Marketing method based on intelligent commodity background matching

Abstract

The invention discloses a marketing method based on intelligent commodity background matching, and relates to the technical field of commodity marketing, and the method specifically comprises the following steps: obtaining backflow behavior new commodity browsing information generated when a user browses a commodity which is not concerned before in a backflow behavior period in real time, and carrying out the analysis after obtaining, respectively generating a backflow behavior interest index and a backflow purchase intention coefficient; constructing an effective evaluation model for the generated reflow behavior interest index and the reflow purchase intention coefficient, and generating an effective evaluation coefficient; and analyzing the generated effective evaluation coefficient, evaluating whether the backflow behavior of the user is effective, and respectively generating an effective signal and a non-effective signal according to an evaluation result. The problem that an existing recommendation system cannot recognize and evaluate the effectiveness of user backflow behaviors is solved, accurate recommendation based on user backflow requirements is achieved by dynamically adjusting the recommendation strategy, and the recommendation accuracy and the platform conversion rate are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of commodity marketing, and in particular to a marketing method based on intelligent commodity background matching. Background Art

[0002] The marketing method based on intelligent product background matching is a marketing method that uses intelligent algorithms and big data technology to accurately match products with target users by analyzing the background information of products (such as usage scenarios, user behavior data, geographic location, time, etc.) and personalized needs of users. This method can push the products or services that best meet the needs of users in specific situations, greatly improving marketing efficiency and user experience. The reason for using intelligent product background matching for marketing is that traditional marketing models usually rely on wide coverage and inefficient promotion methods, while background matching technology can deeply explore the core value of products and the hidden needs of users, avoid information overload and waste of resources, and at the same time improve user recognition and purchase rate of products through accurate push, thereby creating higher commercial value for enterprises.

[0003] Existing marketing technologies based on intelligent product background matching conduct marketing in the following ways: First, the system collects product background data, including product attributes, usage scenarios, target user portraits, etc., and obtains user behavior data, such as browsing history, purchase history, geographic location, real-time environment (such as weather, time), etc. Then, using big data analysis and machine learning algorithms, the product background data is intelligently matched with user data to generate personalized recommendation content. The system analyzes the matching results to determine the user's potential needs and pushes related products or services at the most suitable time, place, and scenario (such as mobile applications, social media ads, emails, etc.). This process usually integrates technologies such as natural language processing, image recognition, and deep learning to more accurately understand user preferences and product characteristics, thereby providing highly relevant personalized marketing plans to enhance user purchase intentions and experience satisfaction.

[0004] The prior art has the following deficiencies: In e-commerce platforms, when a user does not make any purchases or browse for a period of time, the system usually assumes that the user is no longer interested in the relevant products. However, in some cases, such as holiday promotions, advertising push or the guidance of specific events, the user's interest may suddenly return and start browsing products or related accessories that they have not paid attention to before. For example, a user bought a smart speaker a few months ago and did not browse similar products until the promotion period, when the user suddenly started browsing the smart speaker and its accessories. Existing recommendation technologies based on smart product background matching often rely only on users' historical purchase and browsing behaviors to generate recommendations, while ignoring the dynamic changes of users' "return behavior", a new demand. Therefore, the system still recommends outdated products based on past purchase records and fails to timely identify new needs generated by users during promotions. Since the existing recommendation technology fails to distinguish and evaluate the effectiveness of this return behavior, it cannot adjust the recommendation logic in time, so that the recommended content is still based on outdated data and fails to push products that meet the user's current needs. This will cause the recommended products to not match the user's actual interests, thereby reducing the accuracy of the recommendation system, causing users to be dissatisfied with the recommendation system, and may even churn, thereby affecting the platform's conversion rate and overall sales efficiency.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a marketing method based on intelligent commodity background matching to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a marketing method based on intelligent commodity background matching, specifically comprising the following steps: By real-time monitoring of the browsing and purchasing behavior of the user on the e-commerce platform, the user's return behavior is identified. When the user's return behavior is identified, whether the user browses a product that has not been previously viewed during the return behavior is detected in real time, and when it is detected that the user browses a product that has not been previously viewed during the return behavior, the following steps are performed; Real-time acquisition of new product browsing information of return behavior generated when users browse products that they have not paid attention to before during the return behavior period, and analysis after acquisition to generate return behavior interest index and return purchase intention coefficient respectively; Construct an effective evaluation model for the generated return behavior interest index and return purchase intention coefficient to generate an effective evaluation coefficient; Analyze the generated effective evaluation coefficient to evaluate whether the user's recirculation behavior is effective, and generate effective signals and ineffective signals respectively according to the evaluation results; In the case of generating a valid signal, continuously obtaining a number of valid evaluation coefficients generated subsequently, analyzing them after generation, generating signals of different levels, and taking corresponding recommended measures according to the generated signals of different levels; Continuously monitor users' return behavior and new product browsing information, and dynamically adjust the parameters and recommendation strategies of the return behavior effective evaluation model based on real-time monitoring results. At the same time, store and analyze historical data to optimize model parameters and recommendation strategies.

[0008] Preferably, the recurrent behavior new product browsing information generated when the user browses products that have not been previously viewed during the recurrent behavior is obtained in real time, and analyzed after acquisition to generate a recurrent behavior interest index and a recurrent purchase intention coefficient, respectively, which specifically includes the following steps: Real-time acquisition of new product browsing information generated by users browsing products that they have not previously viewed during the recurrent behavior, and pre-processing after acquisition; Extracting return behavior interest information and return purchase intention information from the preprocessed return behavior new product browsing information; The extracted return behavior interest information and return purchase intention information are analyzed to generate the return behavior interest index and return purchase intention coefficient respectively.

[0009] Preferably, the logic for obtaining the return behavior interest index is as follows: Extract the return behavior interest information from the preprocessed return behavior new product browsing information, including the number of times users browse new products in different time periods during the return period, the time they stay on new products, and the ratio of the number of clicks to the number of impressions of new products, and mark them as , and , Indicates that the user is during the return period Number of views of new products in a certain period of time, Indicates that the user is during the return period The dwell time of the new product in the time period, Indicates that the user is during the return period The ratio of clicks to impressions for new products in a given time period. , is a positive integer; The user's stay time for the new product in different time periods during the return period is constructed into a set, and the maximum value in the set is marked as ; Calculate the return behavior interest index. The specific calculation formula is as follows: In the formula, The index of interest in return behavior.

[0010] Preferably, the logic for obtaining the return purchase intention coefficient is as follows: Extract the return purchase intention information from the preprocessed return behavior new product browsing information, including the number of times users add new products to the shopping cart at different time periods during the return period, the conversion rate of purchasing new products, and the number of times new products are collected, and mark them as , and , Indicates that the user is during the return period The number of times new products are added to the shopping cart during a certain period of time. Indicates that the user is during the return period Conversion rate of new products purchased in a certain time period, Indicates that the user is during the return period The number of times a new product is collected in a certain period of time. , is a positive integer; The conversion rate of users purchasing new products in different time periods during the return period is constructed into a set, and the maximum value in the set is marked as ; Calculate the return purchase intention coefficient. The specific calculation formula is as follows: In the formula, is the return purchase intention coefficient.

[0011] Preferably, the generated recurrent behavior index and return purchase intention coefficient Construct an effective evaluation model and generate an effective evaluation coefficient through weighted summation. The specific calculation formula is as follows: In the formula, is the effective evaluation coefficient, and Recurrent Behavior Interest Index and return purchase intention coefficient The non-zero weight coefficient of .

[0012] Preferably, a preset effective evaluation coefficient threshold is determined , and after determination, the effective evaluation coefficient generated Compare and evaluate whether the user's return behavior is effective based on the comparison results. The specific comparison analysis is as follows: like , the user's return behavior is invalid; like , the user's recirculation behavior is valid.

[0013] Preferably, a valid signal and an invalid signal are generated respectively according to the evaluation result, specifically: If the evaluation result shows that the user's recirculation behavior is invalid, an invalid signal is generated; If the evaluation result shows that the user's recirculation behavior is valid, a valid signal is generated.

[0014] Preferably, when a valid signal is generated, several valid evaluation coefficients generated subsequently are continuously obtained and recalibrated as , is the number of valid evaluation coefficients subsequently generated when a valid signal is generated, , is a positive integer; Calculate several subsequent effective evaluation coefficients The average , according to the formula: ; Calculate several subsequent effective evaluation coefficients Standard Deviation , according to the formula: .

[0015] Preferably, a preset average value threshold of a plurality of effective evaluation coefficients subsequently generated when a valid signal is generated is determined. and preset standard deviation threshold , and after determination, respectively with the average value of several effective evaluation coefficients and standard deviation Perform a comparison and generate different levels of signals based on the comparison results. The specific comparison analysis is as follows: like and , generating a first level signal; like and , generating a second level signal; like and , generating a third level signal; like and , generating a fourth level signal.

[0016] Preferably, corresponding recommended measures are taken according to the generated signals of different levels, specifically: If a first-level signal is generated, the recommended measures taken are: increase the recommendation frequency, improve the exposure of new products in the user's recommendation list, and push them to the platform's homepage and popular recommendations; If a second-level signal is generated, the recommended measures are: reduce the frequency of recommendations, keep recommending new products, but reduce their display priority and frequency; If a third-level signal is generated, the recommended measures taken are: maintain the existing recommendation frequency and provide new product recommendations based on the user's current interest intensity; If a fourth-level signal is generated, the recommended action taken is: stop recommending and remove the new product from the user's recommendation list.

[0017] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. The present invention effectively solves the blind spot problem of existing e-commerce platform recommendation systems when user interests change by monitoring user return behaviors in real time and combining return behaviors with new product browsing information. Traditional recommendation systems usually make product recommendations based on historical data, assuming that user interests are fixed. This approach ignores user return behaviors, that is, after a period of no purchase or browsing, user interests may recover or change due to promotions, advertising push or other events. By capturing and analyzing interest fluctuations during user return in real time, the technical solution can identify and process new user needs, thereby ensuring that the recommendation system responds to users' dynamic needs more flexibly and accurately.

[0018] 2. The present invention realizes a multi-dimensional evaluation of user reflow behavior by generating a reflow behavior interest index and a reflow purchase intention coefficient, and combining the two to build an effective evaluation model. These evaluation coefficients can accurately measure the user's interest intensity and purchase potential for new products, avoiding static analysis that only relies on historical behavior data. By generating effective evaluation coefficients through weighted summation and comparing them with set thresholds, the scheme can effectively distinguish between effective and invalid reflow behaviors, ensure real-time adjustment of recommendation logic, reduce resource waste, and improve recommendation accuracy and user satisfaction.

[0019] 3. The present invention can realize personalized recommendation strategy adjustment by generating signals of different levels and taking corresponding recommendation measures according to these signals. The recommendation measures corresponding to different signal levels include increasing recommendation frequency, adjusting recommendation priority, maintaining existing recommendation frequency and stopping recommendation, which enables the system to adjust the recommendation intensity in time according to changes in user feedback behavior, thereby effectively improving the conversion rate and sales efficiency of the platform. By continuously monitoring and analyzing historical data, the system can also optimize the parameters of the feedback behavior evaluation model and dynamically adjust the recommendation strategy, so that the platform can adapt to long-term changes in user needs and improve the adaptability and flexibility of the system. On the whole, the technical solution not only improves the accuracy of the recommendation system, but also ensures the efficient use of resources, improves user experience and enhances the competitiveness of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0021] Figure 1 The figure is a flow chart of the marketing method based on intelligent commodity background matching of the present invention. DETAILED DESCRIPTION

[0022] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0023] The present invention provides Figure 1 The marketing method based on intelligent commodity background matching shown in the figure specifically includes the following steps: By real-time monitoring of the browsing and purchasing behavior of the user on the e-commerce platform, the user's return behavior is identified. When the user's return behavior is identified, whether the user browses a product that has not been previously viewed during the return behavior is detected in real time, and when it is detected that the user browses a product that has not been previously viewed during the return behavior, the following steps are performed; Real-time monitoring of users' browsing and purchasing behaviors on e-commerce platforms can be achieved by collecting and analyzing every interaction data of users on the platform. Specifically, the system can integrate an event tracking module in the background to capture users' browsing history, purchase history, shopping cart, favorites and other behaviors in real time, and store these data in the database. Whenever a user visits the platform, the system will compare the user ID, behavior timestamp and historical activity data. If a user does not perform any behavior within a certain period of time (for example, 30 days or more), and then revisits the platform and starts browsing products, it can be considered that the user has recurrent behavior. This process automatically triggers the detection of recurrent behavior by comparing the time interval between the user's historical data and the current activity.

[0024] When the system recognizes the user's return behavior, it needs to detect in real time whether the products browsed by the user during the return period are products that the user has not paid attention to before. This can be achieved by establishing a user preference model, which identifies the user's preferred product categories, brands, and types based on the user's past browsing and purchase data. During return, the system will compare the products browsed by the current user with the user's historical browsing or purchase records to see if there is any overlap in products. If the currently browsed product is not in the historical record, it is considered that the user has browsed a product that he has not paid attention to before. This process can be quickly compared through a recommendation engine combined with collaborative filtering or content recommendation algorithms to ensure that each returning user can accurately match his current interests.

[0025] This implementation process can solve the problem that existing recommendation systems are too dependent on historical data and fail to dynamically respond to changes in user needs. Under the guidance of holiday promotions, advertising push or specific events, users' interests often change. Traditional recommendation systems will continue to push products based on outdated historical behaviors and fail to meet users' new needs. By real-time detection of users' browsing behavior for unfocused products during the reflow period, the system can capture these new points of interest in a timely manner and make corresponding adjustments in the recommendation logic, thereby pushing products that meet users' current needs. This not only improves the accuracy of recommendations, but also enhances the user experience, avoiding user loss and low conversion rates caused by the lack of dynamic response of the recommendation system.

[0026] Real-time acquisition of new product browsing information of return behavior generated when users browse products that they have not paid attention to before during the return behavior period, and analysis after acquisition to generate return behavior interest index and return purchase intention coefficient respectively; In this embodiment, the recurrent behavior new product browsing information generated when the user browses products that have not been previously viewed during the recurrent behavior is obtained in real time, and analyzed after acquisition to generate a recurrent behavior interest index and a recurrent purchase intention coefficient, respectively, which specifically includes the following steps: Real-time acquisition of new product browsing information generated by users browsing products that they have not previously viewed during the recurrent behavior, and pre-processing after acquisition; To obtain in real time the browsing information of new products generated by users browsing products that they have not paid attention to before during the return behavior, it can be achieved by integrating a real-time event tracking system on the e-commerce platform. The system can automatically record the user's behavior data every time the user visits the platform, such as the user's clicks, browsing, and length of stay. Specifically, when the user returns to the platform, the system will capture and store the user's browsing behavior for products in real time, especially the first browsing of products that have not been paid attention to. This can be achieved through user ID-based tracking. The system analyzes the user's historical browsing records and compares them with the current behavior to identify whether the currently browsed product is a product that the user has not browsed before. In addition, the system will transmit this information to the backend database in real time to ensure the timeliness and accuracy of the data and avoid delays affecting the data processing results.

[0027] Preprocessing is an important step in data analysis, and its purpose is to convert the original behavior data into a form suitable for analysis and calculation. In the real-time browsing information of new products in the return behavior, there may be noise data (such as abnormal clicks, invalid browsing, etc.) or inconsistent data formats, so preprocessing operations such as denoising, normalization and standardization are required. The denoising operation can be completed by setting filtering rules (for example, excluding records with invalid browsing time below a certain threshold) to ensure the validity of the data. Normalization converts data of different ranges (such as browsing time and click frequency) into a unified scale to ensure comparability between different data types. Standardization is to process the data so that different user behavior data are calculated under the same standard to avoid data deviation due to large differences in user behavior. Through these preprocessing steps, the system can ensure the accuracy of subsequent data analysis and evaluation, and provide clean and standardized input data for generating the return behavior interest index and the return purchase intention coefficient.

[0028] Extracting return behavior interest information and return purchase intention information from the preprocessed return behavior new product browsing information; Extracting the return behavior interest information and return purchase intention information from the preprocessed return behavior new product browsing information can be achieved through data mining and feature extraction methods. First, the return behavior interest information can be extracted by analyzing the user's browsing behavior during the return period. This includes filtering out features such as user browsing frequency, stay time, page switching frequency, etc. from the preprocessed data to measure the user's interest in the new product during the return period. Specifically, the system can calculate the interest intensity index based on the user's browsing time and access frequency of each product, and integrate these data into an interest index through a set algorithm (such as weighted average method). For the return purchase intention information, the extraction process focuses on the user's shopping cart behavior, purchase conversion rate, and collection behavior. The system determines the user's purchase potential by analyzing whether the user adds new products to the shopping cart, completes the purchase, and adds the product to the favorites during the return period. In order to extract this information, the system will capture user behavior logs from the database in real time and mark specific actions involving purchase intentions, such as the number of times the user adds to the shopping cart, the conversion rate from browsing to purchase, and the frequency of collection behavior. Through these steps, the system can quantify the user's purchasing potential during the return period, providing the necessary basis for subsequent return behavior analysis. This process relies entirely on data mining algorithms and real-time data analysis technology to ensure that the extracted features accurately reflect the user's current interests and purchase intentions.

[0029] The extracted return behavior interest information and return purchase intention information are analyzed to generate the return behavior interest index and return purchase intention coefficient respectively.

[0030] In this embodiment, the logic for obtaining the return behavior interest index is as follows: Extract the return behavior interest information from the preprocessed return behavior new product browsing information, including the number of times users browse new products in different time periods during the return period, the time they stay on new products, and the ratio of the number of clicks to the number of impressions of new products, and mark them as , and , Indicates that the user is during the return period Number of views of new products in a certain period of time, Indicates that the user is during the return period The dwell time of the new product in the time period, Indicates that the user is during the return period The ratio of clicks to impressions for new products in a given time period. , is a positive integer; To obtain "the number of times users browse new products in different time periods during the reflow period, the length of time they stay on new products, and the ratio of clicks to impressions of new products", this can be achieved by integrating a user behavior data collection and analysis system into the e-commerce platform. This system needs to monitor every page visit and interaction of users in real time and record relevant data in the background. Specifically, the number of views can be obtained by recording the event of each user visiting a certain product page. Each time a user loads a product details page, the system will increase the browsing counter of the product and associate it with the user's unique identifier (such as user ID). The dwell time can be obtained by calculating the time period that a user stays on a certain product page. This is usually done by starting the timer after the user enters the product details page and recording the length of time the user stays on the page until the user leaves the page or jumps to another page. The ratio of clicks to impressions (click-through rate) can be obtained by recording the number of times the product is displayed when the product page is loaded, increasing the click count each time a user clicks on the product, and finally calculating the ratio of clicks to impressions. All of this data will be captured in real time through event tracking systems (such as JavaScript event listeners) when user behavior occurs, and transmitted to the backend database for storage and management through data collection systems (such as Google Analytics, Mixpanel, etc.). Then, the system will organize and count these raw data, cut and classify the data according to time periods (such as every hour, every half hour, etc.) for subsequent analysis and calculation. Ultimately, these processed data will be used to generate return behavior interest information, and evaluate the user's interest in new products by calculating browsing frequency, dwell time, and click-through rate.

[0031] The user's stay time for the new product in different time periods during the return period is constructed into a set, and the maximum value in the set is marked as ; Calculate the return behavior interest index. The specific calculation formula is as follows: In the formula, The index of interest in return behavior.

[0032] The return behavior interest index The calculation formula of uses a combination of logarithmic and exponential operations to comprehensively and accurately measure the user's interest in new products during the return period. First, use exponential operations The purpose of processing the dwell time is to emphasize the impact of products with longer dwell time on user interest. Dwell time reflects the user's in-depth attention to a product. Longer dwell time usually means higher interest. Using exponential calculation can nonlinearly amplify the intensity of this interest and highlight the user's deep interaction. The logarithmic operation is used to balance the relationship between the number of items viewed by users and the intensity of their interest. The logarithmic function can suppress the excessive impact of high-frequency browsing on the interest index, while ensuring that frequently viewed items are properly reflected in the index. The calculation method is designed to amplify the impact of click-through rate on interest while avoiding division by zero errors. Click-through rate itself is a direct expression of user interest in a product. A high click-through rate means that the user is highly interested in the product. Therefore, the contribution of products with high click-through rates to the final interest index is enhanced through logarithmic operations. In addition, the average calculation in the formula can average user interests at different stages in multiple time periods, further ensuring a comprehensive evaluation of user behavior during the reflow period. In short, through these calculation methods, RII can better reflect the user's comprehensive interest in new products during the reflow period, ensuring that the recommendation system can accurately capture changes in user interests and make corresponding adjustments.

[0033] The size of the reflow behavior interest index (RII) directly reflects the user's interest in new products during the reflow period. Therefore, there is a close relationship between the value of RII and "evaluating whether the user's reflow behavior is effective." When the RII value is high, it means that the user shows a strong interest in the new product during the reflow period, which usually means that the user has a high purchase potential and strong engagement when reflowing. Therefore, the system will consider the reflow behavior to be effective and adjust the recommendation strategy accordingly, push more products that meet the user's interests, and improve the accuracy of the recommendation. On the contrary, when the RII value is low, it indicates that the user has a weak interest in the new product, which may be just an accidental browsing or unintentional interaction. The system will judge that the reflow behavior is invalid or has low interest, and the recommendation system will reduce the recommendation of the product or maintain the existing recommendation strategy. Therefore, the level of RII provides a quantitative basis for the system. By evaluating this value, the system can effectively distinguish which reflow behaviors truly reflect the user's current needs, and then optimize the recommended content and improve the platform's conversion rate and user experience.

[0034] In this embodiment, the logic for obtaining the return purchase intention coefficient is as follows: Extract the return purchase intention information from the preprocessed return behavior new product browsing information, including the number of times users add new products to the shopping cart at different time periods during the return period, the conversion rate of purchasing new products, and the number of times new products are collected, and mark them as , and , Indicates that the user is during the return period The number of times new products are added to the shopping cart during a certain period of time. Indicates that the user is during the return period Conversion rate of new products purchased in a certain time period, Indicates that the user is during the return period The number of times a new product is collected in a certain period of time. , is a positive integer; To obtain the number of times users add new products to their shopping carts during different time periods during the reflow period, the conversion rate of new products, and the number of times new products are collected, this can be achieved by integrating user behavior data collection and event tracking systems in the e-commerce platform. First, the number of times a product is added to the shopping cart can be calculated by the system recording each event of adding the product to the shopping cart in real time in the background when the user clicks on the product and adds it to the shopping cart. Every time a user clicks the "Add to Cart" button on the product page, the system records the event and marks each product with a user ID and timestamp, and groups the behavior according to time periods (such as hours, days, etc.) to calculate the number of additions in each time period. Secondly, the purchase conversion rate is obtained through purchase data. The system records the number of views and purchases of each product, and then calculates the conversion rate by calculating the ratio of the number of purchases to the number of views of the product during the reflow period. Products with high conversion rates usually indicate that users have strong purchasing intentions during the reflow period. Finally, the number of collections can be recorded by the event of the user clicking the "Add to Favorites" or "Add to Wishlist" button of the product. Each time a user adds a product to the favorites, the system will update the number of collections of the product in real time, and count it by time period, and finally obtain the collection behavior data of the user for the new product during the return period. These data can be captured in real time through event tracking systems (such as JavaScript event listeners or custom tracking codes), and stored and analyzed through data analysis platforms (such as Google Analytics, Mixpanel or self-built data warehouses). The system will associate these behavioral data with user ID, time period and product ID to conduct detailed analysis and evaluation of each user's behavior during the return period, and finally help generate the return purchase intention coefficient. In this way, the platform can accurately capture and record users' shopping behavior in real time without interfering with the user experience.

[0035] The conversion rate of users purchasing new products in different time periods during the return period is constructed into a set, and the maximum value in the set is marked as ; Calculate the return purchase intention coefficient. The specific calculation formula is as follows: In the formula, is the return purchase intention coefficient.

[0036] Return purchase intention coefficient The calculation formula of is designed to more accurately measure the user's purchase intention during the return period through a combination of exponential and logarithmic operations, and comprehensively consider the impact of shopping cart behavior, purchase conversion rate and collection behavior on purchase potential. First, By analyzing the purchase conversion rate ( ) to perform exponential calculations, emphasizing that when the conversion rate is high, the user's purchase intention is stronger. This operation amplifies the impact of high conversion rate on return purchase intention and avoids the interference of low conversion rate on intention evaluation. A standardized method was used to handle the number of favorites ( ), which makes the collection behavior have a nonlinear amplification effect in the calculation process, but avoids exaggerating the impact of frequent collection of goods on purchase intention. This makes the collection behavior play a reasonable auxiliary role in the intention assessment, and will not be distorted by the extreme value of the number of collections. Finally, Logarithmic calculation is used to measure the number of shopping cart additions ( ), preventing high-frequency shopping cart additions from overly affecting the calculation results, while ensuring that frequently added items are given higher weights when evaluating purchase intentions. Overall, the formula can accurately capture the purchase potential of users when they return by integrating the weighting, standardization, and nonlinear amplification of these behavioral data, and avoid the deviation of a single indicator in evaluating purchase intentions, thereby improving the accuracy and effectiveness of the recommendation system.

[0037] Return purchase intention coefficient There is a direct relationship between the size of and "evaluating whether the user's return behavior is effective". A high value means that the user has shown a strong purchase intention during the return period. This is usually because the user frequently adds items to the shopping cart, has a high conversion rate, and has a large number of collections, indicating that the user has a strong purchase potential. Therefore, the system will consider the return behavior to be effective and push more products that meet the user's current interests and purchase intentions. On the contrary, if A low value indicates that the user's purchase intention is weak during the return period, and the user may just browse by chance or have no clear purchase plan. This indicates that the return behavior is ineffective or the interest is low, and the system will reduce the recommendation of the product or maintain the original recommendation strategy. The system can quantify the user's purchasing potential and thus determine whether the return behavior is effective, helping to accurately adjust the recommendation strategy, ensuring that the products seen by users are more in line with their actual needs, and improving the platform's conversion rate and user satisfaction.

[0038] Construct an effective evaluation model for the generated return behavior interest index and return purchase intention coefficient to generate an effective evaluation coefficient; In this embodiment, the generated return behavior interest index and return purchase intention coefficient Construct an effective evaluation model and generate an effective evaluation coefficient through weighted summation. The specific calculation formula is as follows: In the formula, is the effective evaluation coefficient, and Recurrent Behavior Interest Index and return purchase intention coefficient The non-zero weight coefficient of .

[0039] The calculation of the effective evaluation coefficient (EEC) combines two important indicators, the return behavior interest index (RII) and the return purchase intention coefficient (RPIC), through weighted summation, with the aim of comprehensively evaluating the effectiveness of user return behavior. The return behavior interest index (RII) measures the user's interest in new products and reflects the user's attention when returning; while the return purchase intention coefficient (RPIC) evaluates the user's purchase potential and indicates the user's actual purchase intention for the product during the return period. These two indicators complement each other and together provide a more accurate user behavior evaluation.

[0040] When calculating EEC, two non-zero weight coefficients are used ( and ) to adjust the influence of RII and RPIC in the effective evaluation coefficient. represents the weight of the return behavior interest index, controlling for the contribution of RII to EEC; Represents the weight of the return purchase intention coefficient, controlling the contribution of RPIC to EEC. The sum of the two weight coefficients must be equal to 1, that is, , which means that the influence of RII and RPIC in the evaluation is relative, but the sum is always 100%. By adjusting these two weight coefficients, the calculation of EEC can be optimized according to business needs or historical data, and the importance of user interest and purchase intention in the evaluation of return behavior can be flexibly adjusted. For example, if the system attaches more importance to the user's purchase potential, it can be appropriately increased. If the user's interest strength is more critical, you can increase In this way, EEC can flexibly adapt to different business scenarios, helping the system to more accurately evaluate the effectiveness of user return behavior and thus optimize the recommendation strategy.

[0041] Analyze the generated effective evaluation coefficient to evaluate whether the user's recirculation behavior is effective, and generate effective signals and ineffective signals respectively according to the evaluation results; In this embodiment, the preset effective evaluation coefficient threshold is determined , and after determination, the effective evaluation coefficient generated Compare and evaluate whether the user's return behavior is effective based on the comparison results. The specific comparison analysis is as follows: like , the user's return behavior is invalid; This means that the return behavior fails to meet the expected user interest or purchase potential, indicating that the user's interest in the product or product category is weak, or their purchase intention has not been effectively stimulated. In this case, the system will stop or reduce the recommendation of the product to avoid wasting recommendation resources and avoid pushing products that do not match the user's actual needs. This helps to improve the efficiency and accuracy of recommendations, ensure that system resources are concentrated on products that are more likely to lead to conversions, optimize the overall recommendation effect, and improve user experience and platform conversion rate.

[0042] like , the user's recirculation behavior is valid.

[0043] This situation indicates that the user has shown strong interest and purchase intention during the return period. The system will determine that the user's return behavior is effective and push products or recommended content that match their interests and purchase potential. At this time, the system will strengthen the recommendation of related products, increase the exposure frequency or push intensity of these products, thereby improving the conversion rate. By effectively identifying the user's interests and purchase intentions, the system can provide more personalized and accurate recommendations, improve the user's purchase experience, increase the interaction between users and the platform, and enhance the platform's sales efficiency and customer satisfaction.

[0044] The pre-set effective evaluation coefficient threshold can be determined by a data-driven approach, and is usually optimized based on historical data analysis and business goals. First, it can be preliminarily set using statistical analysis and machine learning models based on user behavior data collected by the platform (such as browsing, purchasing, collecting, adding to shopping cart, etc.) and related information on reflow behavior. Specifically, the system can find the dividing point between effective reflow behavior and ineffective reflow behavior by analyzing the effective evaluation coefficients of these two types of behaviors based on the distribution of historical reflow data. Usually, this threshold is based on the statistical distribution of the reflow behavior interest index (RII) and the reflow purchase intention coefficient (RPIC). For example, by calculating the mean and standard deviation of these indicators, the threshold is set above the distribution so that the evaluation coefficients of most effective reflow behaviors are above the threshold. In addition, the setting of the threshold can also be verified and optimized through A / B testing. In actual operation, the system can conduct experiments on the setting of different thresholds, evaluate their impact on user behavior and recommendation effect, and gradually adjust and optimize the threshold to achieve the best recommendation effect and platform conversion rate. During the experiment, the platform can monitor the conversion rate, click-through rate, purchase rate and other key indicators of the return behavior in real time, and continuously adjust the threshold according to the actual results to ensure that it can effectively distinguish between the effective return behavior and the invalid return behavior of the user. Through such data analysis and experimental optimization, the system can automatically determine the most appropriate effective evaluation coefficient threshold to a certain extent, ensuring the accuracy of the recommendation system and user satisfaction.

[0045] In this embodiment, valid signals and invalid signals are generated respectively according to the evaluation results, specifically: If the evaluation result shows that the user's recirculation behavior is invalid, an invalid signal is generated; To achieve the generation of ineffective signals and their subsequent impact, the recommendation strategy can be dynamically adjusted through the recommendation engine and user behavior analysis module in the system. Specifically, when the system determines that the return behavior is ineffective, it first triggers an ineffective signal by comparing the generated effective evaluation coefficient with the preset threshold. This signal can adjust the recommendation strength through the algorithm, that is, reduce the recommendation frequency of the product or remove it from the user's recommendation list. This can be achieved by modifying the recommendation weight of the product or reducing the relevance of the product in the recommendation model. For example, if the return behavior does not show enough interest or purchase intention, the system will reduce the display of the product and concentrate resources on recommending other products that better meet the user's current needs, thereby improving the overall accuracy and efficiency of the recommendation system.

[0046] If the evaluation result shows that the user's recirculation behavior is valid, a valid signal is generated.

[0047] The generation of effective signals is achieved by increasing the strength of product recommendations for effective return behaviors. The specific approach is to score the interest and purchase potential of return behaviors in the system, and after generating effective signals, increase the recommendation strength of related products based on the signals. For example, the system can increase the weight of products in the recommendation algorithm so that they appear more frequently in the user's recommendation list, or increase the display frequency of products in prominent locations such as the home page and category pages. In addition, the system can also analyze the user's return behavior patterns through algorithms, and push more products that match their interests and purchase potential in real time when users return. This can improve the accuracy of recommendations, enhance user satisfaction, and promote the platform's conversion rate and user stickiness.

[0048] In the case of generating a valid signal, continuously obtaining a number of valid evaluation coefficients generated subsequently, analyzing them after generation, generating signals of different levels, and taking corresponding recommended measures according to the generated signals of different levels; In this embodiment, when a valid signal is generated, several valid evaluation coefficients generated subsequently are continuously obtained and recalibrated as , is the number of valid evaluation coefficients subsequently generated when a valid signal is generated, , is a positive integer; To continuously obtain several subsequent effective evaluation coefficients, real-time data stream processing and event-driven architecture can be used. This process can be achieved by integrating a real-time data monitoring and collection module in the platform's recommendation system, which continuously captures and records the effective evaluation coefficients in each reflow cycle when the user's reflow behavior occurs. Specifically, the system automatically calculates and generates the corresponding effective evaluation coefficients each time the user performs a reflow operation (such as browsing products, adding to shopping carts, clicking to buy, etc.), and stores these coefficients in the background database. In order to ensure the real-time nature of the data, the system can transmit the effective evaluation coefficients generated each time to the data processing layer in real time through a message queue or a streaming data processing platform (such as Apache Kafka, Apache Flink, etc.). In this way, the system can continuously and quickly obtain new effective evaluation coefficients and use them for subsequent analysis and recommendation strategy adjustments. In addition, the system can set up a sliding window mechanism to continuously track and update these effective evaluation coefficients within a certain time range (such as every hour, every day). Whenever a new round of reflow behavior data is generated, the system automatically processes and analyzes the newly generated evaluation coefficients together with the previous coefficients to ensure the dynamic and timeliness of the evaluation process. In this way, the system can ensure that effective evaluation coefficients are obtained in real time and continuously when valid signals are generated, providing accurate data support for subsequent analysis and signal generation.

[0049] Calculate several subsequent effective evaluation coefficients The average , according to the formula: ; Calculate several subsequent effective evaluation coefficients Standard Deviation , according to the formula: .

[0050] In this embodiment, the preset average value threshold of several effective evaluation coefficients generated subsequently when a valid signal is generated is determined. and preset standard deviation threshold , and after determination, respectively with the average value of several effective evaluation coefficients and standard deviation Perform a comparison and generate different levels of signals based on the comparison results. The specific comparison analysis is as follows: like and , generating a first level signal; The first level signal (when the average value is greater than or equal to the preset average value threshold, and the standard deviation is greater than or equal to the preset standard deviation threshold) means that the user shows a strong and stable interest in the product during the reflow period. The system will recognize this signal and regard it as the user has a high purchase intention and strong demand. This situation usually means that the reflow behavior is very effective, and the user's interest is relatively concentrated and sustained. The system should strengthen the recommendation, increase the exposure frequency of related products, or increase the weight of the product in the recommendation algorithm to ensure that users can receive more product recommendations related to their interests, thereby improving conversion rate and sales efficiency.

[0051] like and , generating a second level signal; The second level signal (when the average value is less than the preset average value threshold, and the standard deviation is greater than or equal to the preset standard deviation threshold) means that the user's return behavior shows weak interest but greater volatility. This situation indicates that the user's interest in the product may not be stable when returning, and may be affected by external factors (such as promotions, advertising push, etc.). In this case, the system should reduce the frequency of recommendations and adjust the recommendation strategy appropriately to avoid excessive push of products that do not match the user's fluctuating interests. By reducing the intensity of recommendations, it is possible to avoid pushing products too frequently and causing user disgust, while ensuring that platform resources are not excessively wasted on uncertain interests.

[0052] like and , generating a third level signal; The third level signal (when the average value is greater than or equal to the preset average value threshold, and the standard deviation is less than the preset standard deviation threshold) means that the user shows a strong but relatively stable interest during the reflow period. This situation shows that the user's interest during the reflow period is concentrated and stable, and although the interest intensity is high, the fluctuation is small. For this situation, the system can moderately push products related to the user's interests instead of significantly increasing the recommendation intensity. Through appropriate recommendation strategies, the system can ensure the continuous provision of products that meet the user's stable interests, avoid excessive push, and improve the user's purchasing experience and platform conversion rate.

[0053] like and , generating a fourth level signal.

[0054] The fourth level signal (when the average value is less than the preset average value threshold, and the standard deviation is less than the preset standard deviation threshold) means that the user's return behavior interest is weak and relatively stable. In this case, the user's return behavior does not show obvious changes in interest, and the intensity of his interest is low. The system should judge the return behavior as invalid return. Based on this signal, the system will reduce or stop recommending related products to avoid wasting recommendation resources. Through this adjustment, the platform can ensure the efficiency of the recommendation system while ensuring that users do not receive too many irrelevant product recommendations, thereby improving the accuracy of the recommendation system and user satisfaction.

[0055] Determine the preset mean value threshold and the preset standard deviation threshold of several effective evaluation coefficients subsequently generated when the effective signal is generated, which can be dynamically set through data analysis and machine learning methods. First, the system can analyze the distribution of effective evaluation coefficients (EEC) based on the data set of historical user reflow behaviors. By calculating the mean value and standard deviation of reflow behaviors in historical data, the system can obtain benchmark values, which can reflect the overall trend and fluctuation degree of reflow behaviors under normal circumstances. On this basis, the system can adjust these benchmark values ​​to ensure the rationality and applicability of the thresholds. The specific method is that the system first statistically analyzes the effective evaluation coefficients in historical data to obtain the mean and standard deviation of the EEC value in each time period. Then, according to the fluctuation range of historical data, the preset standard deviation threshold is set, which is usually set to 1-2 times the range of the benchmark value to ensure that most reflow behaviors within the normal range can be covered. The preset mean value threshold can be set by the median or quantile in the historical data, or optimized by A / B testing, that is, in actual operation, the most suitable value is adjusted by experimenting with the impact of different thresholds on the recommendation effect. In addition, the system can use a dynamic adjustment mechanism to fine-tune the threshold according to the effective evaluation coefficient generated in real time. As the platform's user behavior and return patterns continue to change, the system can regularly update data models and threshold adjustment strategies to adapt to changes in user behavior, ensuring that the thresholds always accurately reflect the user's current interests and purchasing potential. In this way, the system can optimize the threshold settings based on historical data and real-time feedback to ensure the accuracy and personalization of recommendations.

[0056] In this embodiment, corresponding recommended measures are taken according to the generated signals of different levels, specifically: If a first-level signal is generated, the recommended measures taken are: increase the recommendation frequency, improve the exposure of new products in the user's recommendation list, and push them to the platform's homepage and popular recommendations to ensure that users can see more products that match their interests; When the first-level signal is generated, the system will increase the exposure of the product in the user's recommendation list by increasing the recommendation frequency of the product. The specific implementation method is to adjust the product weight in the recommendation algorithm so that the product can obtain a higher ranking in the recommendation engine. The system can push the product to prominent positions such as home page recommendations, popular recommendations, and related products by adjusting the weight parameters in the recommendation algorithm in real time, thereby increasing its display probability on the user's page. In addition, the recommendation system can also use user interest models and personalized recommendation algorithms to ensure that these products are highly matched with the user's current interests through analysis of historical behaviors and interests, thereby increasing the click-through rate and purchase conversion rate of the products. The purpose of this is to ensure that when users return, they can see more products that meet their interests and needs, thereby increasing user engagement and the platform's conversion rate.

[0057] If a second-level signal is generated, the recommended measures are: reduce the frequency of recommendations, keep recommending new products, but lower their display priority and frequency, and recommend them to user groups who are uncertain about their interest in the product, thereby avoiding excessive recommendations of invalid products; When the second-level signal is generated, the system will reduce the frequency of recommendations, maintain the recommendation of new products, but reduce the display priority and frequency of the products. The specific implementation method is to adjust the weight in the recommendation algorithm to reduce the recommendation frequency of the product and reduce its display opportunities in the homepage recommendations and related products. The system can recommend the product to user groups who are not clear about their interest in the product, or push it to a lower priority recommendation position. For example, through the personalized algorithm, it is analyzed that the product has a weak relationship with the interests of some users, so that it will appear in the recommendation list of these users. A lower weight. This can avoid excessively recommending invalid products to users, ensure that the resources of the recommendation system are used efficiently, and avoid disturbing users with irrelevant product recommendations, thereby improving the accuracy of recommendations.

[0058] If a third-level signal is generated, the recommended measures are: maintain the existing recommendation frequency and provide new product recommendations based on the user's current interest intensity. New products can be presented to users in a relatively stable recommendation manner to ensure that users continue to obtain products that meet their interests; When the third-level signal is generated, the system will maintain the existing recommendation frequency to ensure that the product is displayed to the user at a moderate frequency in the recommendation list. This means that the recommendation strength of the product remains unchanged and will not increase or decrease, ensuring that the product is in a stable recommendation state. The specific implementation method is through real-time analysis of the interest and purchase intention of the product. The system calculates the user's interest strength and then determines the recommendation frequency of the product based on the strength. If the user's interest in the product remains stable during the reflow period, the system will maintain the frequency of the product in the recommendation list without excessive intervention, ensuring that the user can continue to see the product within the normal recommendation range. This method ensures the balance of the recommendation system, neither over-pushing products nor ignoring users' long-term interests, thereby improving user satisfaction and maintaining a relatively stable conversion rate.

[0059] If a fourth-level signal is generated, the recommended measures taken are: stop recommending and remove the new product from the user's recommendation list to avoid invalid recommendations, concentrate resources on recommending products that better meet the user's current needs, and improve the system's recommendation efficiency.

[0060] When a fourth-level signal is generated, the system stops recommending the product and removes it from the user's recommendation list. This is achieved by adjusting the weights in the recommendation algorithm to reduce the product's recommendation weight to zero, or completely removing it from related products and recommended positions. In order to prevent invalid product recommendations from interfering with users, the system can determine inefficient product recommendations through backflow behavior analysis and stop pushing them, thereby saving recommendation resources and focusing on products that are more in line with user interests. The core purpose of this approach is to ensure that system resources are concentrated on high-potential products, improve the efficiency of the recommendation system, prevent users from receiving products that do not match their interests, and improve the platform's user experience and the accuracy of recommendations.

[0061] Continuously monitor users' return behavior and new product browsing information, and dynamically adjust the parameters and recommendation strategies of the return behavior effective evaluation model based on real-time monitoring results. At the same time, store and analyze historical data to optimize model parameters and recommendation strategies.

[0062] To continuously monitor the user's return behavior and the browsing information of new products in the return behavior, it can be achieved by integrating a real-time data collection and analysis system in the e-commerce platform and combining it with a dynamically adjusted recommendation algorithm. Specifically, the system will continuously capture the relevant data of each user's return behavior, such as browsing time, number of clicks, length of stay, etc., through the user behavior tracking module in the background, and transmit it to the data processing platform in real time. The system will use event-driven architecture and streaming data processing platforms (such as Apache Kafka or Apache Flink) for real-time processing to ensure that the data of each user return and browsing behavior can be captured and analyzed in a timely manner.

[0063] Based on real-time monitoring, the system will dynamically adjust the parameters of the effective evaluation model of the feedback behavior, such as optimizing the model by adjusting the weights of the interest index and the purchase intention coefficient to ensure that the model can respond to changes in user interests and purchase behaviors in real time. The recommendation system will update the parameters in the algorithm based on real-time data, improve the recommendation logic, and enhance the accuracy of the recommendation. At the same time, the system will also store these real-time data in the database, conduct historical data analysis, analyze the changing trends of user behavior patterns, and optimize parameter settings and recommendation strategies. By storing and analyzing historical data, the system can identify long-term effective user behavior characteristics, further improve the feedback behavior evaluation model, and improve the model's predictive ability and the accuracy of the recommendation strategy. The purpose of this is to ensure that the recommendation system can continue to adapt to changes in user needs, improve recommendation efficiency, and enhance user experience and platform conversion rates.

[0064] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0065] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0066] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0067] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0068] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0069] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0070] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0071] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A marketing method based on intelligent commodity background matching, characterized in that: The specific steps include: By real-time monitoring of the browsing and purchasing behavior of the user on the e-commerce platform, the user's return behavior is identified. When the user's return behavior is identified, whether the user browses a product that has not been previously viewed during the return behavior is detected in real time, and when it is detected that the user browses a product that has not been previously viewed during the return behavior, the following steps are performed; Real-time acquisition of new product browsing information of return behavior generated when users browse products that they have not paid attention to before during the return behavior period, and analysis after acquisition to generate return behavior interest index and return purchase intention coefficient respectively; Construct an effective evaluation model for the generated return behavior interest index and return purchase intention coefficient to generate an effective evaluation coefficient; Analyze the generated effective evaluation coefficient to evaluate whether the user's recirculation behavior is effective, and generate effective signals and ineffective signals according to the evaluation results; In the case of generating a valid signal, continuously obtaining a number of valid evaluation coefficients generated subsequently, analyzing them after generation, generating signals of different levels, and taking corresponding recommended measures according to the generated signals of different levels; Continuously monitor users' return behavior and new product browsing information, and dynamically adjust the parameters and recommendation strategies of the return behavior effective evaluation model based on real-time monitoring results. At the same time, store and analyze historical data to optimize model parameters and recommendation strategies.

2. The marketing method based on intelligent commodity background matching according to claim 1, characterized in that: The recurrent behavior new product browsing information generated when the user browses products that have not been previously viewed during the recurrent behavior is obtained in real time, and analyzed after acquisition to generate a recurrent behavior interest index and a recurrent purchase intention coefficient, respectively. Specifically, the following steps are included: Real-time acquisition of new product browsing information generated by users browsing products that they have not previously viewed during the recurrent behavior, and pre-processing after acquisition; Extracting return behavior interest information and return purchase intention information from the preprocessed return behavior new product browsing information; The extracted return behavior interest information and return purchase intention information are analyzed to generate the return behavior interest index and return purchase intention coefficient respectively.

3. The marketing method based on intelligent commodity background matching according to claim 2 is characterized in that: The logic for obtaining the return behavior interest index is as follows: Extract the return behavior interest information from the preprocessed return behavior new product browsing information, including the number of times users browse new products in different time periods during the return period, the time they stay on new products, and the ratio of the number of clicks to the number of impressions of new products, and mark them as , and , Indicates that the user is during the return period Number of views of new products in a certain period of time, Indicates that the user is during the return period The time period for new products, Indicates that the user is during the return period The ratio of clicks to impressions for new products in a given time period. , is a positive integer; The user's stay time for the new product in different time periods during the return period is constructed into a set, and the maximum value in the set is marked as ; Calculate the return behavior interest index. The specific calculation formula is as follows: In the formula, The index of interest in return behavior.

4. The marketing method based on intelligent commodity background matching according to claim 3 is characterized in that: The logic for obtaining the return purchase intention coefficient is as follows: Extract the return purchase intention information from the preprocessed return behavior new product browsing information, including the number of times users add new products to the shopping cart at different time periods during the return period, the conversion rate of purchasing new products, and the number of times new products are collected, and mark them as , and , Indicates that the user is during the return period The number of times new products are added to the shopping cart during a certain period of time. Indicates that the user is during the return period Conversion rate of new products purchased in a certain time period, Indicates that the user is during the return period The number of times a new product is collected in a certain period of time. , is a positive integer; The conversion rate of users purchasing new products in different time periods during the return period is constructed into a set, and the maximum value in the set is marked as ; Calculate the return purchase intention coefficient. The specific calculation formula is as follows: In the formula, is the return purchase intention coefficient.

5. The marketing method based on intelligent commodity background matching according to claim 4 is characterized in that: Interest index for the generated recurrent behavior and return purchase intention coefficient Construct an effective evaluation model and generate an effective evaluation coefficient through weighted summation. The specific calculation formula is as follows: In the formula, is the effective evaluation coefficient, and Recurrent Behavior Interest Index and return purchase intention coefficient The non-zero weight coefficient of .

6. The marketing method based on intelligent commodity background matching according to claim 5, characterized in that: Determine the pre-set effective evaluation coefficient threshold , and after determination, the effective evaluation coefficient generated Compare and evaluate whether the user's return behavior is effective based on the comparison results. The specific comparison analysis is as follows: like , the user's return behavior is invalid; like , the user's recirculation behavior is valid.

7. The marketing method based on intelligent commodity background matching according to claim 6, characterized in that: According to the evaluation results, valid signals and invalid signals are generated respectively, specifically: If the evaluation result shows that the user's recirculation behavior is invalid, an invalid signal is generated; If the evaluation result shows that the user's recirculation behavior is valid, a valid signal is generated.

8. The marketing method based on intelligent commodity background matching according to claim 7, characterized in that: When a valid signal is generated, several valid evaluation coefficients generated subsequently are continuously obtained and recalibrated as , is the number of valid evaluation coefficients subsequently generated when a valid signal is generated, , is a positive integer; Calculate several subsequent effective evaluation coefficients The average , according to the formula: ; Calculate several subsequent effective evaluation coefficients Standard Deviation , according to the formula: .

9. The marketing method based on intelligent commodity background matching according to claim 8, characterized in that: Determine the preset average value threshold of several valid evaluation coefficients generated subsequently when a valid signal is generated and preset standard deviation threshold , and after determination, respectively with the average value of several effective evaluation coefficients and standard deviation Perform a comparison and generate different levels of signals based on the comparison results. The specific comparison analysis is as follows: like and , generating a first level signal; like and , generating a second level signal; like and , generating a third level signal; like and , generating a fourth level signal.

10. The marketing method based on intelligent commodity background matching according to claim 9, characterized in that: Take corresponding recommended measures according to the different levels of signals generated, specifically: If a first-level signal is generated, the recommended measures taken are: increase the recommendation frequency, improve the exposure of new products in the user's recommendation list, and push them to the platform's homepage and popular recommendations; If a second-level signal is generated, the recommended measures are: reduce the frequency of recommendations, keep recommending new products, but reduce their display priority and frequency; If a third-level signal is generated, the recommended measures are: maintain the existing recommendation frequency and provide new product recommendations based on the user's current interest intensity; If a fourth-level signal is generated, the recommended action taken is: stop recommending and remove the new product from the user's recommendation list.

Images