A new media advertisement accurate pushing method and system based on user portrait

Abstract

The application discloses a new media advertisement accurate pushing method and system based on user portrait, relates to the technical field of new media advertisement pushing, and comprises the following steps: obtaining user behavior data and performing feature processing, and constructing a user intention state space; performing semantic modeling on advertisement content, constructing an advertisement semantic representation space, and establishing a cross-domain alignment relationship between user intention and advertisement semantics; predicting a user intention state variable of a target time frame based on a user intention state change trend, and performing semantic verification; generating semantic constraint information according to a verification result, matching advertisement data under the constraint of the cross-domain alignment relationship, and generating a target advertisement pushing set; and updating the user intention state and the semantic mapping relationship based on user feedback data. Through dynamic intention prediction, cross-domain semantic alignment and constraint matching, the application realizes accurate association of information content and advertisement content, and improves advertisement pushing accuracy.

Description

Technical Field

[0001] This invention relates to the field of new media advertising technology, specifically to a method and system for precise new media advertising based on user profiles. Background Technology

[0002] With the development of information distribution technology, the integration of content and commercial advertising within the same information flow scenario is constantly increasing. Users are often exposed to multiple types of commercial information simultaneously while browsing content. Therefore, how to achieve collaborative matching and accurate delivery between content and advertising content while ensuring information relevance has become an important research direction in the current field of information recommendation.

[0003] Current technologies for ad delivery typically build user interest models based on users' historical behavior data and combine them with content tags to achieve matching recommendations. However, in cross-content domain push scenarios, the semantic structures of news content and advertised products differ significantly. News texts focus on event narration and opinion expression, while advertising texts focus on product attributes and consumption guidance. They belong to different domains in terms of semantic expression, attribute description granularity, and emotional tone, making it difficult to establish an effective cross-domain mapping relationship between user reading intent and advertising semantics. Furthermore, existing solutions often rely on static feature representations, summarizing user interests into a set of fixed tags or a single vector. This makes it difficult to effectively characterize the user's interest migration process in different information scenarios and to handle the implicit, non-explicit correlation between reading intent and advertised product attributes within the context.

[0004] Existing technologies suffer from difficulties in cross-domain semantic mapping between user intent and advertising, insufficient characterization of dynamic migration of user interests, and limited accuracy of push matching. There is an urgent need for an advertising push method that can establish a cross-domain mapping relationship between user intent and advertising semantics, dynamically capture changes in user intent, and achieve accurate matching. Summary of the Invention

[0005] To address the aforementioned technical issues, this paper provides a method and system for precise new media advertising based on user profiles. This technical solution solves the problems mentioned in the background section regarding the difficulty in cross-domain semantic mapping between user intent and advertising, insufficient characterization of dynamic migration of user interests, and limited accuracy of push matching in existing technologies.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for precise new media advertising based on user profiles includes: S1. Obtain the target user's behavioral data, and perform feature extraction and fusion processing on the behavioral data to construct the user intent state space and map the user behavioral features into state variables. S2. Perform semantic modeling on the advertising content, construct the advertising semantic representation space, and establish a cross-domain alignment relationship between the user intent state space and the advertising semantic representation space; S3. Based on the change trend information of the user intent state space, predict the change of the user intent state in the target time frame to obtain the predicted user intent state variable. S4. Within the semantic representation space of the advertisement, the advertisement data is matched based on the predicted user intent state variable, and cross-domain alignment relationship is introduced during the matching process to obtain the target advertisement push set. S5. Obtain user feedback behavior data on the target ad push set, and update the user intent state space and user intent state prediction direction based on the feedback behavior data.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: In this invention, by constructing an advertising core matrix, the core semantics of information content and advertising content are extracted and corresponding relationships are established under the same semantic framework, forming an advertising semantic representation space. Based on semantic similarity calculation, a cross-domain alignment relationship is established between user behavior features and core semantics. This makes the event narrative semantics of the content and the product attribute semantics of the advertised goods directly comparable and computable in a unified latent space, avoiding the semantic mapping offset and matching attenuation problems caused by the different domains of the content and the advertisement in traditional methods. Under the constraint of the cross-domain alignment relationship, the correspondence between the purchase intention implied in the context and the attributes of the advertised goods can be accurately identified. Even when the semantic structures of the content and the advertised goods are significantly different, the matching results can still maintain a high degree of relevance, thus ensuring the stability of the push quality from the root.

[0008] This invention generates a user intent change curve and identifies the curve's turning points. It then performs semantic element decomposition and item-by-item comparison of the core semantics of the advertisement at the turning points and the adjacent time frames before and after them, forming a semantic consistency analysis result across the time dimension. The predicted user intent state variable is then checked for directional consistency and deviation consistency with this result. The prediction results of trend extrapolation are independently verified through historical semantic evolution paths. When there is a deviation between trend prediction and semantic evidence, the predicted variable is automatically corrected, avoiding the prediction inaccuracies that may arise from simply relying on statistical trend extrapolation. This ensures that the predicted user intent state variable can both continue the long-term evolution trend and be promptly corrected by short-term semantic attention changes, providing more reliable input conditions for subsequent matching.

[0009] This invention transforms the verification results into semantic constraint information and introduces them into the advertising semantic representation space. By performing semantic constraint mapping on the predicted user intent state variables through cross-domain alignment relationships, it accurately maps them from the user intent state space to the target semantic region in the semantic representation space. The semantic offset boundary is determined based on the mapping distance distribution or historical mapping sample statistics, and results exceeding the boundary are truncated or remapped. The matching process is upgraded from traditional tag matching or vector similarity fuzzy matching to semantically precise matching subject to multiple constraints. This ensures that the selection of candidate advertisements simultaneously meets semantic constraints and cross-domain alignment mapping constraints, effectively preventing the matching results from deviating from the user's true intent due to semantic drift. This ensures that the final pushed advertisements are highly consistent with the user's current browsing focus and consumption intent at the semantic level. Attached Figure Description

[0010] Figure 1 This is a flowchart of the new media advertising precision push method in this invention; Figure 2 This is a flowchart of the steps for obtaining user intent state variables in this invention; Figure 3 This is a flowchart of the intention prediction and verification steps in this invention; Figure 4 This is a framework diagram of the new media advertising precision push system in this invention. Detailed Implementation

[0011] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0012] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0013] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0014] Reference Figures 1-4As shown, a method for precise targeting of new media advertisements based on user profiles specifically includes: S1. Obtain the target user's behavioral data, and perform feature extraction and fusion processing on the behavioral data to construct the user intent state space and map the user behavioral features into state variables. S2. Perform semantic modeling on the advertising content, construct the advertising semantic representation space, and establish a cross-domain alignment relationship between the user intent state space and the advertising semantic representation space; S3. Based on the change trend information of the user intent state space, predict the change of the user intent state in the target time frame to obtain the predicted user intent state variable. S4. Within the semantic representation space of the advertisement, the advertisement data is matched based on the predicted user intent state variable, and cross-domain alignment relationship is introduced during the matching process to obtain the target advertisement push set. S5. Obtain user feedback behavior data on the target ad push set, and update the user intent state space and user intent state prediction direction based on the feedback behavior data.

[0015] Optionally, step S1 specifically includes: Acquire historical behavior data of target users and historical advertising data corresponding to the historical behavior data, and perform correlation and matching processing on the historical behavior data and historical advertising data according to time sequence and event correspondence to form a data set corresponding to historical behavior and historical push. Statistical processing is performed on historical behavior data, including counting the number of occurrences of behavior, calculating the time interval between occurrences of behavior, summarizing the duration of behavior, and performing time aggregation processing on behavior within different time windows, and classifying behavior types to obtain a feature set composed of time behavior features, content preference features, and behavior feedback features. The historical push content in the historical advertising data is structured and the historical push feedback data is quantified. The historical push content and the corresponding historical push feedback and feature set are fused together to form a historical push related data set. Based on the corresponding data set of historical behavior and historical push, as well as the related data set of historical push, a mapping structure for the correspondence between behavioral features and push results is established, and a user intent state space is constructed based on the mapping structure. The user behavior features in the feature set are mapped to state variables, and the state variables can change over time. In an optional embodiment, a correlation structure between user behavior and ad push is first established based on time sequence and event correspondence, so that data from different sources can form an alignable data foundation in the time dimension; Based on this, user behaviors are statistically aggregated according to the time sequence of their occurrence. The statistically processed user behaviors are then summarized in chronological order to extract time behavior features, content preference features, and behavior feedback features. These features are then combined to form a feature set. Among them, the occurrence time of each behavior record, the time interval between the corresponding behavior record and the previous behavior record, and the duration of the corresponding behavior are used as the three time behavior feature components corresponding to the behavior record. The time behavior feature components of all behavior records are arranged in the order of occurrence time to form the time behavior feature. For each content category, the number of times a user's behavior occurs in the corresponding category is counted, and the proportion of the number of behaviors in the corresponding category to the total number of behaviors in all categories is calculated to obtain the frequency percentage. At the same time, the total duration of the user's behavior in the corresponding category is counted, and the proportion of the total duration of the behavior in the corresponding category to the total duration of all categories is calculated to obtain the duration percentage. The frequency percentage and the duration percentage are added together and the average value is taken as the user's preference intensity for the content category. The preference intensity of each content category together constitutes the content preference feature. Based on user feedback records of historical push content, the feedback behaviors corresponding to historical push content are statistically analyzed. Feedback behaviors include user actions such as clicking on historical push content, continuing to browse historical push content, and exiting historical push content. The user's feedback on historical push content is determined based on the number of records corresponding to different feedback behaviors, and behavioral feedback features are generated. Furthermore, the push content and feedback information in the historical advertising data are structured to form a corresponding mapping between advertising content and user behavior characteristics. Based on the mapping, a user intent state space is constructed, thereby unifying the mapping of the user's historical behavior and advertising response relationship into a state expression structure. Subsequently, the features in the feature set that reflect the pattern of changes in user behavior are abstracted into state variables that change over time, in order to describe the evolution of user intent over time. For example, when target users click and stay on "sports ads", "financial information ads" and "e-commerce promotion ads" in different time periods, the above behaviors are aligned with the corresponding ad push records in chronological order, and the number of clicks and average stay time of users in each type of ad are counted to form time behavior characteristics corresponding to sports, finance and e-commerce categories respectively. Meanwhile, based on the occurrence time corresponding to the user behavior record, the historical behavior records arranged in chronological order are aggregated and processed. The number of behaviors generated by users for different content categories under each time sequence and the corresponding duration of the behaviors are counted. Based on the number of behaviors and the duration of the behaviors, the preference degree corresponding to each content category is determined. The preference degree corresponding to each content category under different time sequences is arranged to form content preference features that include time change relationships. Subsequently, the aforementioned behavioral characteristics are integrated with the advertising push feedback results and mapped together to the user intent state space. Finally, the time-varying characteristics in the user intent state space are expressed as state variables to characterize the user's intent change process from "sports interest-driven" to "e-commerce interest-enhanced".

[0016] Optionally, step S2 specifically includes: Extract the core semantics corresponding to each advertisement in the historical advertising data, and bind the core semantics with the corresponding advertisements in the historical advertising data to construct an advertising core matrix; Obtain the state components of user behavior features corresponding to the user intent state variables, and map the core semantics in the core advertising matrix to the state components of the user behavior features to construct an advertising semantic representation space; Associating user behavior characteristics with core semantics to establish cross-domain alignment relationships; In an optional embodiment, semantic analysis is performed on historical advertising data to extract the corresponding core semantic information for each historical advertising data, and the core semantic information is bound to the corresponding advertising to form an advertising core matrix indexed by semantics. The advertising core matrix is ​​used to uniformly represent the structural relationship of different advertisements in the semantic dimension, so that different advertisements are comparable under a unified semantic framework. The core semantics can be extracted using one or more methods, such as keyword extraction and topic model analysis, and represented in the form of semantic vectors. The dimensions of each semantic vector are consistent with the semantic representation space of the subsequent advertisement. Based on this, corresponding state variables are obtained according to the time of user behavior occurrence. The state variables are formed by mapping time behavior features, content preference features and behavior feedback features, and are used to represent the behavioral state of the target user at the corresponding behavior time. The state variable includes multiple state components, each of which corresponds to different types of user interests, behavioral activity levels, and historical feedback responses. Since the user intent state variable is formed by the mapping of time behavior features, content preference features, and behavioral feedback features, when performing ad semantic matching, the corresponding user behavior feature state components in the state variable are directly obtained, and the state components are used to represent the user's current behavioral tendency. Subsequently, the state variables are associated with each core semantic in the core advertising matrix one by one. The correspondence is determined based on the semantic similarity between the state variables and each core semantic, and the advertising semantic representation space is constructed based on the correspondence. Specifically, state variables are used as input for a unified semantic mapping. Correlation calculations are performed on each semantic vector in the core ad matrix to generate a corresponding correlation weight for each core semantic. The correlation weight is used to represent the degree of matching of the corresponding ad semantic under the current user intent state. Each core semantic is combined with the correlation weight to form the ad semantic representation space, so that the ad semantic representation space can reflect the cross-domain alignment relationship between the current user intent state and different ad semantics. Furthermore, the aforementioned association weights are used as a quantitative expression of cross-domain alignment relationships to establish cross-domain alignment relationships between the user intent state space and the advertising semantic representation space; Specifically, the advertising text is converted into a semantic vector through semantic encoding, the state variables are converted into a state vector through feature encoding, and the semantic similarity is calculated based on the angle relationship between the semantic vector and the state vector. The semantic similarity can be calculated using cosine similarity or a metric based on the distance in the vector space, and association weights are generated based on the semantic similarity results to establish a cross-domain alignment relationship between the user intent state space and the advertising semantic representation space. Semantic similarity can be calculated using cosine similarity or a distance-based method in vector space. The association weight is directly generated based on the semantic similarity result: when using cosine similarity, the calculated cosine value is directly used as the association weight; when using a distance-based method in vector space, the reciprocal of the distance value is used as the association weight. For example, in the target user's historical behavior records at multiple consecutive time points, in the early stage, they mainly browsed sports event information and occasionally clicked on sports equipment advertisements; then they gradually increased their browsing and clicking on e-commerce promotional information, while their attention to sports content gradually decreased. Once completed, corresponding state variables are generated at the target time nodes. These state variables comprehensively include time behavior characteristics, content preference characteristics, and behavioral feedback characteristics. For example, they can be represented as: high level of interest in sports, gradually increasing level of interest in e-commerce, moderate level of acceptance of advertising, and high level of activity at night. Subsequently, core semantics were extracted from historical advertising data, such as "sports event analysis", "sports equipment recommendations", "e-commerce promotional activities", and "digital product discounts", and an advertising core matrix was constructed. Then, the state variables corresponding to the target time nodes are associated with the core semantics mentioned above to calculate the association weights for each core semantic. For example, the association weight for sports event analysis is 0.63, the association weight for sports equipment recommendation is 0.71, the association weight for e-commerce promotion activities is 0.78, and the association weight for digital product discounts is 0.46. Based on this, an advertising semantic representation space is constructed. Meanwhile, the aforementioned association weights are used as the correspondence between the user intent state space and the advertising semantic representation space. Based on the content preference features and behavioral feedback features contained in the state variables, the corresponding semantic features in the advertising semantic representation space are matched to determine the advertising semantic representation corresponding to the user's current state variable.

[0017] Optionally, step S3 specifically includes: Extract the target user's intent change data from the user intent state space, merge the intent change data into a user intent change curve, and extract the curve inflection points in the user intent change curve. Based on the time frame of the curve inflection point, extract the corresponding advertisement from the historical advertising data, and the advertisement contained in the time frame before and after the corresponding advertisement. Extract the core semantics of the corresponding advertisement, and compare the core semantics of the three advertisements. A state evolution sequence is formed based on state variables according to a time series, and user intent change trend information is generated based on the state evolution sequence; Based on the changing trend information of the user intent state space, the changes of the user intent state in time frames are predicted to obtain the predicted user intent state variables. Obtain the comparison results of the core semantics, verify the predicted user intent state variables with the comparison results, and obtain the verification results; In an optional embodiment, based on the state variables corresponding to different time nodes in the user intent state space, the state change records of the target user at different time nodes are processed by time serialization, and a state evolution sequence is formed on this basis to represent the dynamic change process of user intent in the time dimension. The state evolution sequence can be further represented as a continuously changing user intent change curve. Based on the state evolution sequence, the changes in user intent are analyzed to identify key change locations. These key change locations correspond to time points when user intent changes significantly, i.e. curve inflection points. The curve inflection points are then matched with time points in the user's historical behavior to determine the historical advertising data corresponding to the time points. After determining the historical advertising data, the semantic information of the advertisements at the corresponding time nodes and adjacent time nodes is extracted, and the semantic information is correlated and compared. The semantic information of different time frames is split and aligned into semantic elements, and the semantic elements are compared item by item to generate a semantic difference set. Based on the semantic difference set, the semantic consistency analysis results across time dimensions are obtained. Furthermore, based on the matching results between the current state evolution sequence and the historical state evolution sequence, the state variables corresponding to multiple consecutive time frames before the target time frame are obtained, and the changes in each state variable between adjacent time frames are calculated to form a state change sequence. According to the consistency of the change direction of the two most recent adjacent changes in the state change sequence, the state variables corresponding to the target time frame are continuously derived. Taking the state variable corresponding to the most recent time frame as the initial value, the change direction of the user intention state is determined according to the change relationship between the state variables corresponding to consecutive time frames, and the association relationship of the corresponding features in the user intention state space is adjusted based on the change direction. By combining the correspondence between the user intent state space and the advertising semantic representation space, the predicted user intent state variables corresponding to the target time are determined, and the prediction results are compared and verified with the semantic consistency analysis results to determine whether the semantic evolution direction corresponding to the predicted user intent state variables is consistent with the historical semantic evolution direction, and whether the predicted user intent state variables correspond to the subsequent semantics in the historical semantic evolution path. For example, target users may exhibit the following behaviors over a continuous period of time: in the early stage, they mainly browse "digital product review content"; in the middle stage, they gradually increase their attention to "smart wearable device recommendation content"; and in the later stage, they focus on clicking on "Bluetooth headphone promotional advertisements". First, based on the user's state variables at different time points, a user intent state evolution sequence is formed in chronological order. By analyzing the changing relationships between adjacent state variables in the state evolution sequence, the locations where significant state changes are identified. In the visualization, the state evolution sequence can be represented as a user intent change curve to observe the changing trend of user intent. The user intent change curve is only used as a graphical representation of the state evolution sequence and does not participate in subsequent calculations. Subsequently, based on the relationship between the changes in state variables in the state evolution sequence, the time position corresponding to the change in the user's intent state is determined, and this time position is matched with the occurrence time corresponding to the historical behavior record to obtain the historical advertising data corresponding to this time position. At this time point and its adjacent time points, the corresponding advertising semantic information is extracted. For example, the previous stage corresponds to "smart device technology trend analysis", the turning point stage corresponds to "smart wearable product recommendation", and the next stage corresponds to "Bluetooth headset promotion information". Furthermore, the semantic information corresponding to the user's historical behavior, the semantic information corresponding to the predicted user intent, and the semantic information corresponding to the advertisement are compared and correlated to determine the correspondence between different semantic information, and a semantic consistency verification result is generated based on the degree of matching between the semantic information. Finally, a consistency judgment is made based on the semantic change relationship of the predicted user intent state variable and the corresponding change relationship in the historical state evolution sequence. When the two are consistent, the predicted user intent state variable is determined to be valid. When the two are inconsistent, the predicted user intent state variable is judged as an abnormal prediction result, and the advertisement semantic matching is performed again.

[0018] Optionally, step S4 specifically includes: The verification results are transformed into semantic constraint information and introduced into the advertising semantic representation space; Using predicted user intent state variables as matching input, semantic matching and filtering of advertising data is performed under the constraint of semantic information. Cross-domain alignment is introduced during the matching process to perform semantic constraint mapping on the predicted user intent state variables, mapping them from the user intent state space to the target semantic region in the advertising semantic representation space, and limiting the semantic offset range during the mapping process. Based on the constrained semantic mapping results, the advertising data is filtered within the advertising semantic representation space to ensure that the matching results meet the semantic consistency of user intent. Generate a target set of ads to push based on the filtering results; In an optional embodiment, the verification result is processed in a structured manner, and semantic consistency identifiers and semantic deviation identifiers are extracted. Semantic constraint information is generated based on the identifiers, and the semantic constraint information is used as a limiting condition in the matching process to participate in semantic filtering. Among them, semantic consistency identifiers and semantic deviation identifiers are obtained based on the similarity between the verification results and the predicted user intent state variables; Furthermore, the semantic consistency relationships and semantic deviation relationships in the verification results are identified and processed to form semantic consistency constraint terms and semantic deviation constraint terms; Based on the correspondence between user behavior features and core semantics, the semantic region where the corresponding core semantics is located is determined, and the semantic region is determined as the semantic region that is allowed to match. The remaining semantic regions that have not established a correspondence are determined as semantic regions that restrict matching, thereby forming semantic region restriction conditions. Simultaneously, semantic constraint information is generated based on the correspondence established between user intent state variables and advertising semantic representations, and the semantic constraint information is used to limit the deviation between candidate advertising semantics and the corresponding semantic representations of user intent state variables, so that the candidate advertising semantics remain within the range defined by the correspondence. Ultimately, semantic consistency constraints, semantic deviation constraints, semantic region restrictions, and semantic evolution constraints together constitute semantic constraint information, and each serves as an independent semantic constraint term within the semantic constraint information. Using the predicted user intent state variable as the matching input, the advertising data is semantically matched and filtered within the constraints of the semantic constraint information. The matching process performs semantic mapping on the predicted user intent state variable through cross-domain alignment relationship, so that it is mapped from the user intent state space to the target semantic region in the advertising semantic representation space, and the range of semantic offset generated in the mapping process is constrained and controlled. In this context, a semantic region represents a set of continuous semantic locations in the advertising semantic representation space, consisting of multiple related core semantics. Each core semantic within the same semantic region is associated with the same user behavior feature and has the same or similar semantic attributes. When multiple core semantics are associated with the same user behavior feature, the continuous semantic locations of the multiple core semantics are determined as the corresponding semantic region. When the semantic offset range exceeds the constraint boundary, the excess part is truncated or remapped to the nearest semantic neighborhood region. Based on the constrained semantic mapping results, the advertising data is filtered to ensure that the filtering results simultaneously satisfy both semantic constraints and cross-domain alignment mapping constraints, and a target advertising push set is generated based on the filtering results. The semantic offset range is determined based on the mapping distance distribution between user intent state variables and semantic representation space in cross-domain alignment relationships, or based on the statistical distribution range of historical mapping samples; For example, based on the prediction of changes in user intent state, it is found that the user has a high intent state variable for "digital consumer content", and the verification result is obtained by semantic comparison of historical time nodes. The verification result indicates that the user has a high consistency in "digital product promotion semantics" at a specific time stage. The verification result is converted into semantic constraint information and introduced into the advertising semantic representation space. This allows the semantic regions related to "digital consumption" and "promotional semantics" to be included in the constraint range. The predicted user intent state variable is used as the matching input to perform semantic matching filtering within the constraint range. Meanwhile, user intent state variables are mapped to the “digital consumption semantic region” in the advertising semantic representation space through cross-domain alignment relationships, and the semantic offset that may occur during the mapping process is restricted so that the mapping result stably corresponds to the semantic region. Finally, under the condition of simultaneously satisfying semantic constraints and mapping constraints, content with semantic consistency with "Bluetooth headphone promotion" and "digital product recommendation" is filtered out from the advertising data, and a target advertising push set is generated.

[0019] Optionally, step S5 specifically includes: Obtain feedback behavior data of target users to the target ad push set, associate the feedback behavior data with the ad identifiers in the target ad push set, and map the association results to the corresponding state variables in the user intent state space to correct the predicted user intent state variables. Based on the corrected predicted user intent state variables, the weight distribution of semantic constraint terms in the semantic constraint information is adjusted, and the constraint structure of the semantic constraint information is updated. Based on feedback behavior data, the cross-domain alignment relationship between user intent and advertising semantics in the user intent state space is corrected, so that the corresponding mapping relationship of the cross-domain alignment relationship in the semantic representation space is updated synchronously. The updated semantic constraint information, predicted user intent state variables, and cross-domain alignment relationships are used as the input structure for the subsequent ad semantic matching process. In an optional embodiment, feedback behavior data of target users to the target ad push set is collected, and the feedback behavior data is associated with the ad identifiers in the target ad push set to determine the specific ad object corresponding to each feedback behavior. The association result is mapped to the corresponding state variable in the user intent state space, thereby correcting the predicted user intent state variable and making the user intent state structurally updated with the feedback result. Based on this, the weight distribution of semantic constraint terms in the semantic constraint information is adjusted according to the corrected predicted user intent state variables, and the semantic constraint structure is updated so that the constraint strength of different semantic constraint terms is redistributed in subsequent matching. Specifically, when the feedback behavior indicates that the user has a positive response to the target advertisement content, the clicks and dwell time generated by this feedback behavior are included in the behavioral statistics of the corresponding content category in the corresponding user intent state variable, the frequency and duration of this content category are recalculated, and the preference intensity of the corresponding content category in the content preference feature is updated; when the feedback behavior indicates that the user has a negative response to the target advertisement content, the negative feedback record is included in the statistical results of the behavioral feedback feature, the negative feedback rate of the content category is recalculated, and the preference intensity of the corresponding content category in the content preference feature is reduced according to the change ratio of the negative feedback rate. Based on feedback behavior data, the cross-domain alignment relationship between user intent and advertising semantics in the user intent state space is further corrected, so that the corresponding structure of the original mapping relationship in the semantic representation space is synchronously adjusted, thereby updating the mapping consistency between user intent and semantic space. Finally, the updated semantic constraint information, predicted user intent state variables, and cross-domain alignment relationships are used as the input structure for the subsequent ad semantic matching process to support the continuous iteration of the subsequent push process. For example, after receiving a set of targeted ads, a target user might click on a "Bluetooth headphone promotion ad" and complete a purchase, only browse a "smartwatch recommendation ad" without taking any further action, and quickly swipe past and ignore a "financial management ad." First, the above feedback behaviors are associated with the corresponding advertising identifiers and mapped to the corresponding state variables in the user intent state space. The weight of the state variable corresponding to "purchase behavior" is increased, and the weight of the state variable corresponding to "ignore behavior" is decreased, thereby correcting the predicted user intent state variables and making structural adjustments to them in the dimensions related to "digital consumption intent". For positive feedback behaviors such as user clicks, favorites, increased dwell time, or completed conversions, the weight of the user intent state variable corresponding to the ad semantics is increased; for negative feedback behaviors such as skipping, closing, or short-term exit, the weight of the corresponding state variable is decreased. The adjustment range is determined based on the intensity of the feedback behavior, that is, the higher the correlation between the feedback and the behavior, the greater the weight adjustment range. Based on the correction results, the semantic constraint information is adjusted. For example, the constraint weight of "digital consumption semantics" is increased, while the constraint priority of "financial management semantics" is reduced, so that the semantic filtering range is redistributed in the subsequent matching process. Furthermore, based on feedback behavior data, the cross-domain alignment relationship is corrected, which strengthens the mapping relationship between "digital consumption intent" in the user intent state space and "digital product promotion semantics" in the advertising semantic representation space, while weakening the mapping relationship with "financial management semantics", thereby updating the correspondence structure between semantic spaces. Ultimately, the updated semantic constraints, user intent state variables, and cross-domain alignment relationships are reused in subsequent ad semantic matching processes to support the update of the input structure for the next round of ad pushes.

[0020] A new media advertising precision targeting system based on user profiles, used to implement the aforementioned new media advertising precision targeting method, specifically includes: The user intent state construction component is used to construct the user intent state space and the state variable expression structure. Semantic representation building components are used to construct the semantic space and semantic structure expression of advertisements; The state evolution prediction component is used for time series modeling and prediction of user intent states; A cross-domain semantic matching component is used to match and filter user intent with advertising semantics; The feedback update component is used to optimize the state space and semantic relationships based on user feedback.

[0021] Optionally, the user intent state construction component specifically includes: The behavior data collection unit is used to acquire historical behavior data of target users and corresponding historical advertising data; The feature fusion unit is used to perform feature fusion processing on behavioral data based on time, content, and feedback data. The state space generation unit is used to construct the user intent state space based on the fused feature set and generate corresponding state variables.

[0022] The semantic representation construction component specifically includes: The core semantic extraction unit is used to extract the core semantic information corresponding to the advertisement; Semantic structuring units are used to bind core semantics with corresponding advertisements to form a set of semantic structures; Semantic space construction unit, used to construct an advertising semantic representation space based on a set of semantic structures.

[0023] Optionally, the state evolution prediction component specifically includes: The state sequence generation unit is used to generate time series structures based on user intent state variables; The change feature extraction unit is used to extract the change direction and change magnitude features between state variables; The state derivation unit is used to derive the user intention state of the target time frame based on the state evolution sequence to obtain the predicted state variables. The cross-domain semantic matching component specifically includes: The semantic constraint generation unit is used to convert the verification results into semantic constraint information; Semantic mapping unit, used to map user intent state variables to semantic representation space through cross-domain alignment relationships; The matching and filtering unit is used to filter advertising data under the combined effect of semantic constraints and mapping constraints, and generate a target push set.

[0024] Optionally, the feedback update component specifically includes: The feedback acquisition unit is used to acquire feedback behavior data of the target user to the push collection; The state correction unit is used to correct and update the user intent state variables based on feedback data; The alignment update unit is used to update the cross-domain alignment relationship between the user intent state space and the semantic representation space.

[0025] The advantages of this invention lie in its construction of a dual-space structure—a user intent state space and an advertising semantic representation space—and the introduction of cross-domain alignment relationships. This enables a mappable association between user behavior features and semantic representations, allowing user intent to be continuously characterized as state variables that evolve over time, rather than being limited to static profiles. Simultaneously, based on the predicted user intent state variables, these variables are used as semantic matching inputs. Combined with semantic constraint information generated from historical semantic consistency verification results, advertising data is subjected to constrained filtering within the semantic representation space. This introduces both "prediction-driven" and "semantic constraint" conditions during the matching process, enhancing its determinism and consistency. Furthermore, cross-domain alignment relationships are used to semantically map user intent state variables, and semantic offsets during the mapping process are constrained and controlled, maintaining structural stability in the transition between the user intent space and the semantic space. Moreover, by synchronously updating the user intent state space, semantic constraint structure, and cross-domain alignment relationships using feedback behavioral data, the prediction direction and semantic mapping relationships can be continuously corrected based on actual interaction results. This forms a closed-loop mechanism of state modeling, intent prediction, semantic constraint matching, and feedback correction, enabling continuous adaptive optimization of the advertising push process at both the semantic and user intent levels.

[0026] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for precise targeting of new media advertisements based on user profiles, characterized in that, include: S1. Obtain the target user's behavior data, and perform feature extraction and fusion processing on the behavior data to construct the user intent state space and map the user behavior features into state variables. S2. Perform semantic modeling on the advertising content, construct the advertising semantic representation space, and establish a cross-domain alignment relationship between the user intent state space and the advertising semantic representation space; S3. Based on the change trend information of the user intent state space, predict the change of the user intent state in the target time frame to obtain the predicted user intent state variable. S4. Within the semantic representation space of the advertisement, the advertisement data is matched based on the predicted user intent state variable, and cross-domain alignment relationship is introduced during the matching process to obtain the target advertisement push set. S5. Obtain user feedback behavior data on the target ad push set, and update the user intent state space and user intent state prediction direction based on the feedback behavior data.

2. The method for precise new media advertising based on user profiles according to claim 1, characterized in that, Step S1 specifically includes: Acquire historical behavior data of target users and historical advertising data corresponding to the historical behavior data, and perform correlation and matching processing on the historical behavior data and historical advertising data according to time sequence and event correspondence to form a data set corresponding to historical behavior and historical push. Statistical processing is performed on historical behavior data, including counting the number of occurrences of behavior, calculating the time interval between occurrences of behavior, summarizing the duration of behavior, and performing time aggregation processing on behavior within different time windows, and classifying behavior types to obtain a feature set composed of time behavior features, content preference features, and behavior feedback features. The historical push content in the historical advertising data is structured and the historical push feedback data is quantified. The historical push content and the corresponding historical push feedback and feature set are fused together to form a historical push related data set. Based on the corresponding data set of historical behavior and historical push, as well as the related data set of historical push, a mapping structure for the correspondence between behavioral features and push results is established, and a user intent state space is constructed based on the mapping structure. User behavior features in the feature set are mapped to state variables, which can change over time.

3. The method for precise new media advertising based on user profiles according to claim 2, characterized in that, Step S2 specifically includes: Extract the core semantics corresponding to each advertisement in the historical advertising data, and bind the core semantics with the corresponding advertisements in the historical advertising data to construct an advertising core matrix; Obtain the state components of user behavior features corresponding to the user intent state variables, and map the core semantics in the core advertising matrix to the state components of the user behavior features to construct an advertising semantic representation space; By associating user behavior characteristics with core semantics, a cross-domain alignment relationship can be established.

4. The method for precise new media advertising based on user profiles according to claim 3, characterized in that, Step S3 specifically includes: Extract the target user's intent change data from the user intent state space, merge the intent change data into a user intent change curve, and extract the curve inflection points in the user intent change curve. Based on the time frame of the curve inflection point, extract the corresponding advertisement from the historical advertising data, and the advertisement contained in the time frame before and after the corresponding advertisement. Extract the core semantics of the corresponding advertisement, and compare the core semantics of the three advertisements. A state evolution sequence is formed based on state variables according to a time series, and user intent change trend information is generated based on the state evolution sequence; Based on the changing trend information of the user intent state space, the changes of the user intent state in time frames are predicted to obtain the predicted user intent state variables. Obtain the comparison results of the core semantics, verify the predicted user intent state variables with the comparison results, and obtain the verification results.

5. The method for precise new media advertising based on user profiles according to claim 4, characterized in that, Step S4 specifically includes: The verification results are transformed into semantic constraint information and introduced into the advertising semantic representation space; Using predicted user intent state variables as matching input, semantic matching and filtering of advertising data is performed under the constraint of semantic information. Cross-domain alignment is introduced during the matching process to perform semantic constraint mapping on the predicted user intent state variables, mapping them from the user intent state space to the target semantic region in the advertising semantic representation space, and limiting the semantic offset range during the mapping process. Based on the constrained semantic mapping results, the advertising data is filtered within the advertising semantic representation space to ensure that the matching results meet the semantic consistency of user intent. And generate a set of target ads to push based on the filtering results.

6. The method for precise new media advertising based on user profiles according to claim 5, characterized in that, Step S5 specifically includes: Obtain feedback behavior data of target users to the target ad push set, associate the feedback behavior data with the ad identifiers in the target ad push set, and map the association results to the corresponding state variables in the user intent state space to correct the predicted user intent state variables. Based on the corrected predicted user intent state variables, the weight distribution of semantic constraint terms in the semantic constraint information is adjusted, and the constraint structure of the semantic constraint information is updated. Based on feedback behavior data, the cross-domain alignment relationship between user intent and advertising semantics in the user intent state space is corrected, so that the corresponding mapping relationship of the cross-domain alignment relationship in the semantic representation space is updated synchronously. The updated semantic constraint information, predicted user intent state variables, and cross-domain alignment relationships are used as the input structure for the subsequent ad semantic matching process.

7. A new media advertising precision push system based on user profiles, used to implement the new media advertising precision push method according to any one of claims 1-6, characterized in that, include: The user intent state construction component is used to construct the user intent state space and the state variable expression structure. Semantic representation building components are used to construct the semantic space and semantic structure expression of advertisements; The state evolution prediction component is used for time series modeling and prediction of user intent states; A cross-domain semantic matching component is used to match and filter user intent with advertising semantics; The feedback update component is used to optimize the state space and semantic relationships based on user feedback.

8. A new media advertising precision push system based on user profiles according to claim 7, characterized in that, The user intent state construction component specifically includes: The behavior data collection unit is used to acquire historical behavior data of target users and corresponding historical advertising data; The feature fusion unit is used to perform feature fusion processing on behavioral data based on time, content, and feedback data. The state space generation unit is used to construct the user intent state space based on the fused feature set and generate corresponding state variables.

9. The semantic representation construction component specifically includes: The core semantic extraction unit is used to extract the core semantic information corresponding to the advertisement; Semantic structuring units are used to bind core semantics with corresponding advertisements to form a set of semantic structures; Semantic space construction unit, used to construct an advertising semantic representation space based on a set of semantic structures.

10. A new media advertising precision push system based on user profiles according to claim 8, characterized in that, The state evolution prediction component specifically includes: The state sequence generation unit is used to generate time series structures based on user intent state variables; The change feature extraction unit is used to extract the change direction and change magnitude features between state variables; The state derivation unit is used to derive the user intention state of the target time frame based on the state evolution sequence to obtain the predicted state variables. The cross-domain semantic matching component specifically includes: The semantic constraint generation unit is used to convert the verification results into semantic constraint information; Semantic mapping unit, used to map user intent state variables to semantic representation space through cross-domain alignment relationships; The matching and filtering unit is used to filter advertising data under the combined effect of semantic constraints and mapping constraints, and generate a target push set.

11. A new media advertising precision push system based on user profiles according to claim 9, characterized in that, The feedback update component specifically includes: The feedback acquisition unit is used to acquire feedback behavior data of the target user to the push collection; The state correction unit is used to correct and update the user intent state variables based on feedback data; The alignment update unit is used to update the cross-domain alignment relationship between the user intent state space and the semantic representation space.

Images