Cross-border e-commerce commodity intelligent recommendation method and system based on AI analysis

Abstract

The invention is suitable for the technical field of cross fusion of artificial intelligence and cross-border e-commerce technology, and provides an intelligent cross-border e-commerce commodity recommendation method and system based on AI analysis, and the method comprises the steps: obtaining the monitoring video data of a target pet, and the initial pet emotional response fitness score of a to-be-recommended product; and analyzing the monitoring video data, identifying whether user intervention exists when the target pet uses other products similar to the to-be-recommended product, and if so, obtaining a reference product most consistent with the background information of the to-be-recommended product and a reference video corresponding to the reference product. According to the method, a pet behavior video analysis and reference model comparison mechanism is introduced, so that the initial emotion response fitness score of the to-be-recommended product is dynamically corrected, and the problem that an existing recommendation system cannot accurately recognize the actual adaptability of the pet is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of cross-border e-commerce and artificial intelligence, and particularly relates to a cross-border e-commerce commodity intelligent recommendation method and system based on AI analysis. BACKGROUND

[0002] At present, in the recommendation process of pet commodities, especially pet emotion improvement products, the cross-border e-commerce platform mainly relies on static information such as user historical purchase records, keyword tags and basic scoring system to match and sort the content. This kind of recommendation method usually focuses on the shopping behavior and subjective evaluation of human users, but it is difficult to touch the individual differences of pets as the final use subject in emotional response and adaptability. Especially for emotion improvement products, their actual effect is highly dependent on the coupling of pet's own state and environmental conditions. The traditional recommendation mechanism based only on product description and user score is difficult to accurately reflect the real adaptability of the target pet, which is easy to cause the recommended result to deviate from the actual demand, affect the user's trust and platform stickiness.

[0003] The existing intelligent recommendation system generally lacks dynamic perception and analysis mechanism for pet behavior, and does not include the user's auxiliary intervention action in the product use process into the judgment basis. In fact, in the use scene of emotion products, the reaction of pets is often directly affected by the user's intervention behavior (such as pacification and guidance), which will significantly change the behavior feedback of pets, and then interfere with the credibility of the score, finally leading to the generation of product recommendation weight factor not being objective enough. However, the existing system usually ignores this factor, so that the evaluation model cannot distinguish between active adaptation and passive cooperation of emotional response. SUMMARY

[0004] The purpose of the present application is to provide a cross-border e-commerce commodity intelligent recommendation method and system based on AI analysis, which aims to solve the problems proposed in the background art.

[0005] The present application is implemented as follows: a cross-border e-commerce commodity intelligent recommendation method based on AI analysis, the method comprising:

[0006] obtaining monitoring video data of a target pet, and an initial pet emotion response adaptability score of a product to be recommended;

[0007] analyzing the monitoring video data to identify whether there is user intervention when the target pet uses other products of the same type as the product to be recommended, if there is, obtaining a reference product most consistent with the background information of the product to be recommended and its corresponding reference video;

[0008] Identify the intervention intensity in the reference video and the reference response behavior score of the target pet, call the reference model to obtain the reference standard response behavior score under the same condition, calculate the reference deviation amplitude between the two, and generate a correction factor according to the reference deviation amplitude;

[0009] According to the correction factor, the initial pet emotional response fitness score is corrected to obtain a corrected pet emotional response fitness score, and the corrected pet emotional response fitness score is applied to the calculation of the recommendation weight factor of the product to be recommended.

[0010] As a further limitation of the technical scheme of the embodiment of the application, the specific category of the product to be recommended is a pet emotion improvement product, including a soothing spray, a soothing therapy pad, an emotion guiding toy or a smart device for relieving anxiety.

[0011] As a further limitation of the technical scheme of the embodiment of the application, the step of analyzing the monitoring video data and identifying whether the target pet exists user intervention when using other products of the same category as the product to be recommended includes:

[0012] Analyzing the monitoring video data and extracting a plurality of sample video clips of the target pet using pet emotion improvement products;

[0013] Analyzing each sample video, identifying whether the target pet exists user intervention action in the process of contacting the corresponding product, if the intervention action is detected and the intervention duration exceeds the preset threshold, the sample video is marked as existing user intervention;

[0014] If the proportion of sample videos with user intervention to all sample videos exceeds the preset proportion, it is determined that the target pet exists user intervention when using other products of the same category as the product to be recommended, and the reference product and its corresponding reference video most consistent with the background information of the product to be recommended are extracted;

[0015] The reference product most consistent with the background information of the product to be recommended refers to the pet emotion improvement product most close to the product to be recommended in terms of product function, applicable emotion category and use situation and other factors.

[0016] As a further limitation of the technical scheme of the embodiment of the application, the reference model is a structured behavior score model constructed based on big data, which indexes and clusters samples according to pet species, gender, age, intervention intensity and other attribute dimensions, and presets standard response behavior score values under different condition combinations.

[0017] As a further limitation of the technical scheme of the embodiment of the present application, the step of identifying the intervention intensity in the reference video and the reference response behavior score of the target pet, calling the reference model to obtain the reference standard response behavior score under the same condition, calculating the reference deviation amplitude between the two, and generating the correction factor according to the reference deviation amplitude comprises:

[0018] Extracting the image sequence of the user intervention stage in the reference video, identifying the intervention action amplitude, duration and frequency of the user, and performing linear weighted combination or outputting through a trained model based on the above parameters to generate the intervention intensity;

[0019] Analyzing the behavior reaction characteristics of the target pet after the user intervention, including gaze duration, approach action, tail movement amplitude, posture stability, and combining the delay time and duration of the behavior reaction to generate the reference response behavior score through a weighted rule or a scoring mechanism based on a trained model;

[0020] Calling a pre-established reference model to match the standard response behavior score corresponding to the sample condition consistent with the species, gender, age and intervention intensity of the target pet;

[0021] Calculating the reference deviation amplitude between the reference response behavior score and the standard response behavior score, and generating the correction factor according to the reference deviation amplitude.

[0022] As a further limitation of the technical scheme of the embodiment of the present application, when correcting the initial pet emotional response adaptability score, a preset score correction model is used;

[0023] The score correction model is: wherein refers to the corrected pet emotional response adaptability score, refers to the initial pet emotional response adaptability score, refers to the correction factor, refers to the corresponding adjustment coefficient, and is greater than 0;

[0024] In the score correction model, wherein refers to the reference response behavior score, refers to the reference standard response behavior score, refers to the reference deviation amplitude.

[0025] An AI analysis-based cross-border e-commerce commodity intelligent recommendation system, the system comprising: a data acquisition module, a reference data acquisition module, a correction factor determination module, and a score correction module, wherein:

[0026] The data acquisition module is configured to acquire monitoring video data of the target pet and an initial pet emotional response fitness score of a product to be recommended; the specific category of the product to be recommended is a pet emotional improvement product, including a calming spray, a soothing therapy pad, an emotional guidance toy, or a smart device for relieving an anxious state.

[0027] The reference data acquisition module is configured to analyze the monitoring video data, identify whether the target pet has user intervention when using other products of the same category as the product to be recommended, and if so, acquire a reference product most consistent with the background information of the product to be recommended and a corresponding reference video of the reference product.

[0028] The correction factor determination module is configured to identify the intervention intensity in the reference video and a reference response behavior score of the target pet, call a reference model to acquire a reference standard response behavior score under the same condition, calculate a reference deviation amplitude between the two, and generate a correction factor according to the reference deviation amplitude; the reference model is a structured behavior score model constructed based on big data, which indexes and clusters samples according to attribute dimensions such as pet category, gender, age, and intervention intensity, and predefines standard response behavior score values under different condition combinations.

[0029] The score correction module is configured to correct the initial pet emotional response fitness score according to the correction factor, to obtain a corrected pet emotional response fitness score, and apply the corrected pet emotional response fitness score in the calculation of a recommendation weight factor of the product to be recommended.

[0030] As a further limitation of the technical scheme of the embodiments of the present application, the reference data acquisition module specifically includes:

[0031] The sample video extraction unit is configured to analyze the monitoring video data and extract a plurality of sample video clips in which the target pet uses pet emotional improvement products.

[0032] The intervention behavior judgment unit is configured to analyze each sample video and identify whether the target pet has user intervention actions during the process of contacting the corresponding product; if the intervention actions are detected and the intervention duration exceeds a preset threshold, the sample video is marked as having user intervention.

[0033] The reference video extraction unit is configured to determine that the target pet has user intervention when using other products of the same category as the product to be recommended if the proportion of sample videos with user intervention among all sample videos exceeds a preset proportion, and extract a reference product most consistent with the background information of the product to be recommended and a corresponding reference video of the reference product.

[0034] The reference product most consistent with the background information of the product to be recommended refers to a pet emotional improvement product that is closest to the product to be recommended in terms of product function, applicable emotional category, and use situation.

[0035] As a further limitation of the technical solutions of the embodiments of the present application, the correction factor determination module specifically comprises:

[0036] An intervention intensity generation unit is configured to extract an image sequence of a user intervention stage in a reference video, identify the intervention action amplitude, duration and frequency of the user, and perform linear weighted combination or output through a trained model based on the above parameters to generate intervention intensity;

[0037] A reference score generation unit is configured to analyze the behavior reaction characteristics of the target pet after the user intervention, including gaze duration, approach action, tail movement amplitude, posture stability and other parameters, and combine the delay time and duration of the behavior reaction to generate a reference response behavior score through a weighted rule or a scoring mechanism based on a trained model;

[0038] A standard score determination unit is configured to call a pre-established reference model, and match the standard response behavior score corresponding to the sample conditions consistent with the species, gender, age and intervention intensity of the target pet;

[0039] A correction factor calculation unit is configured to calculate the reference deviation amplitude between the reference response behavior score and the standard response behavior score, and generate a correction factor according to the reference deviation amplitude.

[0040] As a further limitation of the technical solutions of the embodiments of the present application, a preset score correction model is used when correcting the initial pet emotional response fitness score;

[0041] The score correction model is: Wherein refers to the corrected pet emotional response fitness score, refers to the initial pet emotional response fitness score, refers to the correction factor, refers to the corresponding adjustment coefficient, and is greater than 0;

[0042] In the score correction model, Wherein refers to the reference response behavior score, refers to the reference standard response behavior score, refers to the reference deviation amplitude.

[0043] Compared with the prior art, the present application has the following beneficial effects:

[0044] The application realizes dynamic correction of initial emotional response adaptability score of recommended products by introducing pet behavior video analysis and reference model comparison mechanism, and solves the identification problem of pet emotion improvement products in individual adaptability recommendation. The prior art mainly depends on static information such as label matching or historical score for recommendation, and ignores the influence of user intervention behavior on the actual emotional response of pets. Especially in emotional products, such intervention often has a significant impact on the adaptability of pets, and if not distinguished, it is easy to cause score deviation. In contrast, the application combines the linkage analysis of user intervention actions and pet response characteristics, constructs a correction factor to dynamically adjust the initial score, not only improves the authenticity of pet response evaluation and the individual accuracy of adaptability judgment, but also makes the generation of recommendation weight factors more objective and reliable. The method is suitable for accurate recommendation of pet emotion improvement products in cross-border e-commerce platforms, and has strong practicality and popularization potential. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A flowchart of the method provided by the embodiment of the application is shown.

[0046] Figure 2 A flowchart of determining a reference product and a corresponding reference video in the method provided by the embodiment of the application is shown.

[0047] Figure 3 A flowchart of generating a correction factor in the method provided by the embodiment of the application is shown.

[0048] Figure 4 An application architecture diagram of the system provided by the embodiment of the application is shown.

[0049] Figure 5 A structural block diagram of a reference data acquisition module in the system provided by the embodiment of the application is shown.

[0050] Figure 6 A structural block diagram of a correction factor determination module in the system provided by the embodiment of the application is shown. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application.

[0052] Figure 1 A flowchart of the method provided by the embodiment of the application is shown.

[0053] Specifically, an AI analysis-based cross-border e-commerce commodity intelligent recommendation method, which specifically comprises the following steps:

[0054] Step S100, obtaining monitoring video data of the target pet and an initial pet emotional response fitness score of the product to be recommended. The specific category of the product to be recommended is a pet emotional improvement product, including a calming spray, a soothing massage pad, an emotional guidance toy, or a smart device for relieving an anxious state.

[0055] In the embodiment of the present application, the monitoring video data is obtained under the premise of user authorization, ensuring that the data collection behavior conforms to the user privacy management specification. The video data can come from a smart camera device or a pet-specific monitoring terminal deployed in a home environment, which is used to record the interaction process between the target pet and the environment and specific products in a natural living state, while retaining the relevant action pictures of the user's possible intervention behavior on the pet. The collected data content includes but is not limited to the target pet's gaze direction, body movement, tail reaction, posture change, and the user's gestures, close behavior, voice, or other action instructions to support the joint modeling and analysis of subsequent behavior response and intervention action.

[0056] The product to be recommended refers to a candidate product that is identified by the system as having recommendation value during the user's browsing, collecting, or adding to the shopping cart behavior on the cross-border e-commerce platform. The product to be recommended focused on by the present application is specifically limited to pet emotional improvement products, including a calming spray, a soothing massage pad, an emotional guidance toy, or a smart device for relieving an anxious state. The reason is that such products are active in cross-border sales, but their adaptation effect is highly dependent on individual differences of specific pets, and traditional recommendation algorithms based on user tags or sales ranking cannot effectively identify the actual response of pet individuals to emotional intervention products, resulting in low matching degree and poor conversion rate. Therefore, the present application preferably takes such products as the research target to improve the recommendation accuracy and behavior credibility by establishing an individual response model.

[0057] The initial pet emotional response fitness score is a basic evaluation of the adaptation between the target pet and the product to be recommended before individual feedback correction. This score can be obtained through historical big data modeling, involving pet basic information (breed, age, personality label), product function label (such as adapting to stress scenarios, odor release method), user usage preference, and product feedback records. In the prior art, the initial score is usually generated based on collaborative filtering, content matching, and portrait-based similarity scoring, but such scoring systems generally lack dynamic verification of behavior response. Therefore, in the present application, the initial score will be used as a reference input, and the real response behavior of the target pet will be identified for dynamic correction, finally participating in the calculation of the recommendation weight factor of the product to be recommended, so that the recommendation result is more adaptive and credible.

[0058] Further, the cross-border e-commerce product intelligent recommendation method based on AI analysis further comprises the following steps:

[0059] Step S200, analyze the monitoring video data to identify whether there is user intervention when the target pet uses other products of the same class as the product to be recommended, and if so, obtain the reference product and its corresponding reference video that are most consistent with the background information of the product to be recommended.

[0060] Specifically, Figure 2 A flowchart for determining the reference product and its corresponding reference video is shown.

[0061] The analysis of the monitoring video data to identify whether there is user intervention when the target pet uses other products of the same class as the product to be recommended, and if so, obtain the reference product and its corresponding reference video that are most consistent with the background information of the product to be recommended specifically includes the following steps:

[0062] Step S201, analyze the monitoring video data to extract several sample video segments of the target pet using pet mood improvement products;

[0063] Step S202, analyze each sample video to identify whether there is user intervention action during the target pet's contact with the corresponding product, and if the intervention action is detected and the intervention duration exceeds a preset threshold, mark the sample video as having user intervention;

[0064] Step S203, if the proportion of sample videos with user intervention to all sample videos exceeds a preset proportion, determine that the target pet has user intervention when using other products of the same class as the product to be recommended, and extract the reference product and its corresponding reference video that are most consistent with the background information of the product to be recommended.

[0065] The reference product that is most consistent with the background information of the product to be recommended refers to a pet mood improvement product that is most similar to the product to be recommended in terms of product function, applicable mood category, and use situation.

[0066] In the embodiment of the present application, in step S201, the system first extracts several sample video segments of the target pet's contact with pet mood improvement products from the complete video data of the target pet. The extraction process can be completed by combining product recognition and time label pairing. Specifically, if the monitoring device accesses a device recognition module bound to pet products, or the system has the ability to detect specific product appearance features through image analysis, the image recognition model (such as YOLOv5, EfficientDet, etc.) can be used to extract the time period of the pet interacting with a certain type of mood improvement product from the video sequence, forming several behavior sample segments with product context.

[0067] In step S202, the system analyzes each sample video frame by frame to identify whether the user has obvious intervention behavior during the pet contacting the product. Intervention actions can include voice prompts, hand gestures, body contact, or other active behaviors within the visual range. In the identification process, a human action recognition model based on key point detection (such as OpenPose, HRNet, etc.) can be used in combination with a time length discrimination mechanism to determine whether it constitutes continuous intervention. If the duration of the detected intervention action exceeds a pre-set threshold (such as 3 seconds), the sample video is marked as having user intervention. This threshold is used to avoid incidental, non-interventional short-term motion interference, thereby improving the accuracy of intervention identification.

[0068] In step S203, the system counts the number of videos in all sample videos that are marked as having user intervention and calculates the proportion. Only when the proportion of videos with intervention is greater than a pre-set proportion threshold (such as 60%), is it determined that the target pet has a stable user intervention habit when using this type of product. This judgment logic aims to improve the representativeness of intervention data and avoid misjudgment of the target pet's long-term controlled use tendency due to individual video intervention. The intervention existence judgment here has a filtering significance for subsequent selection of reference samples: only in the presence of observable intervention history behavior of the pet, can the reference model perform comparative scoring under the same "intervention" condition, ensuring the adaptability and credibility of the scoring correction.

[0069] Once it is determined that the target pet has intervention behavior, the system will select the reference product and its corresponding video segment that best matches the background information of the product to be recommended from the videos with user intervention. The matching process is based on semantic similarity calculation of product attributes, and the background information involved includes product function categories (such as releasing odor, providing pressure, cold or hot compress), applicable emotional states (such as anxiety, alertness, and depression), and use situations (such as night soothing, adaptation to unfamiliar environments, and travel companionship, etc.). A product label embedding method based on vector representation (such as word2vec, BERT product description vector) can be used to calculate the semantic vector distance between the historical sample product and the current product to be recommended, and the closest one is selected as the reference product, and its corresponding video is retrieved as the reference video for subsequent behavior scoring and model invocation. This matching process ensures the consistency of the selected reference data in terms of intervention method and product background, thereby providing effective control for scoring correction.

[0070] Further, the AI analysis-based cross-border e-commerce commodity intelligent recommendation method further includes the following steps:

[0071] Step S300, identify the intervention intensity in the reference video and the reference response behavior score of the target pet, call the reference model to obtain the reference standard response behavior score under the same condition, calculate the reference deviation amplitude between the two, and generate a correction factor based on the reference deviation amplitude.

[0072] The reference model is a structured behavior scoring model constructed based on big data, which indexes and clusters samples according to attributes such as pet species, gender, age and intervention intensity, and presets standard response behavior score values under different condition combinations.

[0073] Specifically, Figure 3 A flowchart for generating the correction factor is shown.

[0074] The reference deviation amplitude between the reference response behavior score and the standard response behavior score under the same condition is calculated, and the correction factor is generated according to the reference deviation amplitude, and specifically includes the following steps:

[0075] Step S301, the image sequence of the user intervention stage in the reference video is extracted, the intervention action amplitude, duration and frequency of the user are identified, and the intervention intensity is generated based on the linear weighted combination of the above parameters or through the output of the training model;

[0076] Step S302, the behavior reaction characteristics of the target pet after the user intervention are analyzed, including the gaze duration, approaching action, tail movement amplitude, posture stability and other parameters, and the delay time and duration of the behavior reaction are combined to generate the reference response behavior score through the weighted rule or the scoring mechanism based on the training model;

[0077] Step S303, the pre-established reference model is called, and the standard response behavior score corresponding to the sample condition consistent with the species, gender, age and intervention intensity of the target pet is matched;

[0078] Step S304, the reference deviation amplitude between the reference response behavior score and the standard response behavior score is calculated, and the correction factor is generated according to the reference deviation amplitude.

[0079] In the embodiment of the application, the system quantitatively analyzes the user intervention behavior and the response behavior of the target pet in the reference video, and combines the standard response behavior score stored in the reference model to generate a correction factor for adjusting the initial emotional response fitness score, thereby improving the credibility and adaptability of the final score. The process includes steps S301 to S304.

[0080] In step S301, the system first locates the start and end time period of the user intervention behavior in the reference video and extracts the image sequence in this time period. With the help of user action recognition technology in the image sequence, the system can detect the upper limb activity trajectory in the user intervention process through a human pose estimation model (such as OpenPose, MediaPipe Pose), thereby quantifying the amplitude of the intervention action. The system synchronously records the action duration and the number of actions per unit time as the occurrence frequency. To generate the intervention intensity, a preset weight coefficient can be set to linearly weight and combine the action amplitude, duration, and frequency, or a supervised learning-based regression model (such as multilayer perception, XGBoost, etc.) can be used, taking the above parameters as input features and outputting the intervention intensity value, to achieve higher-precision personalized intensity estimation.

[0081] In step S302, the system focuses on the time window after the user intervention behavior and analyzes the behavior response parameters of the target pet. The behavior characteristics include but are not limited to gaze target duration (analyzed by combining eye orientation and head orientation), body approaching action (detected based on forelimb / gait action displacement), tail swing amplitude (identified by tail region contour change), and posture stability (calculated based on the swing amplitude of the skeleton key points on the time axis). The system further analyzes the immediacy and maintainability of the response through time delay and behavior duration. In the score generation process, a score function based on feature weighting rules can be constructed, or a trained neural network scoring model can be used to comprehensively evaluate the multi-dimensional behavior characteristics, and finally output the reference response behavior score.

[0082] Step S303 involves calling a reference model to obtain a standard response behavior score. The reference model is a structured behavior scoring model built by the system in the training stage through large-scale samples. The model stores standard scores under various pet species (such as dogs, cats, etc.), genders, age groups, and intervention intensity conditions. The score value is based on the average behavior performance of multiple pet samples that are stable and controllable in historical data, generated through a combination of manual annotation and model analysis.

[0083] In step S304, the system compares the reference response behavior score with the standard response behavior score, calculates the deviation amplitude between the two, and the commonly used calculation method is the normalized difference, such as wherein refers to the reference response behavior score, Refers to the reference standard response behavior score. The deviation reflects the relative deviation between the actual response of the current pet and the reference response. This deviation is used to generate a correction factor. The purpose of introducing the correction factor is: if the target pet shows significantly better response behavior than the reference model in real interaction, the initial score should be adjusted upward; otherwise, it should be adjusted downward to ensure the sensitivity and matching degree of the recommended results to the actual behavior. In this way, the system realizes adaptive adjustment of the recommendation adaptability based on intervention response data, effectively improving the individualized precision and credibility of pet emotion improvement product recommendations in cross-border e-commerce.

[0084] Further, the AI analysis-based intelligent recommendation method for cross-border e-commerce goods further comprises the following steps:

[0085] Step S400, correct the initial pet emotional response adaptability score according to the correction factor to obtain a corrected pet emotional response adaptability score, and apply the corrected pet emotional response adaptability score to the calculation of the recommendation weight factor of the product to be recommended.

[0086] In the correction of the initial pet emotional response adaptability score, a preset score correction model is used;

[0087] The score correction model is: Wherein Refers to the corrected pet emotional response adaptability score, Refers to the initial pet emotional response adaptability score, Refers to the correction factor, Refers to The corresponding adjustment coefficient, and Greater than 0;

[0088] In the score correction model, Wherein Refers to the reference response behavior score, Refers to the reference standard response behavior score, Refers to the reference deviation amplitude.

[0089] In the embodiment of the present application, in step S400, the system corrects the initial pet emotional response fitness score according to the correction factor, so as to obtain a score result that is more suitable for the actual behavior performance of the target pet, and takes the corrected score as the core basis for calculating the recommended weight factor. In the recommendation process, the system will first generate an initial pet emotional response fitness score based on the existing technology, which comprehensively considers product attributes, user preferences, pet basic information and other static factors, and is used to preliminarily judge the matching degree between the product to be recommended and the target pet. However, in actual use, users often intervene in pets through voice, action and other ways. Such intervention behavior may enhance the acceptance of pets to the product, or may cause additional stress and reduce adaptability. Therefore, relying only on the initial score may not accurately reflect the real feedback of the pet in the actual situation.

[0090] Based on this, the present application introduces a score correction mechanism. After identifying the user intervention behavior, the actual response performance of the pet in this intervention context is analyzed and compared with the average response performance in the reference model, the score deviation is quantified, and a correction factor is generated to adjust the original score. In this way, the system not only considers the static adaptability of the product itself, but also introduces the dynamic influence of user behavior on the emotional response of the pet, thereby greatly improving the personalization and credibility of the recommendation decision. The correction mechanism can effectively solve the problem of "ignoring behavior feedback" in traditional recommendation systems, making the recommendation result more close to the real use situation.

[0091] The core benefit of using the score correction mechanism is that it can compensate for the deviation of the initial score when there is a lack of reference for the individual behavior of the target pet, and dynamically adjust the recommendation strength through the behavior response differences of specific pets, thereby improving the judgment quality of "whether to recommend, priority ranking". Finally, the corrected score is used as one of the input items of the recommended weight factor of the product to be recommended, participating in the sorting logic or distribution mechanism, which helps the recommendation engine to achieve more fine adaptation recommendation when facing individual behavior data rich, thereby improving the conversion rate of goods and the satisfaction of recommendation.

[0092] The score correction model used in the present application is a linear correction mechanism with clear structure and intuitive logic, which has the advantages of low calculation cost, transparent parameter structure and strong adaptability. The model effectively maps the deviation signal of the behavior data into the amplification or compression process of the score value by multiplying the initial score by a linear coefficient containing a deviation correction factor. In addition to the model used in the present application, other score correction methods can also be considered, such as exponential decay model, Bayesian score correction, weight redistribution type score fusion model or end-to-end feedback score mechanism under the deep learning framework. These methods can be replaced according to the scene in actual deployment, but the model of the present application has sufficient expressiveness in the structured recommendation scene.

[0093] The meaning of the score correction model can be described as follows: the initial pet emotional response fitness score is multiplied by a correction coefficient composed of a correction factor, and the value of the coefficient is equal to one plus an adjustment parameter multiplied by a score deviation factor, wherein the score deviation factor is the reference response behavior score of the target pet minus the standard response behavior score, and then divided by the standard response behavior score.

[0094] The effect achieved by the technical scheme of the present application is intuitively shown by an example as follows:

[0095] Suppose that the initial pet emotional response fitness score of a product to be recommended is 0.65. After analyzing the historical video data, the system extracts the response behavior score of the target pet when using a reference product similar to the product background as 0.80, and the standard score in the reference model under the same species, gender, age and intervention intensity condition is 0.50. The calculated score deviation factor is (0.80-0.50) / 0.50, and the result is 0.60. If the system sets the adjustment parameter as 0.5, the final correction coefficient is 1+0.5x0.60, i.e. 1+0.30=1.30. The final corrected score is the initial score 0.65x1.30=0.845. The corrected score will be used to update the recommendation weight factor of the product, improving its ranking priority in the recommendation system. The increase of the score not only reflects that the pet shows a stronger adaptive response in the use situation with user intervention, but also indicates that the actual effect of the product on the target pet individual is better than the average level.

[0096] In other words, the final corrected score is higher than the initial score, which indicates that under the premise that the user has stable intervention habits, the pet's response to the product is more positive, so it is reasonable and necessary to improve the score. In this example, by introducing the behavior data correction process, a product with an average adaptation degree is successfully adjusted to a high adaptation individual recommendation candidate, effectively improving the credibility and personalization level of the recommendation.

[0097] Further, Figure 4 The application architecture diagram of the system provided by the embodiment of the present application is shown.

[0098] In another preferred embodiment provided by the present application, an AI analysis-based cross-border e-commerce commodity intelligent recommendation system comprises:

[0099] The data acquisition module 100 is configured to acquire the monitoring video data of the target pet and the initial pet emotional response fitness score of the product to be recommended. The specific category of the product to be recommended is a pet emotional improvement product, including a calming spray, a soothing therapy pad, an emotional guidance toy or a smart device for relieving anxiety.

[0100] Further, the AI analysis-based cross-border e-commerce commodity intelligent recommendation system further comprises:

[0101] The reference data acquisition module 200 is configured to analyze the monitoring video data, identify whether user intervention exists when the target pet uses other products of the same type as the product to be recommended, and acquire the reference product and the corresponding reference video that are most consistent with the background information of the product to be recommended if user intervention exists.

[0102] Specifically, Figure 5 The structure block diagram of the reference data acquisition module 200 in the system provided by the embodiment of the application is shown.

[0103] In the preferred embodiment provided by the application, the reference data acquisition module 200 specifically comprises:

[0104] The sample video extraction unit 201 is configured to analyze the monitoring video data and extract a plurality of sample video clips in which the target pet uses pet emotion improvement products.

[0105] The intervention behavior judgment unit 202 is configured to analyze each sample video and identify whether user intervention actions exist in the process in which the target pet contacts the corresponding product, and mark the sample video as existing user intervention if the intervention actions are detected and the intervention duration exceeds a preset threshold.

[0106] The reference video extraction unit 203 is configured to determine that user intervention exists when the target pet uses other products of the same type as the product to be recommended if the proportion of sample videos in which user intervention exists exceeds a preset proportion, and extract the reference product and the corresponding reference video that are most consistent with the background information of the product to be recommended.

[0107] The reference product that is most consistent with the background information of the product to be recommended refers to a pet emotion improvement product that is closest to the product to be recommended in terms of product function, applicable emotion category, use situation and other factors.

[0108] Further, the AI analysis-based cross-border e-commerce commodity intelligent recommendation system further comprises:

[0109] The correction factor determination module 300 is configured to identify the intervention intensity in the reference video and the reference response behavior score of the target pet, call the reference model to obtain the reference standard response behavior score under the same condition, calculate the reference deviation amplitude between the two, and generate a correction factor according to the reference deviation amplitude.

[0110] The reference model is a structured behavior score model constructed based on big data, indexes and clusters samples according to attribute dimensions such as pet species, gender, age and intervention intensity, and presets standard response behavior score values under different condition combinations.

[0111] Specifically, Figure 6 A structural block diagram of the correction factor determination module 300 in the system provided by the embodiment of the present application is shown.

[0112] In the preferred embodiment provided by the present application, the correction factor determination module 300 specifically comprises:

[0113] An intervention intensity generation unit 301 is configured to extract an image sequence of a user intervention stage in a reference video, identify intervention action amplitude, duration and frequency of the user, and generate intervention intensity based on linear weighted combination of the above parameters or output through a trained model;

[0114] A reference score generation unit 302 is configured to analyze behavior reaction features of a target pet after user intervention, including gaze duration, approaching action, tail movement amplitude, posture stability and other parameters, and generate a reference response behavior score through a weighted rule or a scoring mechanism based on a trained model in combination with delay time and duration of the behavior reaction;

[0115] A standard score determination unit 303 is configured to call a pre-established reference model, and match a standard response behavior score corresponding to a sample condition in which the species, gender and age of the target pet are consistent and the intervention intensity is consistent;

[0116] A correction factor calculation unit is configured to calculate a reference deviation amplitude between the reference response behavior score and the standard response behavior score, and generate a correction factor according to the reference deviation amplitude.

[0117] Further, the AI analysis-based cross-border e-commerce commodity intelligent recommendation system further comprises:

[0118] A score correction module 400 is configured to correct an initial pet emotional response fitness score according to the correction factor, obtain a corrected pet emotional response fitness score, and apply the corrected pet emotional response fitness score in calculation of a recommendation weight factor of a product to be recommended.

[0119] In the correction of the initial pet emotional response fitness score, a preset score correction model is adopted;

[0120] The score correction model is: wherein the corrected pet emotional response fitness score, the initial pet emotional response fitness score, the correction factor, the a corresponding adjustment coefficient, and greater than 0;

[0121] In the score correction model, wherein refers to a reference response behavior score, refers to a reference standard response behavior score, refers to a reference bias magnitude.

[0122] It should be understood that although the steps in the flowcharts of the embodiments of the present application are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, at least some of the steps in the embodiments can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least some of the other steps or sub-steps or stages of other steps.

[0123] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a non-volatile computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of the method. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0124] Any combination of the technical features of the above-mentioned embodiments can be made. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0125] The above embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but should not be understood as a limitation on the patent scope of the present application. It should be noted that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

[0126] The above merely describes the preferred embodiments of the present application and should not be used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for AI analysis-based intelligent recommendation of cross-border e-commerce goods, characterized in that, The method comprises: acquiring monitoring video data of a target pet and an initial pet emotional response fitness score of a product to be recommended; analyzing the monitoring video data to identify whether there is user intervention when the target pet uses other products of the same type as the product to be recommended, and if so, acquiring a reference product and its corresponding reference video that are most consistent with the background information of the product to be recommended; identifying the intervention intensity in the reference video and the reference response behavior score of the target pet, calling a reference model to obtain the reference standard response behavior score under the same conditions, calculating the reference deviation amplitude between the two, and generating a correction factor based on the reference deviation amplitude; correcting the initial pet emotional response fitness score based on the correction factor to obtain a corrected pet emotional response fitness score, and applying the corrected pet emotional response fitness score in the calculation of the recommendation weight factor of the product to be recommended. 2.The AI analysis-based cross-border e-commerce commodity intelligent recommendation method according to claim 1, characterized in that, The specific category of the product to be recommended is a pet emotional improvement product, including a calming spray, a soothing therapy pad, an emotional guidance toy, or a smart device for relieving anxiety. 3.The AI analysis-based cross-border e-commerce commodity intelligent recommendation method according to claim 2, characterized in that, The step of analyzing the monitoring video data to identify whether there is user intervention when the target pet uses other products of the same type as the product to be recommended, and if so, acquiring a reference product and its corresponding reference video that are most consistent with the background information of the product to be recommended, comprises: analyzing the monitoring video data to extract a plurality of sample video clips of the target pet using pet emotional improvement products; parsing each sample video to identify whether there is user intervention action during the target pet's contact with the corresponding product, and if the intervention action is detected and the intervention duration exceeds a preset threshold, marking the sample video as having user intervention; if the proportion of sample videos with user intervention to all sample videos exceeds a preset proportion, it is determined that the target pet has user intervention when using other products of the same type as the product to be recommended, and the reference product and its corresponding reference video that are most consistent with the background information of the product to be recommended are extracted. The reference product that is most consistent with the background information of the product to be recommended refers to a pet emotional improvement product that is closest to the product to be recommended in terms of product function, applicable emotion category, and use situation. 4.The AI analysis-based cross-border e-commerce commodity intelligent recommendation method according to claim 3, characterized in that, The reference model is a structured behavior score model constructed based on big data, which indexes and clusters samples according to pet species, gender, age, and intervention intensity attribute dimensions, and presets standard response behavior score values under different condition combinations. 5.The AI analysis-based cross-border e-commerce commodity intelligent recommendation method according to claim 4, characterized in that, The steps of identifying the intervention intensity in the reference video and the reference response behavior score of the target pet, calling a reference model to obtain the reference standard response behavior score under the same conditions, calculating the reference deviation amplitude between the two, and generating a correction factor based on the reference deviation amplitude, comprise: extracting image sequences of the user intervention stage in the reference video, identifying the intervention action amplitude, duration, and frequency of the user, and generating the intervention intensity based on linear weighted combination of the above parameters or through a trained model output; The behavior reaction characteristics of the target pet after user intervention are analyzed, including gaze duration, approach action, tail movement amplitude, posture stability, and the like, and a reference response behavior score is generated through a weighted rule or a scoring mechanism based on a trained model, in combination with the delay time and duration of the behavior reaction; A pre-established reference model is called to match the standard response behavior score corresponding to the sample conditions consistent with the species, gender, and age of the target pet and the uniform intervention intensity; The reference deviation amplitude between the reference response behavior score and the standard response behavior score is calculated, and a correction factor is generated based on the reference deviation amplitude. 6.The AI analysis-based cross-border e-commerce commodity intelligent recommendation method according to claim 5, characterized in that, In the correction of the initial pet emotional response fitness score, a preset score correction model is used; The score correction model is: wherein refers to the corrected pet emotional response fitness score, refers to the initial pet emotional response fitness score, refers to the correction factor, refers to the corresponding adjustment coefficient, and is greater than 0; In the score correction model, wherein refers to a reference response behavior score, refers to a reference standard response behavior score, refers to a reference bias magnitude.

7. An AI analysis-based cross-border e-commerce commodity intelligent recommendation system, characterized in that, The system comprises a data acquisition module, a reference data acquisition module, a correction factor determination module, and a score correction module, wherein: The data acquisition module is configured to acquire monitoring video data of a target pet and an initial pet emotional response fitness score of a product to be recommended; the specific category of the product to be recommended is a pet emotional improvement product, including a calming spray, a soothing therapy pad, an emotional guidance toy, or a smart device for relieving anxiety; The reference data acquisition module is configured to analyze the monitoring video data, identify whether there is user intervention when the target pet uses other products of the same category as the product to be recommended, and if so, acquire a reference product most consistent with the background information of the product to be recommended and a corresponding reference video; The correction factor determination module is configured to identify the intervention intensity in the reference video and the reference response behavior score of the target pet, call a reference model to acquire a reference standard response behavior score under the same conditions, calculate the reference deviation amplitude between the two, and generate a correction factor based on the reference deviation amplitude; the reference model is a structured behavior score model constructed based on big data, which indexes and clusters samples according to pet species, gender, age, and intervention intensity; The score correction module is configured to correct the initial pet emotional response fitness score based on the correction factor to obtain a corrected pet emotional response fitness score, and apply the corrected pet emotional response fitness score in the calculation of the recommendation weight factor of the product to be recommended. 8.The AI analysis-based cross-border e-commerce commodity intelligent recommendation system according to claim 7, characterized in that, The reference data acquisition module specifically comprises: A sample video extraction unit is configured to analyze the monitoring video data and extract a plurality of sample video segments in which the target pet uses pet emotional improvement products; An intervention behavior judgment unit is configured to analyze each sample video, identify whether there is a user intervention action in the process of the target pet contacting the corresponding product, and if the intervention action is detected and the intervention duration exceeds a preset threshold, mark the sample video as having user intervention; A reference video extraction unit is configured to determine that there is user intervention when the target pet uses other products of the same category as the product to be recommended if the proportion of sample videos with user intervention to all sample videos exceeds a preset proportion, and extract a reference product most consistent with the background information of the product to be recommended and a corresponding reference video. The reference product most consistent with the background information of the product to be recommended refers to a pet emotion improvement product most close to the product to be recommended in the comprehensive aspects of product functions, applicable mood categories, and use situations. 9.The AI analysis-based cross-border e-commerce commodity intelligent recommendation system according to claim 8, characterized in that, The correction factor determination module specifically comprises: An intervention intensity generation unit configured to extract an image sequence of a user intervention stage in the reference video, identify intervention action amplitude, duration, and frequency of the user, and generate intervention intensity based on linear weighted combination of the parameters or output of a trained model; A reference score generation unit configured to analyze behavior reaction characteristics of the target pet after the user intervention, including gaze duration, approaching action, tail movement amplitude, posture stability, and combined with delay time and duration of the behavior reaction, generate a reference response behavior score through a weighted rule or a scoring mechanism based on a trained model; A standard score determination unit configured to call a pre-established reference model, and match a standard response behavior score corresponding to a sample condition of consistent species, gender, age, and intervention intensity of the target pet; A correction factor calculation unit configured to calculate a reference deviation amplitude between the reference response behavior score and the standard response behavior score, and generate a correction factor according to the reference deviation amplitude. 10.The AI analysis-based cross-border e-commerce commodity intelligent recommendation system according to claim 9, characterized in that, In the correction of the initial pet emotion response fitness score, a preset score correction model is adopted. The score correction model is: wherein refers to the corrected pet emotional response fitness score, refers to the initial pet emotional response fitness score, refers to the correction factor, refers to the corresponding adjustment coefficient, and is greater than 0; In the score correction model, wherein refers to a reference response behavior score, refers to a reference standard response behavior score, refers to a reference bias magnitude.

Images