An AI-driven customer segmentation and personalized recommendation system

Abstract

The present invention relates to the field of information recommendation technology, and in particular to an AI-driven customer segmentation and personalized recommendation system, comprising a user collection module, a data classification module, a user division module, a training module, a preference link generation module, a retrieval link generation module, a push output module, a monitoring and verification module, and an analysis module. Users are segmented into different groups and several user categories, and personalized link recommendations are provided for each user category. By collecting and analyzing a large amount of user data, accurate user segmentation and personalized recommendations are achieved, thereby improving data processing efficiency.

Description

Technical Field

[0001] The present invention relates to the field of information recommendation technology, and in particular to an AI-driven customer segmentation and personalized recommendation system. Background Art

[0002] With the rapid development of the Internet, people's daily lives are increasingly closely connected with the Internet, such as listening to music, watching movies, shopping, reading, chatting, etc. At the same time, a huge amount of user and item data is constantly generated on the Internet every day, which makes it difficult or even impossible for Internet users to quickly find the unknown information they need or are interested in. Therefore, personalized recommendation technology came into being and is constantly being innovated. Personalized recommendation technology aims to model the user's interest preferences based on the user's own characteristics, and then recommend items that meet the user's personalized preferences.

[0003] Chinese Patent Publication No.: CN104537114B discloses a personalized recommendation method, which relates to the field of computer data processing technology, including: obtaining user usage behavior data on items; generating a user usage behavior sublist according to the obtained usage behavior data; traversing the generated user usage behavior sublist to estimate the one-step transition probability matrix of the item; establishing a personalized recommendation model based on the user's forgetting process of interest in the item and the Markov model; using the gradient descent method to estimate the user's personalized parameters in the interest forgetting process, so as to make recommendations for the user according to the user usage behavior sublist; it can be seen that the above technical solution has the following problems: in the face of a large number of users, the personalized parameters of each user are determined one by one, resulting in low data processing efficiency. Summary of the invention

[0004] To this end, the present invention provides an AI-driven customer segmentation and personalized recommendation system to overcome the problem in the prior art of low data processing efficiency caused by determining personalized parameters of each user one by one when facing a large number of users.

[0005] To achieve the above objectives, the present invention provides an AI-driven customer segmentation and personalized recommendation system, comprising:

[0006] A user acquisition module, which is used to obtain search information of each user, the search information including search terms and each valid keyword determined based on the search terms;

[0007] A data classification module, which is connected to the user acquisition module and is used to determine the search preference of each user based on the keywords in each search cycle;

[0008] A user classification module, which is connected to the data classification module, is used to determine the user's development preference based on the user's various search preferences during the training cycle, and to classify each user into a user category based on the development preference;

[0009] A training module, which is connected to the user classification module, and is used to use the search terms of each user in a single user category as a training set, and input the training set into a training model to obtain a push preference for a single user category;

[0010] A preference link generation module, which is connected to the training module and is used to select a number of corresponding preference links in the database based on the push preference;

[0011] A search link generation module, which is connected to the user acquisition module and is used to select a number of corresponding search links in the database based on the search terms of a single user;

[0012] A push output module, which is connected to the preference link generation module, the search link generation module and the user classification module, and is used to classify push miscellaneous categories of a single user based on user categories, and output a push page based on the push miscellaneous categories;

[0013] A monitoring and verification module, which is connected to the push output module and is used to obtain the click rate of each user on the push page;

[0014] An analysis module, which is respectively connected to the data classification module, the user segmentation module, the push output module and the monitoring and verification module, and is used to determine whether the operating parameters of the data classification module are qualified based on the click rate and the click difference parameters, and when it is determined that the operating parameters of the data classification module are abnormal, determine the processing method for the data classification module, wherein the processing method includes adjusting the duration of the training cycle to a corresponding value, or adjusting the duration of the retrieval cycle to a corresponding value.

[0015] Furthermore, the data classification module is used to determine the search preference of a single user in a corresponding search period, obtain each valid keyword of the single user in a single search period, obtain the number of searches for each valid keyword, and determine the keyword with the largest number of searches as the search preference of the single user in the corresponding search period;

[0016] The user classification module is used to classify users into categories based on their search preferences during a training period, including:

[0017] Obtaining the retrieval preferences of a single user in each retrieval cycle within a training cycle, and determining the retrieval preferences arranged in sequence as the development preferences of the single user;

[0018] Users with the same development preference are classified into the same user category.

[0019] Furthermore, the push output module is used to classify push notifications of a single user into different categories based on user categories, including:

[0020] Get the search preferences corresponding to a single user;

[0021] Solve the overlap of search preferences, calculate the ratio of the number of identical search preferences in the user category corresponding to a single user and the search preferences of other user categories to the total number of search preferences of the two user categories, and determine the ratio as the overlap of search preferences corresponding to the two user categories;

[0022] Determine two user categories whose search preference overlap is greater than a preset overlap as similar categories;

[0023] If the number of similar categories corresponding to a single user is zero, the push doping category of the single user is determined to be a weak push doping category;

[0024] If the number of similar categories corresponding to a single user is greater than zero, the push doping category of the single user is determined to be a strong push doping category;

[0025] The push output module is used to output a push page based on the push doping category, including:

[0026] When the push doping category of a single user is a weak push doping category, a link to a push page is selected according to a preset doping ratio;

[0027] When the push doping category of a single user is a strong push doping category, a push page is output according to a preset doping ratio, and a selection method of a preferred link is determined based on the alternative push ratio;

[0028] The preset doping ratio is the ratio of the number of preference links generated by the preference link generation module to the number of retrieval links generated by the retrieval link generation module in a single push page;

[0029] The method for selecting the preferred link based on the alternative push ratio is:

[0030] Based on the ratio of alternative pushes, select the corresponding preferred links in each similar category and add them to the push page;

[0031] The alternative push ratio is the ratio of the preferred links selected from each similar category.

[0032] Furthermore, the analysis module is used to periodically determine whether the operating parameters of the data classification module are qualified based on the average value of each click rate corresponding to each strong push doping category, including:

[0033] If the average value is less than or equal to the first preset average value, it is determined that the operating parameters of the data classification module are abnormal, and a processing method for the data classification module is determined based on the user growth rate;

[0034] If the average value is less than or equal to the second preset average value and greater than the first preset average value, determining whether the operating parameters of the data classification module determined based on the click difference parameter are qualified;

[0035] If the average value is greater than the second preset average value, it is determined that the operating parameters of the data classification module are qualified, and the data classification module is controlled to continue to operate using the current operating parameters.

[0036] Furthermore, the analysis module is used to determine whether the operation parameters of the data classification module are qualified based on the click difference parameter, including:

[0037] To determine the click difference parameter, obtain the variance of each click rate corresponding to each strong push mixed category, and obtain the click difference parameter;

[0038] If the click difference parameter is less than or equal to the preset click difference parameter, it is determined that the operating parameters of the data classification module are abnormal, and a processing method for the data classification module is determined based on the user growth rate;

[0039] If the click difference parameter is greater than the preset click difference parameter, the operating parameters of the data classification module are determined to be qualified, and the preset doping ratio of the single user is adjusted to a corresponding value based on the number of similar categories of the single user category.

[0040] Further, the analysis module is used to adjust the preset doping ratio of a single user to a corresponding value based on the number of each similar category of the single user category, wherein:

[0041] The increase range of the preset doping ratio of a single user determined based on the number of each similar category of the single user category is proportional to the number of each similar category of the single user category.

[0042] Furthermore, the analysis module is used to determine the user growth rate, obtain the ratio of the number of new users in the current training cycle to the average number of new users in each historical training cycle, and obtain the user growth rate.

[0043] Furthermore, the analysis module is used to determine a processing method for the data classification module based on the user growth rate, including:

[0044] If the user growth rate is less than or equal to the preset growth rate, adjusting the duration of the training cycle to a corresponding value based on the user growth rate;

[0045] If the user growth rate is greater than the preset growth rate, the duration of the retrieval cycle is adjusted to a corresponding value based on the click difference parameter.

[0046] Further, the analysis module is used to adjust the duration of the training cycle to a corresponding value based on the user growth rate, wherein:

[0047] The increase in the duration of the training cycle determined based on the user growth rate is negatively correlated with the user growth rate.

[0048] Furthermore, the analysis module is used to adjust the duration of the retrieval cycle to a corresponding value based on the click difference parameter, wherein:

[0049] The reduction in the duration of the retrieval cycle determined based on the click difference parameter is inversely proportional to the click difference parameter.

[0050] Compared with the prior art, the beneficial effect of the present invention is that it divides users into different groups and several user categories, and provides personalized link recommendations for each user category. By collecting and analyzing a large amount of user data, accurate user segmentation and personalized recommendations are achieved, thereby improving data processing efficiency.

[0051] Furthermore, the development preference of a single user is determined based on the search preferences arranged in sequence, and each user is divided into several user categories based on the development preference. The push preference of a single user category is determined based on the search data of each user in the single user category, thereby achieving accurate user segmentation and personalized recommendation, effectively reducing the amount of data processing and improving the accuracy of push.

[0052] Furthermore, the number of similar categories corresponding to a single user is determined. The number of similar categories represents the situation of user categories that are similar to the search preferences of a single user. When the push doping category of a single user is a strong push doping category, there are several similar categories. The preference links corresponding to the similar categories are integrated to output the push page of a single user. While improving the richness of the push page information, the links in the push page are made more in line with the user's preferences. However, in the case of a strong push doping category, the user's preferences are more complicated. At this time, it is necessary to further monitor the user's clicks and take corresponding processing methods in time when it is determined that the operating parameters of the data classification module are abnormal. The situation of the strong push doping category is monitored in a targeted manner, which effectively reduces the data processing volume and further improves the data processing efficiency.

[0053] Furthermore, the click rate of each strongly pushed and mixed category is monitored periodically to comprehensively determine whether the operating parameters of the data classification module are qualified. When the average value is less than or equal to the second preset average value and greater than the first preset average value, the click rate is low. At this time, the click difference parameter is determined to comprehensively determine whether the operating parameters of the data classification module are qualified. When the click difference parameter is less than or equal to the preset click difference parameter, the click rate of each strongly pushed and mixed category is generally low. In this case, the operating parameters of the data classification module are determined to be abnormal; when the click difference parameter is greater than the preset click difference parameter, the click rate deviation of each user is large. When the average value of each click rate is low, only some users' recommendation pages appear inaccurate abnormal situations. At this time, based on the number of similar categories corresponding to each user, the preset mixed ratio of each user is adjusted to the corresponding value, and the proportion of push link types is dynamically adjusted to make the pushed links more in line with user preferences, further improving data processing efficiency.

[0054] Furthermore, based on determining that the operating parameters of the data classification module are abnormal, the data classification module adjusts the processing method of the data collected by the user collection module. When the user growth rate is greater than the preset growth rate, the retrieval information of each user is more complex due to the rapid growth of users in a short period of time. At this time, the retrieval cycle is shortened to make the user category division more refined; when the user growth rate is less than or equal to the preset growth rate, each module runs smoothly, but there is a deviation in the push of preferred links, resulting in a too low click-through rate of users. At this time, the training cycle is adjusted to increase the data volume of the training set to improve the accuracy of the push preference, thereby further improving the data processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a module block diagram of the AI-driven customer segmentation and personalized recommendation system according to an embodiment of the present invention;

[0056] Figure 2 A logic determination diagram of whether the operating parameters of the data classification module are qualified based on the average value of each click rate corresponding to each strong push doping category by the analysis module in the embodiment of the present invention;

[0057] Figure 3 A logical decision diagram for determining whether the operating parameters of the data classification module are qualified based on the click difference parameter by the analysis module in an embodiment of the present invention;

[0058] Figure 4 This is a logical decision diagram for an analysis module according to an embodiment of the present invention to determine a processing method for a data classification module based on a user growth rate. DETAILED DESCRIPTION

[0059] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0060] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0061] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0062] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0063] See also Figure 1 , Figure 2 , Figure 3 as well as Figure 4 As shown, they are respectively a module block diagram of an AI-driven customer segmentation and personalized recommendation system according to an embodiment of the present invention, a logic determination diagram for determining whether the operating parameters of a data classification module are qualified based on the average value of each click rate corresponding to each strongly pushed doped category by the analysis module, a logic determination diagram for determining whether the operating parameters of a data classification module are qualified based on a click difference parameter by the analysis module, and a logic determination diagram for determining a processing method for a data classification module based on a user growth rate by the analysis module; an AI-driven customer segmentation and personalized recommendation system according to an embodiment of the present invention comprises:

[0064] A user acquisition module, which is used to obtain search information of each user, the search information including search terms and each valid keyword determined based on the search terms;

[0065] A data classification module, which is connected to the user acquisition module and is used to determine the search preference of each user based on the keywords in each search cycle;

[0066] A user classification module, which is connected to the data classification module, is used to determine the user's development preference based on the user's various search preferences during the training cycle, and to classify each user into a user category based on the development preference;

[0067] A training module, which is connected to the user classification module, and is used to use the search terms of each user in a single user category as a training set, and input the training set into a training model to obtain a push preference for a single user category;

[0068] A preference link generation module, which is connected to the training module and is used to select a number of corresponding preference links in the database based on the push preference;

[0069] A search link generation module, which is connected to the user acquisition module and is used to select a number of corresponding search links in the database based on the search terms of a single user;

[0070] A push output module, which is connected to the preference link generation module, the search link generation module and the user classification module, and is used to classify push miscellaneous categories of a single user based on user categories, and output a push page based on the push miscellaneous categories;

[0071] A monitoring and verification module, which is connected to the push output module and is used to obtain the click rate of each user on the push page;

[0072] An analysis module, which is respectively connected to the data classification module, the user segmentation module, the push output module and the monitoring and verification module, and is used to determine whether the operating parameters of the data classification module are qualified based on the click rate and the click difference parameters, and when it is determined that the operating parameters of the data classification module are abnormal, determine the processing method for the data classification module, wherein the processing method includes adjusting the duration of the training cycle to a corresponding value, or adjusting the duration of the retrieval cycle to a corresponding value.

[0073] Specifically, there is no limitation on the specific method of determining effective keywords in the search terms. Natural language processing can be used to segment the search terms into several words or phrases, determine the part of speech (noun, verb, etc.) of each word or phrase, determine nouns and adjectives as effective keywords, identify proper nouns in the search terms, such as names of people, places, and organizations, and determine the identified proper nouns as effective keywords. This is existing technology and will not be elaborated on.

[0074] Specifically, the click rate of each user on the pushed page is obtained by obtaining the ratio of the number of links clicked by a single user to the number of links browsed for each link pushed once, and obtaining the click rate.

[0075] Specifically, the training module obtains the search terms of each user in a single user category, and cleans and preprocesses them, including removing special characters and converting them into a unified format; the preprocessed search terms are combined into a training set, and the terms are vectorized using a bag-of-words model, TF-IDF or Word2Vec, and the model is trained using the constructed training set, and the trained model is applied to new search term data. The result of the model output is the push preference for the single user category. The trained model is the training model, which will not be repeated here.

[0076] Specifically, users are segmented into different groups and divided into several user categories, and personalized link recommendations are provided for each user category. By collecting and analyzing a large amount of user data, accurate user segmentation and personalized recommendations are achieved, thereby improving data processing efficiency.

[0077] Specifically, the data classification module is used to determine the search preference of a single user in a corresponding search cycle, obtain each valid keyword of the single user in a single search cycle, obtain the search times of each valid keyword, and determine the keyword with the largest search times as the search preference of the single user in the corresponding search cycle;

[0078] The user classification module is used to classify users into categories based on their search preferences during a training period, including:

[0079] Obtaining the retrieval preferences of a single user in each retrieval cycle within a training cycle, and determining the retrieval preferences arranged in sequence as the development preferences of the single user;

[0080] Users with the same development preference are classified into the same user category.

[0081] Specifically, the development preference of a single user is determined based on the search preferences arranged in sequence, and each user is divided into several user categories based on the development preference. The push preference of a single user category is determined based on the search data of each user in the single user category, thereby achieving accurate user segmentation and personalized recommendation, effectively reducing the amount of data processing and improving the accuracy of push.

[0082] Specifically, the push output module is used to classify push notification categories of a single user based on user categories, including:

[0083] Get the search preferences corresponding to a single user;

[0084] Solve the overlap of search preferences, calculate the ratio of the number of identical search preferences in the user category corresponding to a single user and the search preferences of other user categories to the total number of search preferences of the two user categories, and determine the ratio as the overlap of search preferences corresponding to the two user categories;

[0085] Determine two user categories whose search preference overlap is greater than a preset overlap as similar categories;

[0086] If the number of similar categories corresponding to a single user is zero, the push doping category of the single user is determined to be a weak push doping category;

[0087] If the number of similar categories corresponding to a single user is greater than zero, the push doping category of the single user is determined to be a strong push doping category;

[0088] The push output module is used to output a push page based on the push doping category, including:

[0089] When the push doping category of a single user is a weak push doping category, a link to a push page is selected according to a preset doping ratio;

[0090] When the push doping category of a single user is a strong push doping category, a push page is output according to a preset doping ratio, and a selection method of a preferred link is determined based on the alternative push ratio;

[0091] The preset doping ratio is the ratio of the number of preference links generated by the preference link generation module to the number of retrieval links generated by the retrieval link generation module in a single push page;

[0092] The method for selecting the preferred link based on the alternative push ratio is:

[0093] Based on the ratio of alternative pushes, select the corresponding preferred links in each similar category and add them to the push page;

[0094] The alternative push ratio is the ratio of the preferred links selected from each similar category.

[0095] Specifically, the number of similar categories corresponding to a single user is determined. The number of similar categories represents the situation of user categories that are similar to the search preferences of a single user. When the push doping category of a single user is a strong push doping category, there are several similar categories. The preference links corresponding to the similar categories are combined to output the push page of a single user. While improving the richness of the push page information, the links in the push page are made more in line with the user's preferences. However, in the case of strong push doping categories, the user's preferences are more complicated. At this time, it is necessary to further monitor the user's clicks and take corresponding processing methods in time when it is determined that the operating parameters of the data classification module are abnormal. The situation of the strong push doping category is monitored in a targeted manner, which effectively reduces the data processing volume and further improves the data processing efficiency.

[0096] Specifically, there is no limitation on the specific method for determining the proportional relationship of the preferred links selected in each similar category. It can be a geometric relationship or a corresponding proportion relationship can be determined based on the overlap of each search preference, which will not be elaborated here.

[0097] Specifically, the analysis module is used to periodically determine whether the operating parameters of the data classification module are qualified based on the average value of each click rate corresponding to each strong push doping category, including:

[0098] If the average value is less than or equal to the first preset average value, it is determined that the operating parameters of the data classification module are abnormal, and a processing method for the data classification module is determined based on the user growth rate;

[0099] If the average value is less than or equal to the second preset average value and greater than the first preset average value, determining whether the operating parameters of the data classification module determined based on the click difference parameter are qualified;

[0100] If the average value is greater than the second preset average value, it is determined that the operating parameters of the data classification module are qualified, and the data classification module is controlled to continue to operate using the current operating parameters.

[0101] Specifically, the first preset average value is selected within the interval [0.12, 0.23], and the second preset average value is selected within the interval [0.47, 0.52].

[0102] Specifically, the analysis module is used to determine whether the operating parameters of the data classification module are qualified based on the click difference parameter, including:

[0103] To determine the click difference parameter, obtain the variance of each click rate corresponding to each strong push mixed category, and obtain the click difference parameter;

[0104] If the click difference parameter is less than or equal to the preset click difference parameter, it is determined that the operating parameters of the data classification module are abnormal, and a processing method for the data classification module is determined based on the user growth rate;

[0105] If the click difference parameter is greater than the preset click difference parameter, the operating parameters of the data classification module are determined to be qualified, and the preset doping ratio of the single user is adjusted to a corresponding value based on the number of similar categories of the single user category.

[0106] Specifically, the preset click difference parameter L0 is in the interval [0.16P 2 , 0.25P 2 ], P is the average of the click rates corresponding to each strongly pushed mixed category.

[0107] Specifically, the click rate of each strongly pushed and mixed category is monitored periodically to comprehensively determine whether the operating parameters of the data classification module are qualified. When the average value is less than or equal to the second preset average value and greater than the first preset average value, the click rate is low. At this time, the click difference parameter is determined to comprehensively determine whether the operating parameters of the data classification module are qualified. When the click difference parameter is less than or equal to the preset click difference parameter, the click rate of each strongly pushed and mixed category is generally low. In this case, the operating parameters of the data classification module are determined to be abnormal; when the click difference parameter is greater than the preset click difference parameter, the click rate deviation of each user is large. When the average value of each click rate is low, only some users' recommendation pages appear inaccurate abnormal situations. At this time, based on the number of similar categories corresponding to each user, the preset mixed ratio of each user is adjusted to the corresponding value, and the proportion of push link types is dynamically adjusted to make the pushed links more in line with user preferences, further improving data processing efficiency.

[0108] Specifically, the analysis module is used to adjust the preset doping ratio of a single user to a corresponding value based on the number of each similar category of the single user category, wherein:

[0109] The increase range of the preset doping ratio of a single user determined based on the number of each similar category of the single user category is proportional to the number of each similar category of the single user category.

[0110] In this embodiment, optionally,

[0111] Recording the number of each similar category of a single user category as a similarity number, and comparing the similarity number with a first preset similarity number and a second preset similarity number;

[0112] If the similarity number is less than or equal to the first preset similarity number, adjusting the preset doping ratio to 1.11 times the initial preset doping ratio;

[0113] If the similarity number is less than or equal to the second preset similarity number and greater than the first preset similarity number, adjusting the preset doping ratio to 1.21 times the initial preset doping ratio;

[0114] If the similarity number is greater than the second preset similarity number, adjusting the preset doping ratio to 1.31 times the initial preset doping ratio;

[0115] Specifically, the first preset similarity number is 1.3N0, the second preset similarity number is 2.5N0, and N0 is the average value of the similarity numbers of each user category.

[0116] Specifically, the analysis module is used to determine the user growth rate, obtain the ratio of the number of new users in the current training cycle to the average number of new users in each historical training cycle, and obtain the user growth rate.

[0117] Specifically, the analysis module is used to determine a processing method for the data classification module based on the user growth rate, including:

[0118] If the user growth rate is less than or equal to the preset growth rate, adjusting the duration of the training cycle to a corresponding value based on the user growth rate;

[0119] If the user growth rate is greater than the preset growth rate, the duration of the retrieval cycle is adjusted to a corresponding value based on the click difference parameter.

[0120] Specifically, the preset growth rate Z0 is selected within the interval [1.13, 1.32].

[0121] Specifically, based on determining that the operating parameters of the data classification module are abnormal, the data classification module adjusts the way it processes the data collected by the user collection module. When the user growth rate is greater than the preset growth rate, the retrieval information of each user is more complex due to the rapid growth of users in a short period of time. At this time, the retrieval cycle is shortened to make the user category division more refined; when the user growth rate is less than or equal to the preset growth rate, each module runs smoothly, but there is a deviation in the push of preferred links, resulting in a too low click-through rate for users. At this time, the training cycle is adjusted to increase the data volume of the training set to improve the accuracy of the push preference, thereby further improving the data processing efficiency.

[0122] Specifically, the analysis module is used to adjust the duration of the training cycle to a corresponding value based on the user growth rate, wherein:

[0123] The increase in the duration of the training cycle determined based on the user growth rate is negatively correlated with the user growth rate.

[0124] In this embodiment, optionally,

[0125] Comparing the user growth rate with a first preset growth comparison threshold and a second preset growth comparison threshold;

[0126] If the user growth rate is less than or equal to the first preset growth comparison threshold, the duration of the training cycle is adjusted to 1.28 times the duration of the initial training cycle;

[0127] If the user growth rate is less than or equal to the second preset growth comparison threshold and greater than the first preset growth comparison threshold, the duration of the training cycle is adjusted to 1.21 times the duration of the initial training cycle;

[0128] If the user growth rate is greater than the second preset growth comparison threshold, the duration of the training cycle is adjusted to 1.11 times the duration of the initial training cycle;

[0129] The first preset growth comparison threshold is 0.72Z0, and the second preset growth comparison threshold is 0.85Z0.

[0130] Specifically, the analysis module is used to adjust the duration of the retrieval cycle to a corresponding value based on the click difference parameter, wherein:

[0131] The reduction in the duration of the retrieval cycle determined based on the click difference parameter is inversely proportional to the click difference parameter.

[0132] In this embodiment, optionally,

[0133] Comparing the click difference parameter with a first preset difference comparison threshold and a second preset difference comparison threshold;

[0134] If the click difference parameter is less than or equal to the first preset difference comparison threshold, the duration of the search cycle is adjusted to 0.72 times the duration of the initial search cycle;

[0135] If the click difference parameter is less than or equal to the second preset difference comparison threshold and greater than the first preset difference comparison threshold, the duration of the search cycle is adjusted to 0.83 times the duration of the initial search cycle;

[0136] If the click difference parameter is greater than the second preset difference comparison threshold, the duration of the search cycle is adjusted to 0.91 times the duration of the initial search cycle;

[0137] The first preset difference comparison threshold is 0.66L0, and the second preset difference comparison threshold is 0.83L0.

[0138] Under the condition that the adjustment of the retrieval cycle is completed, the preset coincidence is adjusted to 1.12 times the initial preset coincidence.

[0139] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, 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 AI-driven customer segmentation and personalized recommendation system, characterized in that: include: A user acquisition module, which is used to obtain search information of each user, the search information including search terms and each valid keyword determined based on the search terms; A data classification module, which is connected to the user acquisition module and is used to determine the search preference of each user based on the keywords in each search cycle; A user classification module, which is connected to the data classification module, is used to determine the user's development preference based on the user's various search preferences during the training cycle, and to classify each user into a user category based on the development preference; A training module, which is connected to the user classification module, and is used to use the search terms of each user in a single user category as a training set, and input the training set into a training model to obtain a push preference for a single user category; A preference link generation module, which is connected to the training module and is used to select a number of corresponding preference links in the database based on the push preference; A search link generation module, which is connected to the user acquisition module and is used to select a number of corresponding search links in the database based on the search terms of a single user; A push output module, which is connected to the preference link generation module, the search link generation module and the user classification module, and is used to classify push miscellaneous categories of a single user based on user categories, and output a push page based on the push miscellaneous categories; A monitoring and verification module, which is connected to the push output module and is used to obtain the click rate of each user on the push page; an analysis module, which is respectively connected to the data classification module, the user classification module, the push output module and the monitoring and verification module, and is used to determine whether the operating parameters of the data classification module are qualified based on the click rate and the click difference parameters, and in the case where it is determined that the operating parameters of the data classification module are abnormal, determine a processing method for the data classification module, wherein the processing method includes adjusting the duration of the training cycle to a corresponding value, or adjusting the duration of the retrieval cycle to a corresponding value; The data classification module is used to determine the search preference of a single user in a corresponding search cycle, obtain each valid keyword of the single user in a single search cycle, obtain the search times of each valid keyword, and determine the keyword with the largest search times as the search preference of the single user in the corresponding search cycle; The user classification module is used to classify users into categories based on their search preferences during a training period, including: Obtaining the retrieval preferences of a single user in each retrieval cycle within a training cycle, and determining the retrieval preferences arranged in sequence as the development preferences of the single user; Users with the same development preference are classified into the same user category.

2. The AI-driven customer segmentation and personalized recommendation system according to claim 1, characterized in that: The push output module is used to classify push notification categories of a single user based on user categories, including: Get the search preferences corresponding to a single user; Solve the overlap of search preferences, calculate the ratio of the number of identical search preferences in the user category corresponding to a single user and the search preferences of other user categories to the total number of search preferences of the two user categories, and determine the ratio as the overlap of search preferences corresponding to the two user categories; Determine two user categories whose search preference overlap is greater than a preset overlap as similar categories; If the number of similar categories corresponding to a single user is zero, the push doping category of the single user is determined to be a weak push doping category; If the number of similar categories corresponding to a single user is greater than zero, the push doping category of the single user is determined to be a strong push doping category; The push output module is used to output a push page based on the push doping category, including: When the push doping category of a single user is a weak push doping category, a link to a push page is selected according to a preset doping ratio; When the push doping category of a single user is a strong push doping category, a push page is output according to a preset doping ratio, and a selection method of a preferred link is determined based on the alternative push ratio; The preset doping ratio is the ratio of the number of preference links generated by the preference link generation module to the number of retrieval links generated by the retrieval link generation module in a single push page; The method for selecting the preferred link based on the alternative push ratio is: Based on the ratio of alternative pushes, select the corresponding preferred links in each similar category and add them to the push page; The alternative push ratio is the ratio of the preferred links selected from each similar category.

3. The AI-driven customer segmentation and personalized recommendation system according to claim 2, characterized in that: The analysis module is used to periodically determine whether the operating parameters of the data classification module are qualified based on the average value of each click rate corresponding to each strong push doping category, including: If the average value is less than or equal to the first preset average value, it is determined that the operating parameters of the data classification module are abnormal, and a processing method for the data classification module is determined based on the user growth rate; If the average value is less than or equal to the second preset average value and greater than the first preset average value, determining whether the operating parameters of the data classification module are qualified based on the click difference parameter; If the average value is greater than the second preset average value, it is determined that the operating parameters of the data classification module are qualified, and the data classification module is controlled to continue to operate using the current operating parameters.

4. The AI-driven customer segmentation and personalized recommendation system according to claim 3, characterized in that: The analysis module is used to determine whether the operation parameters of the data classification module are qualified based on the click difference parameter, including: To determine the click difference parameter, obtain the variance of each click rate corresponding to each strong push mixed category, and obtain the click difference parameter; If the click difference parameter is less than or equal to the preset click difference parameter, it is determined that the operating parameters of the data classification module are abnormal, and a processing method for the data classification module is determined based on the user growth rate; If the click difference parameter is greater than the preset click difference parameter, the operating parameters of the data classification module are determined to be qualified, and the preset doping ratio of the single user is adjusted to a corresponding value based on the number of similar categories of the single user category.

5. The AI-driven customer segmentation and personalized recommendation system according to claim 4, characterized in that: The analysis module is used to adjust the preset doping ratio of a single user to a corresponding value based on the number of each similar category of the single user category, wherein: The increase range of the preset doping ratio of a single user determined based on the number of each similar category of the single user category is proportional to the number of each similar category of the single user category.

6. The AI-driven customer segmentation and personalized recommendation system according to claim 5, characterized in that: The analysis module is used to determine the user growth rate, obtain the ratio of the number of new users in the current training cycle to the average number of new users in each historical training cycle, and obtain the user growth rate.

7. The AI-driven customer segmentation and personalized recommendation system according to claim 6, characterized in that: The analysis module is used to determine a processing method for the data classification module based on the user growth rate, including: If the user growth rate is less than or equal to the preset growth rate, adjusting the duration of the training cycle to a corresponding value based on the user growth rate; If the user growth rate is greater than the preset growth rate, the duration of the retrieval cycle is adjusted to a corresponding value based on the click difference parameter.

8. The AI-driven customer segmentation and personalized recommendation system according to claim 7, characterized in that: The analysis module is used to adjust the duration of the training cycle to a corresponding value based on the user growth rate, wherein: The increase in the duration of the training cycle determined based on the user growth rate is negatively correlated with the user growth rate.

9. The AI-driven customer segmentation and personalized recommendation system according to claim 8, characterized in that: The analysis module is used to adjust the duration of the retrieval cycle to a corresponding value based on the click difference parameter, wherein: The reduction in the duration of the retrieval cycle determined based on the click difference parameter is inversely proportional to the click difference parameter.

Images