Retail customer interaction content generating and pushing system based on AIGC

Abstract

The invention is suitable for the technical field of intelligent content generation and personalized recommendation control, and provides an AIGC-based retail client interaction content generation and push system, which comprises an information acquisition module, a track recognition module, a behavior association judgment module, a correction parameter generation module and a content complexity regulation and control module. According to the method, the user behavior track and the wearable equipment data are fused, and a content complexity level correction mechanism based on the combined driving of the fixation time trend and the physiological state change trend is constructed. Different from an existing mode of setting a recommendation strategy only according to a behavior label or a single-dimensional feature, the method introduces a final correction parameter to dynamically adjust the complexity level of the original content, so that the generated content better fits the current attention intention and cognitive state of the user in the aspects of expression level, information density and logic structure.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent content generation and personalized recommendation control, and particularly relates to a retail customer interaction content generation and push system based on AIGC. Background Art

[0002] In the field of interactive content generation, especially in intelligent recommendation systems driven by AIGC (Artificial Intelligence Generated Content) technology, existing technologies usually rely on static features such as user portraits, tag information, or click behaviors, and combine rule engines or lightweight models for content screening and pushing. Although such methods have a certain recommendation efficiency, the complexity level of the generated content is often controlled by fixed templates or preset parameters, lacking the ability to dynamically respond to the user's current true cognitive state, and it is difficult to meet the adaptability requirements of personalized recommendations in complex scenarios.

[0003] Some existing systems have tried to introduce real-time user behavior data, such as trajectory location, page stay time, etc. to optimize content recommendation strategies, but there are still two common problems: one is that it is impossible to accurately extract the user's preference trend for content complexity from the behavior sequence, and the other is that it fails to combine the changes in the user's physiological state to judge their cognitive load or acceptance ability. As a result, in the process of generating content, there is often a mismatch between the pushed content and the user's state, resulting in the content being too complex or too simple, causing user cognitive pressure, bounce behavior, or a decline in interaction efficiency. Summary of the Invention

[0004] The purpose of the present invention is to provide a retail customer interaction content generation and push system based on AIGC, aiming to solve the problems raised in the background art.

[0005] The present invention is implemented as follows. A retail customer interaction content generation and push system based on AIGC, the system includes: an information collection module, a trajectory recognition module, a behavior association determination module, a correction parameter generation module, and a content complexity regulation module, wherein: The system is built based on an AIGC model, used to dynamically generate personalized recommendation content according to the user's state, and the generation process of the recommendation content is controlled by the content complexity level; The information collection module is used to obtain the behavior trajectory data, wearable device data, and the original content complexity level of the recommended content within a preset time period before the node when the user enters the experience space and triggers the recommended content push node; The trajectory recognition module is used to identify several target areas browsed by the user based on the behavior trajectory data, and extract the concentrated fixation time in each target area and the corresponding area interaction complexity value of each target area; A behavior correlation determination module, which is used to sort the regional interaction complexity values from small to large, construct an interaction complexity sequence and a corresponding fixation time sequence, and determine whether the change trends of the two conform to a preset behavior trend pattern; A correction parameter generation module, which is used to generate a correction factor based on the change trend of the fixation time sequence under the condition that the behavior trend pattern is determined to be met, and generate an adjustment weight based on the analysis of the user's physiological state change trend from wearable device data, and apply the adjustment weight to the correction factor to obtain a final correction parameter; A content complexity regulation module, which is used to dynamically correct the original content complexity level by using the final correction parameter, and use the corrected content complexity level to control the expression level, language style or information structure of the recommended content generated based on the AIGC model.

[0006] As a further limitation of the technical solution of the embodiment of the present invention, the experience space is pre-divided into several independent pre-divided regions, and a regional interaction complexity value is set for each pre-divided region, and the target region is a pre-divided region where the user actually stays and the stay time reaches a preset threshold during the browsing process.

[0007] As a further limitation of the technical solution of the embodiment of the present invention, the trajectory recognition module specifically includes: A video data analysis unit, which is used to analyze the behavior trajectory data based on image recognition and target tracking technologies, and identify several target regions browsed by the user; A fixation time determination unit, which is used to determine the concentrated fixation time of the user in each target region through head pose estimation and line-of-sight direction analysis technologies, and associate the regional interaction complexity value corresponding to the target region.

[0008] As a further limitation of the technical solution of the embodiment of the present invention, the behavior correlation determination module specifically includes: A data sorting unit, which is used to sort the regional interaction complexity values corresponding to several target regions from small to large, construct an interaction complexity sequence, and sequentially extract the concentrated fixation time corresponding to each target region in the sequence to form a fixation time sequence; A behavior correlation determination unit, which is used to analyze the change trend of the fixation time sequence and determine whether it shows a pattern of gradually increasing or gradually decreasing. If so, it is determined that the change trends between the interaction complexity sequence and the fixation time sequence conform to the preset behavior trend pattern.

[0009] As a further limitation of the technical solution of the embodiment of the present invention, the correction parameter generation module specifically includes: A correction factor determination unit, configured to calculate an average slope of the fixation time series and use the average slope as a correction factor when it is determined that the change trend between the interaction complexity series and the fixation time series conforms to a preset behavior trend pattern; A state trend acquisition unit, configured to extract heart rate variability data or skin conductance response data of a user within a preset time period based on wearable device data, perform time series arrangement on the data, and obtain a physiological state change trend of the user with the change of browsing time; A final parameter determination unit, configured to calculate an average slope of the physiological state change trend, use it as an adjustment weight to act on the correction factor, and obtain a final correction parameter.

[0010] As a further limitation of the technical solution of the embodiment of the present invention, the content complexity regulation module specifically includes: A complexity level correction unit, configured to retrieve a preset complexity level correction model, and dynamically adjust the original content complexity level in combination with the final correction parameter to obtain a corrected content complexity level; A complexity level application unit, configured to apply the corrected content complexity level to the generation process of recommended content to control the semantic depth, information density or expression structure of the generated content.

[0011] As a further limitation of the technical solution of the embodiment of the present invention, the complexity level correction model is: , where refers to the corrected content complexity level, refers to the original content complexity level, refers to the final correction parameter, refers to an adjustment coefficient of the final correction parameter, used to control the sensitivity or amplification ratio of the correction amplitude; In the complexity level correction model, , where refers to the average slope of the physiological state change trend, the average slope of the fixation time series.

[0012] Compared with the prior art, the present invention has the following beneficial effects: The present invention constructs a content complexity level correction mechanism jointly driven by the gaze time trend and the physiological state change trend by integrating the user behavior trajectory and the wearable device data. Different from the existing method of setting the recommendation strategy only based on behavior tags or single-dimensional features, the present invention introduces a final correction parameter to dynamically adjust the original content complexity level, so that the generated content is more in line with the user's current attention intention and cognitive state in terms of the expression level, information density and logical structure. This mechanism can avoid the cognitive burden or insufficient information caused by too high or too low content complexity while maintaining the recommendation effect, and achieve more accurate and flexible personalized generation control. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is the application architecture diagram of the system provided by the embodiment of the present invention; Figure 2 is the structural block diagram of the trajectory recognition module in the system provided by the embodiment of the present invention; Figure 3 is the structural block diagram of the behavior association determination module in the system provided by the embodiment of the present invention; Figure 4 is the structural block diagram of the correction parameter generation module in the system provided by the embodiment of the present invention; Figure 5 is the structural block diagram of the content complexity regulation module in the system provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0015] Furthermore, Figure 1 shows the application architecture diagram of the system provided by the embodiment of the present invention.

[0016] Among them, in another preferred embodiment provided by the present invention, a retail customer interaction content generation and push system based on AIGC includes: The system is constructed based on an AIGC model and is used to dynamically generate personalized recommendation content according to the user state, and the generation process of the recommendation content is controlled by the content complexity level.

[0017] An information collection module 100, configured to obtain the behavior trajectory data, wearable device data, and the original content complexity level of the recommendation content within a preset time period before the node when the user enters the experience space and triggers the recommendation content push node.

[0018] In the embodiments of the present invention, the experience space refers to the physical retail scenario where users can enter and independently browse, interact, and stay, usually an offline store environment equipped with multi-type product display areas, visual information terminals, and behavior perception devices. Such places allow users to contact products, observe information, and generate behavioral feedback in a natural behavior state, with obvious immersive and interactive retail characteristics, such as smart home experience halls, comprehensive digital device display halls, future retail unmanned stores, etc.

[0019] The user group targeted by the present invention is the experiencers who have wearable devices and have authorized the system to access their physiological state data. Such users usually wear smart bracelets, watches, or portable devices with physiological data sensing functions, and can continuously collect physiological data such as heart rate variability and skin conductance response during the experience process and upload them to the experience system. When designing this system, it is assumed that the data access of such users is clearly authorized and legally used, ensuring that subsequent push strategies and content adjustment are based on the real state of the users.

[0020] The so-called "triggering the recommended content push node" refers to the logical time point when the system automatically starts the content push process when the user's behavior in the experience space reaches a certain condition threshold. This node usually includes but is not limited to the following triggering methods: the user continuously stays in a certain area for more than the set time, completes the interactive browsing of multiple devices, the movement path in the store satisfies a certain high-attention trajectory pattern, actively wakes up the voice guide assistant or scans the code to log in to the system, etc. In the prior art, similar triggering mechanisms widely exist in scenarios such as trajectory-based recommendation, commodity hot zone push, or immersive retail experience, such as the "stay + gaze linkage" advertising guidance system in some smart stores.

[0021] The preset time period refers to the reference time window used by the system to analyze the user's state before the triggering node occurs, usually between 2 and 5 minutes, set according to the average browsing cycle of users in the actual scenario, which can not only cover rich enough behavioral trajectories but also avoid introducing overly old data resulting in delayed state judgment. The specific time period length can be dynamically set by factors such as operation strategies, scenario density, and the number of products.

[0022] The original content complexity level of the recommended content refers to the initial complexity index given to the recommended content by the system according to the user profile, historical behavior, or default template before content push. For example, this level can be represented by numerical values for parameters such as the information level of the generated content, the number of professional terms, and the depth of the logical structure. This index has been widely used in the prior art, especially in AIGC content generation and natural language summary recommendation, where there are already prior parameter setting methods for controlling the style and difficulty of the generated content, usually achieved through methods such as Prompt template classification, generation temperature parameter regulation, and guiding word selection.

[0023] Behavior trajectory data can be obtained in real time through image acquisition devices and vision recognition algorithms (such as object tracking, multi-camera fusion, path mapping) deployed in the experience space. The data content includes the user's displacement path, staying area, gaze direction, head posture, and regional staying duration, etc. Wearable device data relies on the devices worn by users and is transmitted to the system through Bluetooth, Wi-Fi, or other communication protocols. The content usually includes heart rate, HRV (heart rate variability), skin conductance response, skin temperature, activity frequency, etc. These two types of data together constitute the multi-modal input basis of the system, which is used to support the judgment and adjustment of subsequent content push strategies.

[0024] Furthermore, the AIGC-based retail customer interaction content generation and push system further includes: A trajectory recognition module 200, configured to identify a plurality of target areas browsed by a user based on behavior trajectory data, and extract the concentrated gaze time of the user in each target area and the regional interaction complexity value corresponding to each target area.

[0025] The experience space is pre-divided into a plurality of independent pre-divided areas, and a regional interaction complexity value is set for each pre-divided area. The target area is a pre-divided area where the user actually stays and the staying time reaches a preset threshold during the browsing process.

[0026] Specifically, Figure 2 The structure block diagram of the trajectory recognition module 200 in the system provided by the embodiment of the present invention is shown.

[0027] Among them, in the preferred implementation manner provided by the present invention, the trajectory recognition module 200 specifically includes: A video data parsing unit 201, configured to parse behavior trajectory data based on image recognition and object tracking technologies, and identify a plurality of target areas browsed by a user; A gaze time determination unit 202, configured to determine the concentrated gaze time of the user in each target area through head posture estimation and line-of-sight direction analysis technologies, and associate the regional interaction complexity value corresponding to the target area.

[0028] In the embodiments of the present invention, the regional interaction complexity value is a predefined complexity index for the devices or interaction contents configured for each pre-divided area in the experience space. This value can be comprehensively evaluated based on factors such as the functional type of the display devices in the area, the number of operation steps, the length of the interaction process, and the complexity of the information presentation method, and scored manually or algorithmically. Such complexity modeling belongs to the category of existing technologies and is widely used in fields such as human factors engineering, user experience evaluation, and retail display layout optimization. For example, in a smart retail or smart home exhibition hall, areas with higher complexity may involve smart panels and multi-functional linkage devices that require users to complete multi-step operations, while areas with lower complexity may only include passive display induction lights, visual introduction terminals, etc. The regional interaction complexity value can be preset by the operator during the system deployment phase, or automatically generated based on an expert knowledge system to build a weight model.

[0029] The setting of the preset threshold is used to determine whether the user has actual interactive attention to a certain area. Specifically, whether the time the user stays in a certain pre-divided area exceeds this threshold is the basis for the system to judge whether the user enters the "target area". The significance of setting this threshold is to eliminate the noise data brought by the user's unconscious passing by or brief glance, and only retain the cognitive or intentional staying behavior, enhancing the effectiveness of the trajectory recognition result. The preset threshold can be set by the system through learning historical data, or configured according to the average residence behavior experience of different display areas, usually fluctuating between 2 and 5 seconds.

[0030] The technical means adopted by the video data analysis unit 201 mainly include the behavior trajectory extraction method based on image recognition and target tracking. Specifically, the system deploys multiple cameras in the experience space, identifies the user targets entering the field of view through multi-object detection (such as based on models such as YOLO and Faster R-CNN), and continuously tracks the position information of the user based on the target tracking algorithm (such as SORT and Deep SORT), so as to construct the trajectory path of the user in the space. Combining the path information with the preset area mapping relationship in the experience space can determine whether the user enters a certain area and record the start and end times of their stay.

[0031] The fixation time determination unit 202 determines the concentrated fixation time of the user in each target area through head pose estimation and line-of-sight direction analysis techniques. The specific method includes obtaining the head pose angle based on facial key point recognition (such as MediaPipe, OpenFace, etc.), estimating the included angle range between the head rotation angle and the plane orientation of the area, combining the fixation direction aggregation of frame continuity, determining whether the user is in a fixation state in this area, and accumulating the fixation duration. If the fixation time continuously meets a certain threshold, it is considered that this area enters the effective attention range and is bound to the area interaction complexity value for subsequent behavior trend analysis and parameter correction module processing.

[0032] Furthermore, the AIGC-based retail customer interaction content generation and push system further includes: A behavior correlation determination module 300, which is used to sort the area interaction complexity values from small to large, construct an interaction complexity sequence and the corresponding fixation time sequence, and determine whether the change trends of the two conform to a preset behavior trend pattern.

[0033] Specifically, Figure 3 The structure block diagram of the behavior correlation determination module 300 in the system provided by the embodiment of the present invention is shown.

[0034] Among them, in the preferred embodiment provided by the present invention, the behavior correlation determination module 300 specifically includes: A data sorting unit 301, which is used to sort the area interaction complexity values corresponding to several target areas from small to large in numerical value, construct an interaction complexity sequence, and sequentially extract the concentrated fixation time corresponding to each target area in this sequence to form a fixation time sequence; A behavior correlation determination unit 302, which is used to analyze the change trend of the fixation time sequence and determine whether it shows a pattern of gradually increasing or gradually decreasing. If so, it is determined that the change trends between the interaction complexity sequence and the fixation time sequence conform to a preset behavior trend pattern.

[0035] In the embodiment of the present invention, the implementation process of the data sorting unit 301 mainly includes the following steps: First, the system receives the area interaction complexity values of multiple target areas and their corresponding concentrated fixation time data from the trajectory recognition module; Subsequently, these target areas are arranged in ascending order according to the numerical values of their area interaction complexity values to construct an interaction complexity sequence; Immediately afterwards, the system sequentially extracts the concentrated fixation time values corresponding to each area according to this sorting order to generate a fixation time sequence. During the sorting process, the system can ensure the consistency of the sorting index based on the bivariate pairing method, that is, the interaction complexity sequence and the fixation time sequence always correspond one by one. The sorting operation can be implemented by quick sorting, merge sorting with higher stability, or other structured sorting methods based on key-value pairs.

[0036] In the behavior correlation determination unit 302, the changing trend of the fixation time series does not require strict monotonic increase or decrease in numerical value. Instead, it is judged whether it shows an increasing or decreasing trend "overall" through the fitting result of the overall trend. In specific implementation, the system can adopt methods such as linear regression, locally weighted regression, or moving average to analyze the trend of the fixation time series, allowing a certain degree of local fluctuation or individual outliers in the series. As long as the overall fitting curve or fitting slope shows a positive or negative direction, it can be determined as a changing trend of "gradually increasing" or "gradually decreasing". This setting improves the robustness of the system in dealing with discreteness and volatility in actual behavior data and avoids misjudgment caused by over-sensitivity.

[0037] The behavior correlation determination unit 302 evaluates whether the corresponding relationship with the interaction complexity series has a consistent or reverse trend characteristic by analyzing the overall changing trend of the fixation time series. If the fitting trend between the two shows a stable positive or negative correlation, that is, as the regional interaction complexity value gradually increases, the fixation time generally shows a corresponding increasing or decreasing tendency, the system can identify that this behavior sequence has shown an interpretable behavior trend pattern accordingly.

[0038] The recognition of this trend relationship plays an important role. It reflects the acceptance characteristics and attention preferences of users for content of different complexities during browsing. For example, when the fixation time gradually extends in a high-complexity area, it may mean that the user has a high willingness and interest in information-intensive content; on the contrary, if the fixation time significantly shortens in an area with increasing complexity, it may indicate that the user has a certain degree of cognitive load or avoidance reaction. Through the above trend matching process, the system can extract the potential preference logic in the user behavior characteristics and provide a logical basis and judgment premise for the generation of subsequent correction factors, further improving the adaptability and accuracy of personalized content complexity regulation.

[0039] Furthermore, the AIGC-based retail customer interaction content generation and push system further includes: A correction parameter generation module 400, configured to generate a correction factor based on the changing trend of the fixation time series under the condition of determining that the behavior trend pattern is met, and analyze the changing trend of the user's physiological state based on the wearable device data to generate an adjustment weight and act on the correction factor to obtain a final correction parameter.

[0040] Specifically, Figure 4 FIG. shows the structural block diagram of the correction parameter generation module 400 in the system provided by the embodiment of the present invention.

[0041] Among them, in the preferred implementation manner provided by the present invention, the correction parameter generation module 400 specifically includes: A correction factor determination unit 401, configured to calculate an average slope of the fixation time series and use the average slope as a correction factor when it is determined that the change trend between the interaction complexity series and the fixation time series conforms to a preset behavior trend pattern; A state trend acquisition unit 402, configured to extract heart rate variability data or skin conductance response data of a user within a preset time period based on wearable device data, perform time series arrangement on the data, and obtain a physiological state change trend of the user as the browsing time changes; A final parameter determination unit 403, configured to calculate an average slope of the physiological state change trend, use it as an adjustment weight to act on the correction factor, and obtain a final correction parameter.

[0042] In an embodiment of the present invention, when the system identifies a trend relationship that meets preset conditions between the interaction complexity series and the fixation time series, it can be considered that the user has shown a certain behavior tendency towards the regional interaction complexity during the browsing process. On this basis, using the average slope of the fixation time series as a correction factor can quantify the change trend of the user's attention degree to the content of different complexity regions as a whole. This slope reflects the sensitivity of the fixation time to the increase in complexity, that is, the larger the slope value, the more obvious the increase in the user's stay time in the complex content region, indicating a higher willingness to process information; conversely, if the slope tends to be flat or negative, it means that the user's attention investment in complex content is limited, with characteristics of cognitive resistance or interest attenuation. Through this correction factor, the system can initially evaluate whether the complexity of the current recommended content needs to be adjusted upward or downward, providing a basic judgment for realizing personalized regulation.

[0043] In the specific implementation process of the state trend acquisition unit 402, by accessing the data stream of the user's wearable device, physiological indicators closely related to the cognitive state within the preset time period are extracted, including but not limited to heart rate variability (HRV) data and skin conductance response (EDA) data. The system arranges and regularizes this type of data in chronological order, eliminates outliers, and uses means such as moving window smoothing to generate a stable time series curve. Then, the system calculates the physiological state change trend of the user during the browsing process based on this time series data, for example, extracts the change direction and change rate through linear regression or locally weighted regression. If the HRV shows a continuous downward trend or the skin conductance continuously rises, it usually reflects that the user is in a state of high cognitive load or stress; conversely, if the HRV rebounds or the EDA slows down, it indicates that the user's state tends to be stable or relaxed.

[0044] Further, the system uses the average slope of the change trend curve as the core parameter to measure the change trend of the physiological state. Among them, the larger the slope value, the faster the improvement speed of the user's state and the stronger the cognitive bearing capacity; the smaller the slope value or even negative, it indicates that the user is in the process of state deterioration or increasing fatigue. This slope not only reflects the quality of the state itself, but also reflects the rate characteristics of its change trend, providing a quantitative basis for the subsequent adjustment of the correction parameter.

[0045] The final parameter determination unit 403 uses the average slope of the above-mentioned physiological state change trend as the adjustment weight to act on the correction factor, aiming to appropriately amplify or suppress the behavior preference signal according to the user's current psychological bearing capacity. When the physiological trend indicates that the user's pressure is gradually increasing, the system reduces the action amplitude of the correction factor to avoid cognitive interference caused by too high content complexity; when the physiological trend indicates that the state is stable or even improving, the system can appropriately amplify the action amplitude of the correction factor to guide the generation of higher-density or hierarchical content. This adjustment method based on the fusion of behavioral and physiological dual states can reflect the balance relationship between the user's willing preference for complex information and the immediate acceptable ability, so as to ensure that the content complexity adjustment not only conforms to interest guidance but also does not exceed the psychological load threshold, realizing a more accurate and user-friendly recommendation and push strategy.

[0046] Furthermore, the AIGC-based retail customer interaction content generation and push system further includes: A content complexity regulation module 500, which is used to dynamically correct the original content complexity level by using the final correction parameter, and use the corrected content complexity level to control the expression level, language style or information structure of the recommended content generated based on the AIGC model.

[0047] Specifically, Figure 5 shows the structural block diagram of the content complexity regulation module 500 in the system provided by the embodiment of the present invention.

[0048] Among them, in the preferred embodiment provided by the present invention, the content complexity regulation module 500 specifically includes: A complexity level correction unit 501, which is used to retrieve the preset complexity level correction model and dynamically adjust the original content complexity level in combination with the final correction parameter to obtain the corrected content complexity level; A complexity level application unit 502, which is used to apply the corrected content complexity level to the generation process of the recommended content to control the semantic depth, information density or expression structure of the generated content.

[0049] The complexity level correction model is: , where refers to the corrected content complexity level, Refers to the original content complexity level, Refers to the final correction parameter, Refers to the adjustment coefficient of the final correction parameter, which is used to control the sensitivity or amplification ratio of the correction amplitude; In the complexity level correction model, , where Refers to the average slope of the change trend of the physiological state, The average slope of the fixation time series.

[0050] In the embodiments of the present invention, this case is mainly applied to the intelligent retail recommendation scenario driven by AIGC, and is particularly suitable for immersive and interactive experience spaces, such as smart home experience halls, digital device exhibition halls or future unmanned stores. The system faces the user group with wearable devices, and by collecting their behavior trajectories and physiological state data, dynamically identifies their preference trends and cognitive load changes for the content in different regions, so as to realize the personalized adjustment of the complexity of the recommended content.

[0051] The core technical problem solved by the present invention is that existing content recommendation systems usually set push strategies only based on historical behavior tags or single behavior events, lacking the comprehensive perception ability of users' real-time interest evolution and psychological state, resulting in the risk that the generated content deviates from the current bearing capacity of users in terms of complexity setting. Especially in scenarios where content generation depends on the output of large models, if the pushed content is too dense, the terms are too complex or the structure is too heavy, it is easy to cause user fatigue, bounce or give up reading. Therefore, how to establish an adaptive complexity control mechanism based on users' immediate behavior patterns and psychological states has become an important technical problem addressed in this case.

[0052] The application prospect of this technical solution is broad, and it can be extended to various content generation scenarios for people, such as intelligent question-and-answer recommendation systems, personalized advertisement generation platforms, voice shopping guide systems, augmented reality interaction experiences, etc. By integrating behavior trend analysis and physiological state perception, the system can dynamically adjust the expression structure, information depth and push rhythm of the generated content with higher precision, improve user acceptance, optimize the content matching effect, thereby enhancing the intelligence of the AIGC system and the naturalness of human-computer interaction, and having significant commercial value and social application potential.

[0053] The complexity level correction model adopted is an intuitive and effective implementation method. By introducing a proportional correction mechanism on the basis of the original complexity level, flexible regulation can be achieved. This model has the advantages of clear structure, lightweight operation, and easy deployment. However, in practical applications, other forms of dynamic adjustment strategies can also be extended and adopted. For example, an exponential response model can be introduced to perform non-linear scaling on the final correction parameters; or the complexity distribution can be dynamically adjusted based on the Bayesian update strategy; or a higher-order complexity prediction and control module can be constructed by introducing the joint fitting of multiple variables through a deep learning model. These alternative solutions can be flexibly selected according to the business scenario, user type, and model deployment resources.

[0054] The meaning represented by the calculation formula of the final correction parameter is as follows: When the average slope of the fixation time series is positive and the average slope of the physiological state change trend is also positive, it indicates that the user's interest in complex content is increasing, and the physiological state tends to be good, with strong cognitive bearing capacity. At this time, the system calculates the correction parameter by multiplying the fixation slope by one plus the physiological state slope, so as to actively amplify the content complexity level and promote the generation and push of higher-quality information.

[0055] When the slope of the fixation time series is positive while the average slope of the physiological state change trend is negative, it shows that although the user remains interested in complex content, their state is tending to fatigue or the cognitive load is increasing. To avoid overly aggressive content pushing, the system adopts a conservative correction method of multiplying the fixation slope by one minus the absolute value of the physiological state slope to control the increase amplitude of the content complexity, relieve the user's pressure while ensuring the pushing quality.

[0056] When the slope of the fixation time series is zero or negative and the average slope of the physiological state change trend is negative, it indicates that the user has a low attention to complex content and the current state is on a downward trend. At this time, the system adopts the method of multiplying the fixation slope by one plus the absolute value of the physiological state slope to significantly lower the complexity level, quickly relieve the cognitive load, and prevent interference caused by content mismatch.

[0057] When the slope of the fixation time series is zero or negative and the average slope of the physiological state change trend is positive, it indicates that although the user has little current interest in complex content, their state is good and they have a certain acceptance ability. In this case, the system adopts the method of multiplying the fixation slope by one minus the physiological state slope to moderately adjust the complexity level, avoid misjudging the pushing strategy due to short-term fluctuations in behavior, and achieve a more gentle reduction in complexity.

[0058] Through the above-mentioned segmented control mechanism, the system can flexibly adapt to different user states and behavioral characteristics, achieve precise adjustment of the content complexity level, and while ensuring the push effect, improve the stability and personalized adaptation ability of the user experience.

[0059] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0060] The above-described embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it cannot be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention should be subject to the appended claims.

[0061] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A retail customer interaction content generation and push system based on AIGC, characterized in that, The system includes: an information collection module, a trajectory recognition module, a behavior association determination module, a correction parameter generation module, and a content complexity regulation module, where: The system is constructed based on an AIGC model and is used to dynamically generate personalized recommendation content according to the user's state. The generation process of the recommendation content is controlled by the content complexity level. The information collection module is used to obtain the behavioral trajectory data, wearable device data, and the original content complexity level of the recommendation content within a preset time period before the node when the user enters the experience space and triggers the recommendation content push node. The trajectory recognition module is used to identify several target areas browsed by the user based on the behavioral trajectory data, and extract the concentrated fixation time of the user in each target area and the corresponding area interaction complexity value of each target area. The behavior association determination module is used to sort the area interaction complexity values from small to large, construct an interaction complexity sequence and the corresponding fixation time sequence, and determine whether the change trends of the two conform to the preset behavior trend pattern. The correction parameter generation module is used to generate a correction factor based on the change trend of the fixation time sequence under the condition that the behavior trend pattern is determined to be met, and analyze the change trend of the user's physiological state based on the wearable device data, generate an adjustment weight and apply it to the correction factor to obtain the final correction parameter. The content complexity regulation module is used to dynamically correct the original content complexity level using the final correction parameter, and use the corrected content complexity level to control the expression level, language style, or information structure of the recommendation content generated based on the AIGC model.

2. The AIGC-based retail customer interaction content generation and push system according to claim 1, wherein The experience space is pre-divided into several independent pre-divided areas, and a corresponding area interaction complexity value is set for each pre-divided area. The target area is a pre-divided area where the user actually stays and the stay time reaches the preset threshold during the browsing process.

3. The AIGC-based retail customer interaction content generation and push system according to claim 2, wherein The trajectory recognition module specifically includes: The video data analysis unit is used to analyze the behavioral trajectory data based on image recognition and target tracking technologies, and identify several target areas browsed by the user. The fixation time determination unit is used to determine the concentrated fixation time of the user in each target area through head pose estimation and gaze direction analysis technologies, and associate the corresponding area interaction complexity value of the target area.

4. The AIGC-based retail customer interaction content generation and push system according to claim 3, wherein The behavior association determination module specifically includes: The data sorting unit is used to sort the area interaction complexity values corresponding to several target areas from small to large, construct an interaction complexity sequence, and sequentially extract the concentrated fixation time corresponding to each target area in the sequence to form a fixation time sequence. The behavior association determination unit is used to analyze the change trend of the fixation time sequence and determine whether it shows a gradually increasing or gradually decreasing pattern. If so, it is determined that the change trends between the interaction complexity sequence and the fixation time sequence conform to the preset behavior trend pattern.

5. The AIGC-based retail customer interaction content generation and push system according to claim 4, wherein The correction parameter generation module specifically includes: The correction factor determination unit is used to calculate the average slope of the fixation time sequence and use the average slope as the correction factor when it is determined that the change trends between the interaction complexity sequence and the fixation time sequence conform to the preset behavior trend pattern. A status trend acquisition unit, configured to extract the heart rate variability data or the galvanic skin response data of a user within a preset time period based on the wearable device data, perform time series arrangement on the data, and obtain the physiological state change trend of the user along with the browsing time change; A final parameter determination unit, configured to calculate the average slope of the physiological state change trend, use the slope as an adjustment weight to act on a correction factor, and obtain a final correction parameter.

6. The AIGC-based retail customer interaction content generation and push system according to claim 1, wherein The content complexity regulation module specifically includes: A complexity level correction unit, configured to retrieve a preset complexity level correction model, and dynamically adjust the original content complexity level in combination with the final correction parameter to obtain a corrected content complexity level; A complexity level application unit, configured to apply the corrected content complexity level to the generation process of the recommended content to control the semantic depth, information density or expression structure of the generated content.

7. The AIGC-based retail customer interaction content generation and push system according to claim 6, wherein, The complexity level correction model is as follows: , where refers to the corrected content complexity level, refers to the original content complexity level, refers to the final correction parameter, refers to the adjustment coefficient of the final correction parameter, which is used to control the sensitivity or amplification ratio of the correction amplitude; In the complexity level correction model, , where refers to the average slope of the change trend of the physiological state, the average slope of the fixation time series.

Images