generated based on the scenario vector and the candidate ad feature vector to characterize the degree of adaptation of the candidate ad to the current interactive scenario. The advertising placement decision in the current interaction scenario is evaluated by combining the preference association evaluation and the scenario adaptation evaluation to obtain the advertising placement strategy decision score. Then, the advertising placement strategy for the target user in the current interaction scenario is adjusted and executed according to the strategy decision score.

[0006] Preferably, acquiring long-term behavioral data of the target user and generating a user interest vector representing their interests and preferences specifically includes: Collect target users' advertising interaction behavior logs within a preset historical period. The advertising interaction behavior logs include at least ad clicks, ad viewing duration, and records of product collections or purchases associated with the ad content. The advertising interaction behavior logs are cleaned and normalized to extract interaction frequency and depth features corresponding to different advertising categories or themes; The interaction frequency and depth features are input into a pre-trained user interest model, which learns users’ implicit preferences for advertising categories or topics based on behavior logs. The user interest model outputs a multi-dimensional vector, which serves as the user interest vector.

[0007] Preferably, collecting multi-dimensional contextual information of the current interaction scenario to generate a scenario vector representing the semantics of the current interaction scenario and the real-time state of the target user specifically includes: Real-time acquisition of environmental status information and current user interaction content; The environmental state information and the current interaction content are respectively extracted and vectorized to obtain the environmental state vector and the user real-time interaction vector. By fusing the environmental state vector and the user's real-time interaction vector, a scene vector representing the semantics of the current interaction scene and the real-time state of the target user is generated.

[0008] Preferably, the process of matching the optimal associated historical scenario from a pre-built historical scenario strategy library based on the scenario vector and obtaining the advertising strategy template corresponding to the optimal associated historical scenario specifically includes: Calculate the similarity between the current scene vector and each historical scene vector in the historical scene strategy library; The historical scene with the highest similarity is selected as the optimal associated historical scene; Read the set of strategy parameters corresponding to the historical scenario and use it as the template for the advertising delivery strategy.

[0009] Preferably, the generation of a preference association evaluation metric, based on the user interest vector and the candidate ad feature vector, to characterize the degree of long-term interest association between the candidate ad and the target user specifically includes: Determine the similarity between the user interest vector and the candidate ad feature vector; The similarity is corrected based on the time decay attribute of user interests; The corrected value is used as a preference association evaluation metric to characterize the degree of association between candidate ads and the long-term interests of target users.

[0010] Preferably, correcting the similarity based on the time decay attribute of user interests specifically includes: assigning decay weights to interest features in different time windows in the user interest vector, and weighting the similarity calculation results based on the decay weights.

[0011] Preferably, the strategy decision score for ad placement in the current interaction scenario is obtained by fusing the preference-related evaluation score and the scenario-adaptation evaluation score: Determine the initial membership degree of the preference association evaluation value and the scenario adaptation evaluation value corresponding to the ad recommendation level, respectively; Based on the collaborative rules defined in the advertising strategy template, a two-dimensional evaluation relationship matrix is ​​constructed to describe the mutual influence between preference-related evaluation quantities and scenario-adaptation evaluation quantities. The numerical values ​​in the two-dimensional evaluation relation matrix are mapped to the corresponding fuzzy sets by using fuzzy membership functions; Based on the initial membership degree and the fuzzy set, a fusion decision fuzzy evaluation matrix is ​​constructed; Based on the fuzzy evaluation matrix of the fusion decision, the advertising placement decision in the current interaction scenario is evaluated by fuzzy comprehensive evaluation, and the strategy decision score of the advertising placement is output after defuzzification.

[0012] Secondly, this application provides an e-commerce advertising placement strategy optimization system based on multi-source perception, comprising: The acquisition module is used to acquire long-term behavioral data of target users and generate user interest vectors that represent their interests and preferences. The acquisition module is also used to collect multi-dimensional context information of the current interaction scenario and generate a scenario vector that represents the semantics of the current interaction scenario and the real-time state of the target user. The processing module is used to match the optimal associated historical scene from the pre-built historical scene strategy library according to the scene vector, and obtain the advertising placement strategy template corresponding to the optimal associated historical scene. The processing module is further configured to generate a preference association evaluation metric, based on the user interest vector and the candidate ad feature vector, a metric representing the degree of long-term interest association between the candidate ad and the target user. The processing module is also used to generate a scene adaptation evaluation quantity, which characterizes the degree of adaptation of the candidate advertisement to the current interactive scene, based on the scene vector and the candidate advertisement feature vector, under the guidance of the advertisement delivery strategy template. The execution module is used to perform a fusion evaluation of the advertising placement decision in the current interaction scenario through the preference association evaluation quantity and the scenario adaptation evaluation quantity, to obtain the advertising placement strategy decision score, and then adjust and execute the advertising placement strategy for the target user in the current interaction scenario based on the strategy decision score.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described method for optimizing e-commerce advertising placement strategies based on multi-source perception.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for optimizing e-commerce advertising placement strategies based on multi-source perception.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In this embodiment, the advertising placement decision in the current interaction scenario is evaluated by combining the preference association evaluation metric and the scenario adaptation evaluation metric to obtain an advertising placement strategy decision score. Then, the advertising placement strategy for the target user in the current interaction scenario is adjusted and executed based on the strategy decision score. First, based on the user interest vector and candidate ad feature vector, a preference association evaluation metric is generated to characterize the long-term interest association between the candidate ad and the target user. This preference association evaluation metric provides a stable interest baseline for user decision-making, effectively combating interest drift caused by short-term data sparsity or random interaction noise, thus laying the foundation for system robustness. Second, guided by the advertising placement strategy template, a scenario adaptation evaluation metric is generated based on the scenario vector and candidate ad feature vector to characterize the adaptability of the candidate ad to the current interaction scenario. This approach enables dynamic contextual adaptation of strategies, preventing static models from failing under sudden scene changes and enhancing the system's real-time response and adaptability to complex and ever-changing environments. Then, by fusing the preference-related evaluation and the scene-adaptation evaluation, a strategy decision score for advertising in the current interaction scenario is obtained. This fusion evaluation uses fuzzy mathematics to handle the nonlinear relationship and decision uncertainty between the two evaluation signals, achieving a smooth and disturbance-resistant comprehensive assessment, fundamentally suppressing abnormal fluctuations in decision output. Finally, the advertising strategy for the target user in the current interaction scenario is adjusted and executed based on the strategy decision score, enabling precise and personalized reach that maintains stable performance in changing environments. In summary, this application's solution improves the robustness of the advertising system through dynamic decision-making based on collaborative analysis of user interests and real-time interaction scenarios. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating an application scenario of an e-commerce advertising placement strategy optimization method based on multi-source perception, as shown in some embodiments of this application. Figure 2 This is an exemplary flowchart of an e-commerce advertising placement strategy optimization method based on multi-source perception, as shown in some embodiments of this application. Figure 3 This is a flowchart illustrating the process of determining preference-related evaluation metrics according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of an e-commerce advertising placement strategy optimization system based on multi-source perception, according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device that implements a method for optimizing e-commerce advertising placement strategies based on multi-source perception, according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] refer to Figure 1 This figure is a schematic diagram of an application scenario for an e-commerce advertising placement strategy optimization method based on multi-source perception, according to some embodiments of this application. The figure includes a data acquisition device, a server, a communication network, and a terminal. The data acquisition device is connected to the server via the network, and the terminal is connected to the server system via the communication network. The server acquires user behavior data and real-time context information provided by the data acquisition device, cleans and standardizes the data, and then inputs it into a pre-built strategy optimization model. The model sequentially performs user long-term interest vector extraction, interaction scenario vector generation, historical scenario strategy matching, preference association and scenario adaptation dual-path evaluation, and strategy fusion decision-making, ultimately generating a personalized advertising placement strategy for the current scenario and user. When the server receives a strategy optimization request sent through the terminal for a specific user group or advertising campaign, it feeds back the recommended strategy, expected effect indicators, and strategy confidence to the terminal for advertising optimizers, marketing decision-makers, or system operation administrators to view and implement adjustments.

[0019] The data acquisition devices may include user behavior log systems, real-time interactive sensing interfaces, third-party data platform access devices, and environmental context acquisition services; the terminals may be, but are not limited to, advertising management back-end systems, marketing data analysis platforms, operation monitoring screens, or mobile decision support terminals; the servers may be strategy computing clusters deployed locally on the enterprise, or distributed advertising decision and optimization service platforms built on public or hybrid clouds.

[0020] refer to Figure 2 The figure is an exemplary flowchart of an e-commerce advertising placement strategy optimization method based on multi-source perception, according to some embodiments of this application. The e-commerce advertising placement strategy optimization method based on multi-source perception mainly includes the following steps: In step 101, long-term behavioral data of the target user is obtained to generate a user interest vector representing their interests and preferences.

[0021] In some embodiments, obtaining long-term behavioral data of target users and generating user interest vectors representing their interests and preferences can be achieved through the following steps: Collect target users' advertising interaction behavior logs within a preset historical period. The advertising interaction behavior logs include at least ad clicks, ad viewing duration, and records of product collections or purchases associated with the ad content. The advertising interaction behavior logs are cleaned and normalized to extract interaction frequency and depth features corresponding to different advertising categories or themes; The interaction frequency and depth features are input into a pre-trained user interest model, which learns users’ implicit preferences for advertising categories or topics based on behavior logs. The user interest model outputs a multi-dimensional vector, which serves as the user interest vector.

[0022] It should be noted that the interaction frequency in this application is an indicator that measures the total number of interactions performed by the target user on a specific advertising category or theme within a preset historical period; the depth feature is an indicator that measures the quality and depth of the target user's interaction with a specific advertising category or theme; and the user interest vector is a feature vector that reflects the strength of the target user's long-term implicit preference across advertising categories or themes.

[0023] In specific implementation, firstly, a tracking system deployed at the front end of the ad display can capture all ad-related interaction events of the target user within the e-commerce platform or application in real time. This includes, but is not limited to, clicks on ad placements, dwell time on ad landing pages, and records of product collection or purchase behavior driven by ads. These raw event data are collected and temporarily stored according to a unified log format, forming a structured record set containing a unique user identifier, event timestamp, event type, associated ad identifier, and associated product identifier—the ad interaction behavior log. Secondly, the acquired ad interaction behavior log undergoes data preprocessing, including removing invalid records, unifying the identifier format, and classifying and labeling each log record according to a preset ad category system or topic tag system. Furthermore, for each ad category or topic, the total number of interactions by the target user within a preset historical period is counted to calculate the interaction frequency characteristics, and the average browsing time, collection rate, and purchase rate are comprehensively calculated. The conversion rate is used to construct the interaction depth feature. Then, the maximum and minimum values ​​of all statistically obtained feature values ​​are normalized to obtain standardized interaction frequency and depth features. Then, the extracted and standardized interaction frequency and depth features are used as input feature vectors and input into a pre-trained user interest model. This user interest model adopts a deep neural network architecture. It has learned the complex mapping relationship from behavioral features to users' implicit interest preferences by supervising or self-supervising training on a large number of users' historical advertising interaction behavior logs and their corresponding features. After receiving the input feature vector, the model performs calculations through its internal multi-layer nonlinear transformations. Finally, after the user interest model completes the forward propagation calculation of the input feature vector, it obtains a fixed-dimensional real number array from its output layer. Each dimension of this array corresponds to a predefined advertising interest dimension or topic, and its value represents the target user's preference strength for that dimension. This real number array is used as the final output user interest vector.

[0024] Preferably, in some embodiments, the user interest model can be pre-trained in the following manner: a training sample set is constructed using a large number of users' historical advertising interaction behavior logs and their corresponding interaction frequency and depth features, wherein each sample uses a pair of user behavior feature vectors and their actual interest tendency labels shown in a subsequent period as input and expected output supervision signals; the model adopts a multilayer perceptron structure, with its input layer dimension matching the behavioral feature dimension, its output layer dimension matching the preset number of interest categories, and one to three hidden layers in the middle, and the number of neurons in each layer can be configured exemplarily between 128 and 512 according to the task complexity; during training, stochastic gradient descent or its variant is used as the optimizer, the learning rate can be set to an example value between 0.001 and 0.01, the batch size can be set to 256 or 512, and iterative training is performed until the loss function converges, thereby learning the mapping relationship from the user's historical behavior features to their stable interest preferences, completing the pre-training of the model for subsequent generation of user interest vectors.

[0025] In step 102, multi-dimensional contextual information of the current interaction scenario is collected to generate a scenario vector that represents the semantics of the current interaction scenario and the real-time state of the target user.

[0026] It should be noted that the multi-dimensional contextual information in this application refers to the collection of environmental state data and real-time user behavior data that are closely related to the user's interaction behavior at the current moment.

[0027] In some embodiments, collecting multi-dimensional contextual information of the current interaction scenario and generating a scenario vector representing the semantics of the current interaction scenario and the real-time state of the target user can be achieved through the following steps: Real-time acquisition of environmental status information and current user interaction content; The environmental state information and the current interaction content are respectively extracted and vectorized to obtain the environmental state vector and the user real-time interaction vector. By fusing the environmental state vector and the user's real-time interaction vector, a scene vector representing the semantics of the current interaction scene and the real-time state of the target user is generated.

[0028] It should be noted that the scene vector in this application is a feature vector that reflects the overall semantics of the current interactive environment and the user's immediate behavioral intent.

[0029] In specific implementation, firstly, by calling the system interface and application interface of the terminal device, a set of data reflecting the objective environment is acquired and encapsulated in real time. This includes a system timestamp accurate to the second, geographical coordinates obtained through GPS or network positioning, the terminal device model and screen size, and the current network connection type and signal strength. Simultaneously, by parsing the document object model and network requests of the currently active application window or browser tab, the Uniform Resource Locator (URL) of the page the user is focusing on, the page title, core text content, and the currently entered search query are captured. These two categories of data are combined as environmental state information and the user's current interaction content. Secondly, the environmental state information acquired in the first step is processed using a combination of one-hot encoding and scalarization. For example, discrete information such as device model and network type is converted into one-hot encoding, timestamps are converted into periodic values ​​within a day, and geographical locations are converted into latitude and longitude scalars. All encoded values ​​are then concatenated into a fixed-length array as the environmental state vector. Finally, the acquired user's current interaction content is used to extract semantic features from the core text of the page and the search terms using a pre-trained natural language processing model (such as a text embedding model). The process involves taking the generated environment state vector and converting it into a fixed-dimensional array of real numbers, which serves as the user's real-time interaction vector. Then, the generated environment state vector and the user's real-time interaction vector are concatenated end-to-end to form a higher-dimensional concatenated vector. This concatenated vector is then input into a pre-trained fully connected neural network layer for non-linear fusion and dimensionality reduction. This neural network layer uses learned weight parameters to weight and integrate the two types of information and compress features, ultimately outputting a fixed-dimensional array of real numbers as a scene vector representing the semantics of the current interaction scenario and the target user's real-time state. It should be further noted that... The pre-trained fully connected neural network layer is used here for two main reasons: first, to achieve efficient information fusion and abstraction; and second, to optimize the performance of subsequent calculations. Directly concatenated vectors only mechanically merge two types of information without capturing their inherent relationship. This neural network layer, through learned weights, can perform non-linear interaction and weighted integration of the two types of vectors, thereby extracting more representative high-order combined features. At the same time, this layer plays a dimensionality reduction role, compressing the high-dimensional concatenated vectors into a unified and compact low-dimensional representation, which significantly reduces the storage and computing power overhead in the subsequent matching and calculation process.

[0030] In step 103, based on the scene vector, the optimal associated historical scene is matched from the pre-built historical scene strategy library, and the advertising placement strategy template corresponding to the optimal associated historical scene is obtained.

[0031] It should be noted that the historical scene strategy library in this application refers to a database that stores aggregated and optimized historical interaction scene representations and their corresponding advertising strategies, supporting efficient similarity retrieval. Its construction process is as follows: First, a large amount of contextual information about historical interaction scenes is collected, and a corresponding historical scene vector is generated for each historical interaction scene using the aforementioned scene vector generation method. Second, unsupervised clustering is performed on all historical scene vectors using a clustering algorithm (e.g., K-means), aggregating semantically similar scenes into multiple categories, with the center vector of each category representing a typical scene. Then, for each category, using historical advertising exposure, click, and conversion data within that category, a set of strategy parameters that maximize advertising performance metrics (such as click-through rate or conversion rate) is searched offline and determined using reinforcement learning or Bayesian optimization methods. Finally, the center vector of each category is bound to its optimized strategy parameters, stored as a complete record in the database, collectively forming the historical scene strategy library for online real-time retrieval and retrieval.

[0032] In some embodiments, matching the optimal associated historical scenario from a pre-built historical scenario strategy library based on the scenario vector and obtaining the advertising strategy template corresponding to the optimal associated historical scenario can be achieved through the following steps: Calculate the similarity between the current scene vector and each historical scene vector in the historical scene strategy library; The historical scene with the highest similarity is selected as the optimal associated historical scene; Read the set of strategy parameters corresponding to the historical scenario and use it as the template for the advertising delivery strategy.

[0033] In specific implementation, firstly, cosine similarity, a technology already in use, can be employed to describe the similarity between the current scene vector and each historical scene vector in the historical scene strategy library. Then, all calculated similarity values ​​are iterated and compared to find the largest similarity value. The historical scene vector corresponding to this largest similarity value and all its associated identifiers in the library are located and obtained. This historical scene vector and its associated information are determined as the optimal associated historical scene for guiding the current advertising strategy. Next, based on the unique identifier of the optimal associated historical scene, an index query is performed in the historical scene strategy library. A set of strategy parameters, pre-optimized offline and bound to this identifier, is retrieved from the library. This parameter set includes configurable items such as scene adaptation weights and fusion rule coefficients. This retrieved set of strategy parameters serves as the advertising strategy template for the current real-time advertising decision. The advertising strategy template refers to a set of adjustable strategy parameters obtained from historical scene matching, used to guide the generation of scene adaptation evaluation metrics in real time.

[0034] In step 104, for candidate advertisements, a preference association evaluation metric is generated based on the user interest vector and the candidate advertisement feature vector to characterize the degree of association between the candidate advertisement and the target user's long-term interests.

[0035] In some embodiments, reference Figure 3 As shown in the figure, this is a flowchart illustrating the process of determining the preference association evaluation metric in some embodiments of this application. In this embodiment, the preference association evaluation metric, which characterizes the degree of long-term interest association between the candidate advertisement and the target user, is generated based on the user interest vector and the candidate advertisement feature vector. This can be achieved through the following steps: In step 1041, the similarity between the user interest vector and the candidate advertisement feature vector is determined; In step 1042, the similarity is corrected according to the time decay attribute of user interests; In step 1043, the corrected value is used as a preference association evaluation metric to characterize the degree of association between the candidate advertisement and the target user's long-term interests.

[0036] In specific implementation, firstly, the generated user interest vector is compared with the advertising feature vector extracted from the candidate advertising content. The similarity is determined by calculating the cosine of the angle between the two vectors. This calculation method involves dividing the inner product of the two vectors by the product of their respective magnitudes, and finally using the calculated cosine value as the similarity between the two vectors. Using cosine similarity primarily aims to eliminate the influence of vector length, thus focusing on measuring the degree of matching between user interests and advertising content in thematic direction. This reflects the essential correlation better than simple inner product calculation. Then, corresponding attenuation weight coefficients are configured for the interest feature components in the user interest vector originating from different historical time windows. These coefficients decrease with the distance from the present time window. The configured attenuation weights are then used to determine the similarity. The calculated similarity is weighted by a weighting factor, and the resulting weighted value is used as the corrected value. This is because user interests evolve or fade over time, and recent behavior better reflects current effective preferences. The introduction of a time decay attribute dynamically corrects the contribution of historical interests, making the evaluation more timely and accurate. Finally, the value obtained after time decay correction is directly mapped to a scalar within a preset scoring range (e.g., between 0 and 1). This scalar value is then used as a preference association evaluation metric to characterize the degree of long-term interest association between candidate advertisements and target users. The preference association evaluation metric is a quantitative indicator that measures the strength of the matching association between candidate advertisements and target users' long-term stable interests after time correction.

[0037] In step 105, under the guidance of the advertising delivery strategy template, a scene adaptation evaluation metric is generated based on the scene vector and the candidate advertising feature vector to characterize the degree of adaptation of the candidate advertising to the current interactive scene.

[0038] In some embodiments, under the guidance of the advertising delivery strategy template, generating a scene adaptation evaluation metric to characterize the degree of adaptation of candidate ads to the current interactive scene based on the scene vector and candidate ad feature vector can be achieved through the following steps: Obtain the scenario adaptation weight parameters from the ad delivery strategy template; The scene vector and the candidate advertisement feature vector are interactively calculated; The interaction calculation results are weighted and scored using the scene adaptation weight parameters, and a scene adaptation evaluation score is output to characterize the degree of adaptation of the candidate advertisement to the current interaction scene.

[0039] It should be noted that the scenario fit evaluation score is a quantitative indicator that measures the degree of matching between the candidate ad and the current specific interaction scenario and the user's immediate intent.

[0040] It should also be noted that traditional advertising scoring models typically use fixed weight parameters, which cannot adapt to the dynamic changes in user intent and attention under different interaction scenarios, resulting in inaccurate scenario fit assessment. This method dynamically obtains scenario fit weight parameters optimized from historical data from a historical scenario strategy library, and applies these parameters to the interaction calculation results of scenario vectors and advertising feature vectors for weighted scoring, thereby achieving scenario adaptation in the evaluation process. Its beneficial effects are mainly reflected in two aspects: First, it significantly improves the accuracy of scenario fit evaluation because the weight parameters carry successful experience from similar historical scenarios, and can more reasonably quantify the importance of different feature dimensions in the current scenario; second, it enhances the overall adaptive capability and effect stability of the system, enabling advertising strategies to be adjusted in a refined and differentiated manner based on the semantic differences of the scenario, ultimately effectively improving ad click-through rates and conversion rates.

[0041] In practice, firstly, based on the optimal historical scenario identifier obtained from the current matching, a set of numerical weight parameters pre-optimized for this type of scenario is read from the corresponding ad placement strategy template record. This set of weight parameters is a real number array whose dimension matches the subsequent interaction calculation results. This read real number array is used as the scenario adaptation weight parameters to guide the current real-time scoring. Obtaining weight parameters from the matched historical strategy template, rather than using fixed parameters, is to enable the scoring mechanism to adapt to the characteristics of different scenario types. Historical data shows that the importance of each dimension to the final adaptation varies after the interaction between scenario information and ad features in different scenarios. The weights in the template are a quantification of this historical experience, which can improve the accuracy of the scoring. Secondly, the scenario vector with the same dimension is multiplied element-wise with the candidate ad feature vector, i.e., the Hadamard product of the two vectors is calculated. This operation produces a new... The vector is a product of the elements of the two original vectors at corresponding positions. This new vector is used as the result of the interaction calculation. Then, the resulting vector is multiplied element-wise with the scene adaptation weight parameters, and all products are summed to obtain a weighted sum. This weighted sum is then input into a Sigmoid function for normalization mapping, constraining its value to a fixed scoring range. This final output scalar value is used as the scene adaptation evaluation metric. Using weights obtained from the template to perform a weighted summation of the interaction results is to highlight the most important matching dimensions in the current scene type based on historical experience. Normalization using the Sigmoid function is to standardize the scoring results, ensuring that they are within the same numerical range as the preference-related evaluation metric, such as between 0 and 1. This ensures that the two can be fairly and effectively compared and calculated in subsequent fusion evaluation steps.

[0042] In step 106, the advertising placement decision in the current interaction scenario is evaluated by combining the preference association evaluation value and the scenario adaptation evaluation value to obtain the advertising placement strategy decision score. Then, the advertising placement strategy for the target user in the current interaction scenario is adjusted and executed according to the strategy decision score.

[0043] It should be noted that this solution introduces a fuzzy evaluation mechanism, primarily based on technical considerations of the inherent uncertainty and nonlinear relationships in the advertising decision-making process. Traditional fusion methods based on precise numerical values ​​(such as weighted summation or linear models) struggle to effectively handle semantic concepts like "preference" and "scenario adaptation," which are inherently continuous and progressive. They also cannot adequately address noise in real-time data or sudden changes in decision-making at evaluation boundaries. This solution constructs a fuzzy evaluation system, mapping precise input quantities to fuzzy linguistic variables via membership functions. It then utilizes a fuzzy rule base defined based on historical strategy knowledge for approximate reasoning, finally outputting a decision score through defuzzification. The beneficial effects of this technical approach are: significantly improved system decision robustness, greater tolerance to input data noise, and ensured smooth and stable score output; enhanced system scenario adaptability by flexibly capturing and modeling the complex, nonlinear dynamic synergistic relationship between long-term interests and immediate scenarios through fuzzy rules; and ultimately, a more scenario-appropriate comprehensive evaluation at the technical level, providing a more reliable and interpretable decision-making basis for improving key indicators such as ad click-through rate and conversion rate at the business level.

[0044] In some embodiments, the strategy decision score for advertising placement in the current interaction scenario is obtained by fusing the preference association evaluation score and the scenario adaptation evaluation score. This can be achieved through the following steps: Determine the initial membership degree of the preference association evaluation value and the scenario adaptation evaluation value corresponding to the ad recommendation level, respectively; Based on the collaborative rules defined in the advertising strategy template, a two-dimensional evaluation relationship matrix is ​​constructed to describe the mutual influence between preference-related evaluation quantities and scenario-adaptation evaluation quantities. The numerical values ​​in the two-dimensional evaluation relation matrix are mapped to the corresponding fuzzy sets by using fuzzy membership functions; Based on the initial membership degree and the fuzzy set, a fusion decision fuzzy evaluation matrix is ​​constructed; Based on the fuzzy evaluation matrix of the fusion decision, the advertising placement decision in the current interaction scenario is evaluated by fuzzy comprehensive evaluation, and the strategy decision score of the advertising placement is output after defuzzification.

[0045] It should be noted that the initial membership degree in this application is a numerical distribution representing the probability that the preference-related evaluation quantity or the scenario-adaptation evaluation quantity belongs to multiple preset advertising recommendation levels; the two-dimensional evaluation relationship matrix refers to a numerical square matrix used to quantify the strength of the dynamic mutual influence relationship between the preference-related evaluation quantity and the scenario-adaptation evaluation quantity; the fusion decision fuzzy evaluation matrix is ​​a composite fuzzy matrix used to integrate the initial membership degree information and the fuzzy influence relationship between the evaluation quantity and to provide complete input information for the final comprehensive evaluation; the strategy decision score is the final decision basis for quantifying the comprehensive placement priority of candidate advertisements.

[0046] In specific implementation, firstly, a set of fuzzy membership functions are defined for the preference-related evaluation quantity and the scenario-adaptation evaluation quantity, with each function corresponding to an ad recommendation level; the precise value of each evaluation quantity is input into its corresponding membership function for calculation to obtain the membership value of the evaluation quantity to each recommendation level, and the two sets of membership values ​​calculated for the two evaluation quantities are used as their initial membership degrees; secondly, a predefined collaboration rule is read from the currently used ad delivery strategy template, which clearly describes whether the two evaluation quantities mutually reinforce, weaken, or are independent when they are in different level combinations; according to this rule, a fuzzy membership function is used to determine the membership degree of each evaluation quantity. The influence of each level of the price quantity on each level of another evaluation quantity is assigned, filling a matrix whose row and column dimensions are equal to the number of recommendation levels. The completed numerical matrix serves as the two-dimensional evaluation relationship matrix. It should be further noted that traditional fusion assumes the two evaluation quantities are independent; this relationship matrix is ​​constructed to model the non-independent, dynamic synergistic or antagonistic effects between them, such as "high scene adaptability may enhance the value of moderate interest relevance," thus making the fusion model more consistent with complex decision-making logic. Then, a set of fuzzy linguistic variables and their corresponding membership functions are defined to describe the strength of the relationship. Each value in the two-dimensional evaluation relationship matrix is ​​then assigned a value. Each numerical value is input into a membership function for calculation, resulting in a membership vector for each value belonging to a strength level of influence relationship. The set of membership vectors corresponding to all values ​​is collectively referred to as the fuzzy set obtained after mapping. While the original numerical values ​​in the relationship matrix are precise, "strength of influence" itself is a fuzzy concept. This step fuzzifies the precise numerical relationships to enable subsequent rule reasoning to be performed at a unified, more tolerant fuzzy semantic level, thus improving the system's robustness. Furthermore, the two initial membership vectors are synthesized with the obtained fuzzy set representing the influence relationship. The synthesis rule is based on the principle of fuzzy relation synthesis. For example... For example, a specific operation, such as the maximum-minimum synthesis method, is performed on the initial membership vector of an evaluation metric and the fuzzy submatrix of the influence relationship involving that evaluation metric to obtain an intermediate matrix. The intermediate matrix, which integrates all the information, is then used to construct the fusion decision fuzzy evaluation matrix. Finally, the fuzzy comprehensive evaluation algorithm (such as weighted average evaluation) is applied to the constructed fusion decision fuzzy evaluation matrix to obtain a fuzzy output set about the final advertising recommendation degree. This fuzzy output set is then defuzzified using a method such as the centroid method to calculate its clarity value. This calculated clarity value is then normalized to a preset scoring range and used as the final strategy decision score output.

[0047] In other embodiments, the strategy decision score for advertising placement in the current interaction scenario is obtained by fusing the preference association evaluation and the scenario adaptation evaluation to perform a fusion evaluation. This can also be achieved in the following ways: fuzzification: the preference association evaluation and the scenario adaptation evaluation are mapped to membership vectors on three linguistic variables, 'low', 'medium', and 'high', respectively, using a preset membership function; fuzzy inference: the two membership vectors are input into a fuzzy rule base for inference. The fuzzy rule base contains multiple rules defined in the form of "if-then" to determine the fuzzy output of the comprehensive recommendation degree; defuzzification: the fuzzy set of the comprehensive recommendation degree output by the fuzzy inference is defuzzified using the centroid method or the maximum membership method to obtain the accurate strategy decision score. This is only an example and is not intended to limit the invention.

[0048] Preferably, in some embodiments, the collaboration rule is used to define: when the scene adaptation evaluation quantity is at a "high" level, the weight factor of the preference association evaluation quantity in the fusion evaluation is reduced; when the preference association evaluation quantity is at a "low" level, the weight factor of the scene adaptation evaluation quantity in the fusion evaluation is increased.

[0049] In some embodiments, adjusting and executing the advertising delivery strategy for the target user in the current interaction scenario based on the strategy decision score can be achieved in the following way: First, a score threshold range associated with advertising performance indicators is set, and the strategy decision score is mapped to the corresponding delivery action level; second, based on the delivery action level, the real-time bidding bid, display order in the ad slot, or creative material selection of candidate ads are dynamically adjusted; then, the adjustment results of all candidate ads are integrated to generate a personalized advertising delivery sequence for the current interaction scenario and target user; finally, the delivery sequence is submitted and executed in real time through the application programming interface of the advertising delivery engine to complete the accurate display and exposure of the ads. For example, if the strategy decision score is higher than 0.8, the display priority of the ad is raised to the highest level, and a premium coefficient is applied to participate in real-time bidding; if the score is lower than 0.3, its bid is directly filtered or significantly reduced, thereby maximizing the overall revenue of advertising while ensuring user experience. This is only an example and is not intended to limit the specific scope of the invention.

[0050] On the other hand, in some embodiments, this application provides an e-commerce advertising placement strategy optimization system based on multi-source perception, referencing... Figure 4 The figure is a schematic diagram of the structure of an e-commerce advertising placement strategy optimization system based on multi-source perception, according to some embodiments of this application. The e-commerce advertising placement strategy optimization system 400 based on multi-source perception includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire long-term behavioral data of the target user and generate a user interest vector representing its interest preferences. In this application, the acquisition module 401 is also used to collect multi-dimensional context information of the current interaction scenario and generate a scenario vector that represents the semantics of the current interaction scenario and the real-time state of the target user. Processing module 402, in this application, is used to match the optimal associated historical scene from the pre-built historical scene strategy library according to the scene vector, and obtain the advertising placement strategy template corresponding to the optimal associated historical scene; In this application, the processing module 402 is also used to generate a preference association evaluation quantity that characterizes the degree of association between the candidate advertisement and the target user's long-term interests, based on the user interest vector and the candidate advertisement feature vector. In this application, the processing module 402 is also used to generate a scene adaptation evaluation quantity to characterize the degree of adaptation of the candidate advertisement to the current interactive scene, based on the scene vector and the candidate advertisement feature vector, under the guidance of the advertisement delivery strategy template. The execution module 403 in this application is mainly used to perform a fusion evaluation of the advertising placement decision in the current interaction scenario through the preference association evaluation quantity and the scenario adaptation evaluation quantity, to obtain the advertising placement strategy decision score, and then adjust and execute the advertising placement strategy for the target user in the current interaction scenario according to the strategy decision score.

[0051] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described method for optimizing e-commerce advertising placement strategies based on multi-source perception.

[0052] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing a multi-source perception-based e-commerce advertising placement strategy optimization method according to some embodiments of this application. The multi-source perception-based e-commerce advertising placement strategy optimization method in the above embodiments can be implemented through... Figure 5 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0053] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0054] The communication bus 502 can be used to transmit information between the aforementioned components.

[0055] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0056] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The e-commerce advertising placement strategy optimization method based on multi-source perception in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0057] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0058] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0059] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0060] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for optimizing e-commerce advertising placement strategies based on multi-source perception.

[0061] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0062] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for optimizing e-commerce advertising placement strategies based on multi-source perception, characterized in that, Includes the following steps: Acquire long-term behavioral data of target users and generate user interest vectors that represent their interests and preferences; Collect multi-dimensional contextual information of the current interaction scenario to generate a scenario vector that represents the semantics of the current interaction scenario and the real-time state of the target user; Based on the scenario vector, the optimal associated historical scenario is matched from the pre-built historical scenario strategy library, and the advertising placement strategy template corresponding to the optimal associated historical scenario is obtained. For candidate ads, a preference association evaluation metric is generated based on the user interest vector and the candidate ad feature vector to characterize the degree of association between the candidate ad and the target user's long-term interests. Guided by the advertising delivery strategy template, a scenario adaptation evaluation metric is generated based on the scenario vector and the candidate ad feature vector to characterize the degree of adaptation of the candidate ad to the current interactive scenario. The advertising placement decision in the current interaction scenario is evaluated by combining the preference association evaluation and the scenario adaptation evaluation to obtain the advertising placement strategy decision score. Then, the advertising placement strategy for the target user in the current interaction scenario is adjusted and executed according to the strategy decision score.

2. The method as described in claim 1, characterized in that, Obtaining long-term behavioral data of target users and generating user interest vectors that represent their interests and preferences specifically includes: Collect target users' advertising interaction behavior logs within a preset historical period. The advertising interaction behavior logs include at least ad clicks, ad viewing duration, and records of product collections or purchases associated with the ad content. The advertising interaction behavior logs are cleaned and normalized to extract interaction frequency and depth features corresponding to different advertising categories or themes; The interaction frequency and depth features are input into a pre-trained user interest model, which learns users’ implicit preferences for advertising categories or topics based on behavior logs. The user interest model outputs a multi-dimensional vector, which serves as the user interest vector.

3. The method as described in claim 1, characterized in that, Collecting multi-dimensional contextual information of the current interaction scenario to generate a scenario vector representing the semantics of the current interaction scenario and the real-time state of the target user specifically includes: Real-time acquisition of environmental status information and current user interaction content; The environmental state information and the current interaction content are respectively extracted and vectorized to obtain the environmental state vector and the user real-time interaction vector. By fusing the environmental state vector and the user's real-time interaction vector, a scene vector representing the semantics of the current interaction scene and the real-time state of the target user is generated.

4. The method as described in claim 1, characterized in that, Based on the scene vector, matching the optimal associated historical scene from the pre-built historical scene strategy library and obtaining the advertising placement strategy template corresponding to the optimal associated historical scene specifically includes: Calculate the similarity between the current scene vector and each historical scene vector in the historical scene strategy library; The historical scene with the highest similarity is selected as the optimal associated historical scene; Read the set of strategy parameters corresponding to the historical scenario and use it as the template for the advertising delivery strategy.

5. The method as described in claim 1, characterized in that, Based on the user interest vector and the candidate ad feature vector, a preference association evaluation metric is generated to characterize the degree of association between the candidate ad and the target user's long-term interests. This specifically includes: Determine the similarity between the user interest vector and the candidate ad feature vector; The similarity is corrected based on the time decay attribute of user interests; The corrected value is used as a preference association evaluation metric to characterize the degree of association between candidate ads and the long-term interests of target users.

6. The method as described in claim 5, characterized in that, Correcting the similarity based on the time decay attribute of user interests specifically includes: assigning decay weights to interest features in different time windows in the user interest vector, and weighting the similarity calculation results based on the decay weights.

7. The method as described in claim 1, characterized in that, The advertising placement decision in the current interaction scenario is evaluated by combining the preference association evaluation score and the scenario adaptation evaluation score, resulting in a strategy decision score for advertising placement. Determine the initial membership degree of the preference association evaluation value and the scenario adaptation evaluation value corresponding to the ad recommendation level, respectively; Based on the collaborative rules defined in the advertising strategy template, a two-dimensional evaluation relationship matrix is ​​constructed to describe the mutual influence between preference-related evaluation quantities and scenario-adaptation evaluation quantities. The numerical values ​​in the two-dimensional evaluation relation matrix are mapped to the corresponding fuzzy sets by using fuzzy membership functions; Based on the initial membership degree and the fuzzy set, a fusion decision fuzzy evaluation matrix is ​​constructed; Based on the fuzzy evaluation matrix of the fusion decision, the advertising placement decision in the current interaction scenario is evaluated by fuzzy comprehensive evaluation, and the strategy decision score of the advertising placement is output after defuzzification.

8. A multi-source perception-based e-commerce advertising placement strategy optimization system, characterized in that, include: The acquisition module is used to acquire long-term behavioral data of target users and generate user interest vectors that represent their interests and preferences. The acquisition module is also used to collect multi-dimensional context information of the current interaction scenario and generate a scenario vector that represents the semantics of the current interaction scenario and the real-time state of the target user. The processing module is used to match the optimal associated historical scene from the pre-built historical scene strategy library according to the scene vector, and obtain the advertising placement strategy template corresponding to the optimal associated historical scene. The processing module is further configured to generate a preference association evaluation metric, based on the user interest vector and the candidate ad feature vector, a metric representing the degree of long-term interest association between the candidate ad and the target user. The processing module is also used to generate a scene adaptation evaluation quantity, which characterizes the degree of adaptation of the candidate advertisement to the current interactive scene, based on the scene vector and the candidate advertisement feature vector, under the guidance of the advertisement delivery strategy template. The execution module is used to perform a fusion evaluation of the advertising placement decision in the current interaction scenario through the preference association evaluation quantity and the scenario adaptation evaluation quantity, to obtain the advertising placement strategy decision score, and then adjust and execute the advertising placement strategy for the target user in the current interaction scenario based on the strategy decision score.

9. A computer device comprising a memory and a processor, the memory storing code, characterized in that, The processor is configured to acquire the code and execute the e-commerce advertising placement strategy optimization method based on multi-source perception as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the e-commerce advertising placement strategy optimization method based on multi-source perception as described in any one of claims 1 to 7.

Images