Precision Marketing Recommendation System Based on AI Large Model

Abstract

The present invention discloses a precise marketing recommendation system based on an AI large model, which relates to the technical field of information recommendation and is used to improve the problem of the decline in product information recommendation efficiency caused by the delay of recommendation tags. It includes collecting user data, analyzing the frequency modulation potential of user information using the K-nearest neighbor algorithm for classification, screening out users with high frequency modulation potential, setting a detection time to collect the recommended information data of users, screening out key tags in the recommended information data and calculating the tag demand coefficient, detecting the user's clearing behavior and performing secondary screening on the key tags to obtain evaluation tags, using the modified geometric mean method to determine whether to perform a demand reduction process on the evaluation tags, marking the evaluation tags subjected to the demand reduction process, obtaining the tag attributes of the marked tags, comprehensively calculating the reduction coefficients of different marked tags based on the tag attributes of the marked tags and the tag demand coefficient, and setting a skip time for the marked tags according to the reduction coefficients.

Description

Technical Field

[0001] The present invention relates to the technical field of information recommendation, and more specifically, to a precise marketing recommendation system based on an AI large model. Background Art

[0002] Information recommendation technology uses data analysis and algorithms to recommend products or services to users on a marketing platform, thereby improving user satisfaction. When information recommendation technology is applied to a marketing recommendation system, product information can be accurately recommended to high-demand users.

[0003] The prior art has the following deficiencies:

[0004] In the past, when a marketing platform recommended product information to users, it determined user needs through user behavior for precise recommendation, without considering the different timeliness of products under different tags. When the user's needs changed, a large number of products in the past tags were still provided to the user, resulting in a decline in the efficiency of product information recommendation. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a precise marketing recommendation system based on an AI large model, which evaluates the user's frequency modulation potential by analyzing the degree of user demand transfer and classifies the users, combines the user category and user behavior to screen product tags multiple times, and sets different skip times for different product tags to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A precise marketing recommendation system based on an AI large model includes a user data collection module, a frequency modulation potential analysis module, a tag demand evaluation module, and an information frequency modulation processing module;

[0008] The user data collection module is used to collect user data and transmit the user data to the frequency modulation potential analysis module. After receiving the user classification result transmitted back by the frequency modulation potential analysis module, the user data collection module sets a detection time to collect recommended information data and transmits it to the tag demand evaluation module;

[0009] The frequency modulation potential analysis module analyzes the information frequency modulation potential of users using the K-nearest neighbor algorithm based on the user data transmitted by the user data collection module and classifies the users, and transmits the user classification result back to the user data collection module;

[0010] The tag demand evaluation module is used to receive recommended information data, screen out key tags according to the recommended information data and calculate the corresponding tag demand coefficient, detect the user's clearing behavior and perform secondary screening on the key tags to obtain evaluation tags, use the modified geometric mean method to determine whether to adjust the demand of the evaluation tags downward, mark the evaluation tags that are to be adjusted downward, and send the tag demand coefficient of the marked tag to the information frequency modulation processing module;

[0011] After receiving the marking tag, the information frequency modulation processing module obtains the tag attribute of the marking tag, calculates the down-adjustment coefficient of the corresponding marking tag based on the tag attribute of the marking tag and the tag requirement coefficient, and sets the skip time for the marking tag according to the down-adjustment coefficient.

[0012] In a preferred embodiment, the user data in the user data collection module includes the number of tag categories occupied by the user's search keywords and the browsing time of tags of different categories;

[0013] When a user enters a search keyword through the search bar, it will be recorded in the search history. In the search history, the search keyword entered by the user is identified and the tag category to which it belongs is determined. The total number of different tag categories is counted to obtain the number of tag categories occupied by the user's search keyword;

[0014] When a user browses in the marketing platform, the tag where the user's browsing information is located is identified and the browsing time is recorded to obtain the browsing time of different categories of tags.

[0015] In a preferred embodiment, the frequency modulation potential analysis module comprehensively analyzes the user's information frequency modulation potential using the K nearest neighbor algorithm based on the number of tag categories occupied by the user's search keywords and the browsing time of different categories and classifies the user. The specific steps are as follows:

[0016] Collect the number of tag categories and browsing time of different categories of search keywords of multiple users, and perform standardization processing respectively, merge the number of tag categories of search keywords into a search data set, merge the browsing time of different categories into a browsing data set, take the ratio of the data in the search data set or the browsing data set to the maximum value in the search data set or the browsing data set as the standardization result of the corresponding data, randomly select a data in the search data set and the browsing data set as the reference data, set the nearest neighbor coefficient K, select K data with the smallest absolute value of the difference with the reference data in each data set as the marked data, calculate the average value of the marked data in the search data set and the browsing data set, and sum them to obtain the user classification threshold;

[0017] Sum the results of normalizing the number of user search keyword categories and the browsing time of different categories to obtain the user's information frequency modulation potential, and compare the information frequency modulation potential with the user classification threshold; when the user's information frequency modulation potential exceeds the user classification threshold, determine that the user is a high information frequency modulation potential user; when the user's information frequency modulation potential is lower than the user classification threshold, determine that the user is a low information frequency modulation potential user.

[0018] In a preferred embodiment, after classifying the user, the frequency modulation potential analysis module transmits the user classification result to the user data collection module. When the user is a high information frequency modulation potential user, the user data collection module sets the detection time to collect the recommended information data, and the recommended information data is the browsing frequency of the user under each tag.

[0019] The user data collection module selects a period of time as the detection time, records and counts the tags to which the recommended content clicked by the user belongs during the detection time, obtains the browsing frequency of the user under each tag during the detection time, and the user data collection module transmits the browsing frequency of the user under each tag to the tag demand evaluation module.

[0020] In a preferred embodiment, the tag demand evaluation module receives the browsing frequency of the user under each tag, selects the median of the browsing frequency under each tag as the screening threshold, screens out the tags whose browsing frequency exceeds the screening threshold as the key tags, and takes the ratio of the browsing frequency of the key tags to the screening threshold as the tag demand coefficient of the corresponding tag.

[0021] The user clearing behavior is that the user actively deletes the keywords in the search history during the detection time. Identify the tags where each keyword in the clearing list is located and match them with the key tags, retain the successfully matched tags as the evaluation tags, calculate the clearing ratio of the evaluation tags, and divide the sum of the number of keywords belonging to the same evaluation tag in the clearing list by the total number of keywords in the clearing list to obtain the tag clearing ratio of the corresponding evaluation tag.

[0022] In a preferred embodiment, the tag demand evaluation module comprehensively evaluates the tag demand coefficient and the tag clearing ratio of the evaluation tags, and uses the modified geometric mean method to determine whether to down-adjust the demand for the evaluation tags. The specific steps are as follows:

[0023] Data conversion: Convert the tag demand coefficient and the tag clearing ratio of the evaluation tags into the same dimension.

[0024] Repeated conversion: Set n conversion parameters to obtain multiple conversion results of the tag demand coefficient or the tag clearing ratio.

[0025] Data connection: Subtract the tag demand coefficient and the tag clearing ratio after data conversion using the same conversion parameter to obtain multiple connection values.

[0026] Connection processing: After performing logarithmic operations on multiple connection values, take the arithmetic mean as the connection processing quantity;

[0027] Antilogarithm processing: Take the antilogarithm of the connection processing quantity to obtain the demand evaluation value;

[0028] When the demand evaluation value of the evaluation label exceeds the preset evaluation threshold, no demand reduction processing is performed on the evaluation label; when the demand evaluation value of the evaluation label is lower than the preset evaluation threshold, demand reduction processing is performed on the evaluation label.

[0029] In a preferred embodiment, the label attributes of the marked label are the product shelf life and the evaluation star rating under the corresponding label; the evaluation star rating is the scoring system of the marketing platform, and different star ratings are set to evaluate different products.

[0030] In a preferred embodiment, when the information frequency modulation processing module calculates the reduction coefficient of the marked label, randomly select multiple products under the marked label as target products, collect the shelf life and evaluation star rating of each target product, calculate the average value of the shelf life of all target products as the label shelf life, and count the total number of evaluation star ratings of all target products as the label star value.

[0031] In a preferred embodiment, take the geometric mean of the normalized label shelf life and label star value of the marked label as the demand adjustment quantity, and calculate the reduction coefficient of the corresponding marked label through the harmonic formula: , where B is the label demand coefficient of the marked label, s is the preset harmonic parameter, h is the demand adjustment quantity of the marked label, and T is the reduction coefficient of the marked label;

[0032] Multiply the default skip time of the marked label by the reduction coefficient to obtain the skip time of the corresponding marked label.

[0033] Technical effects and advantages of the precise marketing recommendation system based on the AI large model of the present invention:

[0034] The present invention classifies users by collecting user data and analyzing their information frequency modulation potential, screens out users with high frequency modulation potential, and uses the screened users to determine the user direction, improving the processing efficiency. It sets a detection time to collect the recommended information data of users, screens out key tags in the recommended information data and calculates the tag demand coefficient, detects the user's clearing behavior, and performs a secondary screening on the key tags to obtain evaluation tags. It judges whether to perform a demand reduction process on the evaluation tags, marks the evaluation tags subjected to the demand reduction process, obtains the tag attributes of the marked tags, comprehensively calculates the reduction coefficients of different marked tags based on the tag attributes of the marked tags and the tag demand coefficient, and sets skip times for the marked tags according to the reduction coefficients, thereby reducing the recommendation of invalid tags or keywords and improving the accuracy of marketing recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the precise marketing recommendation system based on the AI large model of the present invention.

[0036] Figure 2 It is a flowchart of the precise marketing recommendation system based on the AI large model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0038] The present invention classifies users by collecting user data and analyzing their information frequency modulation potential, screens out users with high frequency modulation potential, sets a detection time to collect the recommended information data of users, screens out key tags in the recommended information data and calculates the tag demand coefficient, detects the user's clearing behavior, and performs a secondary screening on the key tags to obtain evaluation tags. It judges whether to perform a demand reduction process on the evaluation tags, marks the evaluation tags subjected to the demand reduction process, obtains the tag attributes of the marked tags, comprehensively calculates the reduction coefficients of different marked tags based on the tag attributes of the marked tags and the tag demand coefficient, and sets skip times for the marked tags according to the reduction coefficients, thereby reducing the recommendation of invalid tags or keywords and improving the accuracy of marketing recommendations.

[0039] Embodiment, a precise marketing recommendation system based on the AI large model, as Figure 1 and Figure 2 shown, includes a user data collection module, a frequency modulation potential analysis module, a tag demand evaluation module, and an information frequency modulation processing module, and the modules are connected by signals;

[0040] The functions of each module are as follows:

[0041] The user data collection module is used to collect user data and transmit the user data to the frequency modulation potential analysis module. After receiving the user classification result transmitted back by the frequency modulation potential analysis module, the user data collection module sets the detection time to collect the recommended information data and transmits it to the label requirement evaluation module;

[0042] The frequency modulation potential analysis module analyzes the information frequency modulation potential of users using the K-nearest neighbor algorithm based on the user data transmitted by the user data collection module, classifies the users, and transmits the user classification result back to the user data collection module;

[0043] The label requirement evaluation module is used to receive the recommended information data, screen out the key labels according to the recommended information data, calculate the corresponding label requirement coefficients, detect the user's clearing behavior, perform secondary screening on the key labels to obtain the evaluation labels, use the modified geometric mean method to judge whether to perform a demand reduction process on the evaluation labels, mark the evaluation labels that have undergone the demand reduction process, and send the label requirement coefficients of the marked labels to the information frequency modulation processing module;

[0044] After receiving the marked labels, the information frequency modulation processing module obtains the label attributes of the marked labels, comprehensively calculates the reduction coefficients of the corresponding marked labels based on the label attributes and label requirement coefficients of the marked labels, and sets the skip time for the marked labels according to the reduction coefficients.

[0045] It should be noted that the skip refund mechanism is used to perform secondary calibration on the skip times of some marked labels to improve the fault tolerance rate of marketing recommendations.

[0046] The user data in the user data collection module includes the number of user search keywords accounting for label categories and the browsing times of different category labels;

[0047] Users input search keywords through the search bar in the marketing platform to obtain their own demand information. In the relational database of the marketing platform, the label categories of all item information and all search keywords corresponding to different labels are stored. Whenever a user inputs a search keyword through the search bar, it will be recorded in the search history. The search keywords input by the user are identified in the search history and the corresponding label categories are determined. The total number of different marked categories is counted to obtain the number of user search keywords accounting for label categories;

[0048] When a user browses in the marketing platform, the label where the user's browsing information is located is identified and the browsing time is recorded to obtain the browsing times of different category labels.

[0049] The FM potential analysis module analyzes the information FM potential of users using the K-nearest neighbor algorithm by integrating the number of search keywords of users accounting for tag categories and the browsing time of different categories, and classifies the users. The specific steps are as follows:

[0050] Collect the number of search keywords of multiple users accounting for tag categories and the browsing time of different categories, and perform standardization processing respectively. Combine the number of search keywords accounting for tag categories into a search data set, and combine the browsing time of different categories into a browsing data set. Take the ratio of the data in the search data set or browsing data set to the maximum value in the search data set or browsing data set as the standardized result of the corresponding data. Randomly select a data in the search data set and browsing data set as the reference data, set the nearest neighbor coefficient K, and select the K data with the smallest absolute value of the difference from the reference data in each data set as the marked data. Calculate the average value of the marked data in the search data set and browsing data set and sum them to obtain the user classification threshold;

[0051] Sum the standardized results of the number of search keyword categories of users and the browsing time of different categories to obtain the information FM potential of users, and compare the information FM potential with the user classification threshold; when the information FM potential of users exceeds the user classification threshold, it is determined that the user is a high information FM potential user; when the information FM potential of users is lower than the user classification threshold, it is determined that the user is a low information FM potential user.

[0052] It should be noted that the higher the number of search keywords accounting for tag categories, the richer the recommended content for users, the more information FM is needed, and the higher the information FM potential; the longer the browsing time of different categories, the richer the recommended content for users, the more information FM is needed, and the higher the information FM potential. Information FM refers to removing or delaying some recommended information tags when there are too many recommended information tags; the nearest neighbor coefficient K can be set according to the actual situation. For example, the nearest neighbor coefficient K is set to 6, which will not be elaborated here.

[0053] After classifying the users, the FM potential analysis module transmits the user classification result to the user data collection module. When the user is a high information FM potential user, the user data collection module sets the detection time to collect recommended information data, and the recommended information data is the browsing frequency of users under each tag;

[0054] The user data collection module selects a period of time as the detection time, records and counts the tags to which the recommended content clicked by the user belongs during the detection time, obtains the browsing frequency of users under each tag during the detection time, and the user data collection module transmits the browsing frequency of users under each tag to the tag demand evaluation module.

[0055] It should be noted that the time interval of the detection time is set by professionals in this field and will not be analyzed here.

[0056] The label demand evaluation module receives the browsing frequencies of the user under each label, selects the median of the browsing frequencies under each label as the screening threshold, screens out the labels with browsing frequencies exceeding the screening threshold as key labels, and takes the ratio of the browsing frequency of the key label to the screening threshold as the label demand coefficient of the corresponding label;

[0057] The user's clearing behavior refers to the user's active deletion behavior of keywords in the search history within the detection time. Identify the labels where each keyword in the clearing list is located and match them with the key labels, retain the successfully matched labels as evaluation labels, calculate the clearing ratio of the evaluation labels, and divide the sum of the number of keywords belonging to the same evaluation label in the clearing list by the total number of keywords in the clearing list to obtain the label clearing ratio of the corresponding evaluation label.

[0058] It should be noted that the clearing list is a temporary storage station for storing deletion information. The keywords deleted by the user within the detection time can be obtained from the clearing list. Since some of the keywords in the clearing list may not be in the key labels, matching and retention are performed as described above.

[0059] The label demand evaluation module comprehensively evaluates the label demand coefficient and label clearing ratio of the evaluation labels, and uses the modified geometric mean method to determine whether to down-adjust the demand for the evaluation labels. The specific steps are as follows:

[0060] Data conversion: Perform data conversion on the label demand coefficient and label clearing ratio of the evaluation labels: , where is the label demand coefficient or label clearing ratio of the evaluation label, is the result after data conversion of the label demand coefficient or label clearing ratio of the corresponding evaluation label, is the conversion parameter used to convert the label demand coefficient and label clearing ratio into the same dimension;

[0061] Repeated conversion: Set n conversion parameters to obtain multiple conversion results of the label demand coefficient or label clearing ratio;

[0062] Data connection: Subtract the label demand coefficient and label clearing ratio after data conversion using the same conversion parameter to obtain multiple connection values;

[0063] Connection processing: Take the arithmetic mean of the logarithms of the multiple connection values as the connection processing quantity;

[0064] Antilogarithm processing: Take the antilogarithm of the connection processing quantity to obtain the demand evaluation value.

[0065] When the demand evaluation value of the evaluation label exceeds the preset evaluation threshold, no demand reduction processing is performed on the evaluation label; when the demand evaluation value of the evaluation label is lower than the preset evaluation threshold, demand reduction processing is performed on the evaluation label.

[0066] The label demand evaluation module marks the evaluation labels determined to undergo demand reduction processing and sends the label demand coefficients of the marked labels to the information frequency modulation processing module.

[0067] It should be noted that when setting the conversion parameters, it is necessary to meet the condition of avoiding the original data from becoming zero or negative after data conversion. The smaller the label demand coefficient, the smaller the demand for the evaluation label, and the larger the label clearing ratio, indicating that the user's demand for the label decreases and the demand for the evaluation label is smaller, and demand reduction processing is required; the demand reduction processing is to improve the recommendation accuracy by delaying the appearance time of the demand reduction label.

[0068] The label attributes of the marked labels are the product shelf life and evaluation star rating under the corresponding labels. The lower the product shelf life, the worse the product timeliness, and the easier it is for the marked labels to become invalid, and the more the skip time needs to be extended. The evaluation star rating is the scoring system of the marketing platform, and different star ratings are set to evaluate different products. The lower the cumulative star rating of each product under the same label, the worse the product timeliness, and the easier it is for the marked labels to become invalid, and the more the skip time needs to be extended.

[0069] Calculate the downward adjustment coefficient of the marked label and set the skip time based on the label attributes of the marked label and the label demand coefficient. The specific steps are as follows:

[0070] Randomly select multiple products under the marked label as target products, collect the shelf life and evaluation star rating of each target product, calculate the average value of the shelf life of all target products as the label shelf life, and count the total number of evaluation star ratings of all target products as the label star value;

[0071] Take the geometric mean of the normalized label shelf life and label star value of the marked label as the demand adjustment amount, and calculate the downward adjustment coefficient of the corresponding marked label through the harmonic formula: , where B is the label demand coefficient of the marked label, s is the preset harmonic parameter, h is the demand adjustment amount of the marked label, and T is the downward adjustment coefficient of the marked label;

[0072] Multiply the default skip time of the marked label by the downward adjustment coefficient as the skip time of the corresponding marked label. By setting the skip time for different marked labels, the recommendation of a large number of invalid labels or keywords can be reduced, improving the accuracy of marketing recommendations. At the same time, each marked label is recorded and saved for convenient subsequent management analysis and call.

[0073] It should be noted that the harmonic parameter is used to convert the downward adjustment coefficient into the skip time ratio for adjusting the marker label. The smaller the label demand coefficient of the marker label or the larger the demand adjustment amount, the larger the downward adjustment coefficient of the marker label, the lower the demand of the marker label, and the corresponding skip time of the marker label is extended. The default skip time is the time period when the marker label initially set in the marketing platform is not recommended. The harmonic parameter is set by professionals in the field, and the harmonic parameter is greater than 1 to ensure the skip time delay effect of the marker label.

[0074] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0075] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application of the technical solution and the invention constraints. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0076] In addition, the functional modules in each embodiment of this application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0077] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0078] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An accurate marketing recommendation system based on an AI large model, characterized in that It includes user data collection module, frequency modulation potential analysis module, label demand evaluation module and information frequency modulation processing module; The user data collection module is used to collect user data and transfer the user data to the frequency modulation potential analysis module. After receiving the user classification result returned by the frequency modulation potential analysis module, the user data collection module sets the detection time to collect the recommended information data and transfer it to the label demand evaluation module; The frequency modulation potential analysis module analyzes the user's information frequency modulation potential using the K nearest neighbor algorithm based on the user data transmitted by the user data collection module and classifies the user, and transmits the user classification result back to the user data collection module; The tag demand evaluation module is used to receive recommended information data, screen out key tags according to the recommended information data and calculate the corresponding tag demand coefficient, detect the user's clearing behavior and perform secondary screening on the key tags to obtain evaluation tags, use the modified geometric mean method to determine whether to adjust the demand of the evaluation tags downward, mark the evaluation tags that are to be adjusted downward, and send the tag demand coefficient of the marked tag to the information frequency modulation processing module; The modified geometric mean method determines whether to adjust the evaluation label downward. The specific steps are as follows: Data conversion: convert the label requirement coefficient and label clearing ratio of the evaluation label into the same dimension; Repeated conversion: Set n conversion parameters to obtain conversion results of multiple label requirement coefficients or label clearing ratios; Data connection: multiple connection values are obtained by subtracting the label requirement coefficient and label clearing ratio after data conversion using the same conversion parameters; Connection processing: perform logarithmic operations on multiple connection values and take the arithmetic mean as the connection processing amount; Antilogarithmic processing: Take the antilogarithm of the connection processing amount to obtain the demand evaluation value; When the demand evaluation value of an evaluation tag exceeds the preset evaluation threshold, the demand for the evaluation tag will not be adjusted downward; when the demand evaluation value of an evaluation tag is lower than the preset evaluation threshold, the demand for the evaluation tag will be adjusted downward; After receiving the marking tag, the information frequency modulation processing module obtains the tag attribute of the marking tag, calculates the down-adjustment coefficient of the corresponding marking tag based on the tag attribute of the marking tag and the tag demand coefficient, and sets the skip time for the marking tag according to the down-adjustment coefficient; The skip time is the time period set in the marketing platform during which the tag is not recommended; User clearing behavior refers to the user's active deletion of keywords in the search history within the detection time.

2. The precision marketing recommendation system based on AI big model according to claim 1 is characterized by: The user data in the user data collection module includes the number of tag categories occupied by the user's search keywords and the browsing time of tags in different categories; When a user enters a search keyword through the search bar, it will be recorded in the search history. In the search history, the search keyword entered by the user is identified and the tag category to which it belongs is determined. The total number of different tag categories is counted to obtain the number of tag categories occupied by the user's search keyword; When a user browses in the marketing platform, the tag where the user's browsing information is located is identified and the browsing time is recorded to obtain the browsing time of different categories of tags.

3. The precise marketing recommendation system based on the AI large model according to claim 2, characterized in that: The frequency modulation potential analysis module comprehensively analyzes the number of search keywords of users accounting for the label categories and the browsing time of different categories, and uses the K-nearest neighbor algorithm to analyze the information frequency modulation potential of users and classify users. The specific steps are as follows: Collect the number of search keywords of multiple users accounting for the label categories and the browsing time of different categories, and perform standardization processing respectively. Combine the number of search keywords accounting for the label categories into a search data set, and combine the browsing time of different categories into a browsing data set. Take the ratio of the data in the search data set or the browsing data set to the maximum value in the search data set or the browsing data set as the standardization result of the corresponding data. Randomly select a data in the search data set and the browsing data set as the reference data, set the nearest neighbor coefficient K, and select the K data with the smallest absolute value of the difference from the reference data in each data set as the marked data. Calculate the average value of the marked data in the search data set and the browsing data set and sum them to obtain the user classification threshold; Sum the standardized results of the number of search keyword categories of users and the browsing time of different categories to obtain the information frequency modulation potential of users, and compare the information frequency modulation potential with the user classification threshold; When the information frequency modulation potential of the user exceeds the user classification threshold, it is determined that the user is a high information frequency modulation potential user; when the information frequency modulation potential of the user is lower than the user classification threshold, it is determined that the user is a low information frequency modulation potential user.

4. The precise marketing recommendation system based on the AI large model according to claim 3, characterized in that: After classifying the users, the frequency modulation potential analysis module transmits the user classification result to the user data collection module. When the user is a high information frequency modulation potential user, the user data collection module sets the detection time to collect the recommended information data, and the recommended information data is the browsing frequency of the user under each label; The user data collection module selects a period of time as the detection time, records and counts the labels to which the recommended content clicked by the user belongs during the detection time, obtains the browsing frequency of the user under each label during the detection time, and the user data collection module transmits the browsing frequency of the user under each label to the label demand evaluation module.

5. The precise marketing recommendation system based on the AI large model according to claim 4, characterized in that: The label demand evaluation module receives the browsing frequency of the user under each label, selects the median of the browsing frequency under each label as the screening threshold, screens out the labels with the browsing frequency exceeding the screening threshold as the key labels, and takes the ratio of the browsing frequency of the key labels to the screening threshold as the label demand coefficient of the corresponding label; The user clearing behavior means that the user actively deletes the keywords in the search history during the detection time. Identify the labels where each keyword in the clearing list is located and match them with the key labels, and retain the successfully matched labels as the evaluation labels. Calculate the clearing ratio of the evaluation labels, and divide the sum of the number of keywords belonging to the same evaluation label in the clearing list by the total number of keywords in the clearing list to obtain the label clearing ratio of the corresponding evaluation label.

6. The precise marketing recommendation system based on the AI large model according to claim 1, characterized in that: The label attributes of the marked labels are the product shelf life and evaluation star rating under the corresponding labels; the evaluation star rating is the scoring system of the marketing platform, and different products are evaluated by setting different star ratings.

7. The precise marketing recommendation system based on the AI large model according to claim 6, characterized in that: When calculating the downward adjustment coefficient of the marked label, the information frequency modulation processing module randomly selects multiple products under the marked label as target products, collects the shelf life and evaluation star rating of each target product, calculates the average value of the shelf life of all target products as the label shelf life, and counts the total number of evaluation star ratings of all target products as the label star value.

8. The precise marketing recommendation system based on the AI large model according to claim 7, characterized in that: The geometric mean of the on-shelf duration and label star value of the marked label is standardized as the demand adjustment volume, and the downward adjustment coefficient of the corresponding marked label is calculated through the harmonic formula: , where B is the label demand coefficient of the marked label, s is the preset harmonic parameter, h is the demand adjustment volume of the marked label, and T is the downward adjustment coefficient of the marked label; The product of the default skip time of the marked label and the downward adjustment coefficient is used as the skip time of the corresponding marked label.

Images