An AI search recommendation method and system for procurement

Abstract

The present invention belongs to the field of intelligent recommendation technology, and specifically relates to an AI search recommendation method and system for procurement. The method includes: determining the degree of association between the two major categories in all the historical orders of a relevant user based on the regularity of the time intervals between historical orders containing goods belonging to two major categories in all the historical orders of the relevant user, and then determining whether the two major categories belong to the associated category. For the two major categories that do not belong to the associated category, further determining the stage of the two major categories in the historical orders based on the regularity of the time intervals between historical orders containing goods belonging to the two major categories in all the historical orders of all the relevant users, and then determining whether the two major categories belong to the advanced category; and recommending products to the user based on the associated category and advanced category of the major category to which the products in the user's reference order belong. The present invention improves the accuracy of recommendations and the breadth of user demand coverage.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recommendation technology. More specifically, the present invention relates to an AI search recommendation method and system for procurement. Background Art

[0002] The recommendation system uses data analysis technology to find out the products that users are most likely to be interested in and recommend them to them.

[0003] Item-based collaborative filtering (Item-CF) is a classic method in the field of recommendation systems. Its core idea is to quantify the similarity between items by calculating users' historical ratings or behavioral data (such as clicks and purchases) on items, and recommend products that users may be interested in based on similar items.

[0004] This method relies on the density of the user-item interaction matrix and usually uses cosine similarity or Pearson correlation coefficient to measure item correlation.

[0005] However, in actual applications, item-based collaborative filtering algorithms only judge similarity through co-purchase or ratings, resulting in recommendation results being limited to products with similar functions, making it difficult to explore cross-category associations and advanced needs. In addition, item-based collaborative filtering algorithms only focus on static co-occurrence situations and lack temporal dynamic analysis, resulting in mismatched recommendation timing. Therefore, item-based collaborative filtering algorithms have the problem of difficulty in exploring cross-category association needs and mismatched recommendation timing, which seriously affects the accuracy of recommendations. Summary of the Invention

[0006] To address the technical issues that the above-mentioned item-based collaborative filtering algorithm has difficulty in mining cross-category related demands and mismatches recommendation timing, which seriously affects recommendation accuracy, the present invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides an AI search and recommendation method for procurement, comprising: obtaining the category to which each commodity belongs through a procurement item classification catalog; taking any category as a target category, and recording any category other than the target category as a reference category; for all historical orders of a relevant user: recording historical orders containing commodities belonging to the target category as target orders, and determining whether the target order has associated orders in all historical orders containing commodities belonging to the reference category based on the interval time period between the historical orders; and determining whether the target order has associated orders based on the number of target orders with associated orders and the size and regularity of the interval time period between the target order with associated orders and its associated orders. Determine the degree of correlation between the target category and the reference category in the historical orders of the relevant user; determine whether the target category and the reference category belong to the associated category based on the degree of correlation between the target category and the reference category in the historical orders of all relevant users; if the target category and the reference category do not belong to the associated category: determine the stage nature of the target category and the reference category in the historical orders based on the number of target orders with associated orders of all relevant users and the regularity of the interval time period between the target orders with associated orders and their associated orders, and then determine whether the target category and the reference category belong to the advanced category; recommend products to the user based on the associated category and the advanced category of the category to which the products in the user's reference orders belong.

[0008] Preferably, the related users are users who have historical orders containing commodities belonging to the target category and historical orders containing commodities belonging to the reference category.

[0009] Preferably, the step of determining whether the target order has an associated order in all historical orders containing goods belonging to a reference category based on the interval time between historical orders includes: recording the target category as the target category , record the reference category as the reference category ; For the Target orders , among all historical orders containing products belonging to the reference category: If all historical orders containing products belonging to the reference category and the target order There are other target orders in the interval time period, then the target order There is no associated order; otherwise, it will be associated with the target order There are multiple historical orders with no other target orders in the interval time period. The historical order with the shortest interval time is recorded as the target order Related orders.

[0010] Preferably, determining the degree of correlation between the target category and the reference category in the historical orders of the relevant user includes: Where, is the degree of correlation between the target category and the reference category in the historical orders of the relevant user, is the average of the intervals between all target orders and their associated orders. is the number of target orders with associated orders, is the quantity of all target orders, is the number of all historical orders containing products belonging to the reference category, is the standard deviation of the interval time between all target orders and their associated orders. Characterizes the regularity of the interval length between the target order and its associated orders. is the Sigmoid function.

[0011] Preferably, the method of determining whether the target category and the reference category belong to the associated class based on the degree of association between the target category and the reference category in the historical orders of all relevant users includes: arranging the degrees of association between the target category and the reference category in the historical orders of all relevant users in ascending order, and obtaining the first quartile after arrangement: if the first quartile is greater than a preset association threshold, the target category and the reference category belong to the associated class; if the first quartile is less than or equal to the preset association threshold, the target category and the reference category do not belong to the associated class.

[0012] Preferably, the step of determining the target category and the reference category in historical orders includes: Where, The stages of target categories and reference categories in historical orders, The duration of the interval between the target order and its associated orders. The number of related users in The duration range of the target category and reference category in historical orders. is the number of all relevant users, is the standard deviation of duration corresponding to the stage duration range, is the natural exponential function, is the Sigmoid function, Characterizes the regularity of the interval time between the target order and its associated orders for all relevant users.

[0013] Preferably, the phased duration range and the duration standard deviation corresponding to the phased duration range include: obtaining the interval time period between the target order and its associated orders of all relevant users, arranging the duration of all interval time periods in ascending order, and calculating the duration in The standard deviation of the duration of all intervals in the range , Indicates the sorted percentile, Indicates the sorted percentile; for Integer in the range, get The smallest value Corresponding range , as the phased duration range of the target category and reference category in historical orders, and The standard deviation of duration corresponding to the stage duration range.

[0014] Preferably, the determining whether the target category and the reference category belong to the advanced category includes: if the stage of the target category and the reference category in the historical orders is greater than a preset stage threshold, then the target category and the reference category belong to the advanced category.

[0015] Preferably, the method for obtaining the user's reference order includes: recording the maximum value of the upper limit of the phased duration range of all two categories with advanced relationships in historical orders as the user's advanced reference duration ; Get all the historical orders of the user, the order time is The historical orders within the range are recorded as the user's first reference order. The current moment; the average of the intervals between the target order and its associated orders in all the user's historical orders is used as the user's associated reference duration ; Get all the historical orders of the user, the order time is The historical orders within the range are recorded as the user's second reference order.

[0016] In a second aspect, the present invention provides an AI search and recommendation system for procurement, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned AI search and recommendation method for procurement is implemented.

[0017] By adopting the above technical solution, the above-mentioned AI search recommendation method for procurement is generated into a computer program and stored in a memory to be loaded and executed by a processor, thereby making a terminal device based on the memory and the processor for easy use.

[0018] The beneficial effects of the present invention are:

[0019] The present invention, through the collaboration of large-category-level association mining, temporal dynamics modeling and group statistical decision-making, not only solves the problem of recommendation homogeneity of traditional methods by distinguishing between association and advanced relationships, short-term and long-term needs, but also makes up for the defect of lack of temporal dynamics through the dynamic time window mechanism, and ultimately achieves a comprehensive improvement in the diversity and timeliness of recommendation results, and significantly improves the recommendation accuracy and the breadth of user demand coverage. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart schematically illustrating an AI search recommendation method for procurement in the present invention;

[0021] Figure 2 is a flowchart schematically showing step S2. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0023] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] The embodiment of the present invention discloses an AI search recommendation method for procurement, referring to Figure 1 , including steps S1 to S3:

[0025] S1. Obtain the major category to which each commodity belongs through the purchase item classification catalog.

[0026] It should be noted that determining the major category to which each commodity belongs lays the foundation for subsequent exploration of the correlation and stage-by-stage relationship between major categories.

[0027] Specifically, the procurement item classification catalog is a detailed catalog of specific classifications of procurement objects and is a concretization of the procurement scope. Therefore, the procurement item classification catalog includes the major category to which each commodity belongs.

[0028] S2. Determine whether the relationships between all major categories belong to associated categories and advanced categories.

[0029] It should be noted that the item-based collaborative filtering algorithm only models similarity based on static co-occurrence relationships (such as pairs of items purchased by the same user), ignoring the time series characteristics of user behavior (such as purchase intervals and stage-by-stage needs). This leads to homogeneous recommendation results, that is, the recommendation results are limited to products with similar functions, and cannot distinguish between related related products (for example, toothpaste and toothbrushes) and stage-by-stage advanced products (for example, baby strollers and walkers), making it difficult to explore cross-category associations and advanced needs.

[0030] Therefore, for all major categories, any major category is taken as the target major category, and any major category other than the target major category is recorded as the reference major category, and it is determined whether the relationship between the target major category and the reference major category belongs to the associated category and the advanced category.

[0031] It should be noted that by traversing every two different categories and using them as target categories and reference categories respectively, all possible category combinations are systematically covered to avoid missing potential correlations or advanced relationships, providing a clear comparison benchmark for subsequent steps.

[0032] See the flowchart of step S2 Figure 2 , including steps S201 to S205, specifically:

[0033] S201: Determine relevant users.

[0034] Specifically, the relevant users refer to users who have historical orders containing products belonging to the target category and historical orders containing products belonging to the reference category.

[0035] S202: Determine the degree of correlation between the target category and the reference category in the historical orders of each relevant user.

[0036] Specifically, for all historical orders of a relevant user: the historical orders containing products belonging to the target category are recorded as target orders; based on the interval time period between two historical orders, it is determined whether the target order has any associated orders in all historical orders containing products belonging to the reference category.

[0037] The interval between two historical orders refers to the time period between the first historical order placement time and the second historical order placement time.

[0038] Among them, the target category is recorded as the target category , record the reference category as the reference category ; For the Target orders , among all historical orders containing products belonging to the reference category:

[0039] (1) If all historical orders and target orders for products belonging to the reference category are included There are other target orders in the interval time period, then the target order There are no associated orders.

[0040] (2) Otherwise, the target order will be There are multiple historical orders with no other target orders in the interval time period. The historical order with the shortest interval time is recorded as the target order Related orders.

[0041] It should be noted that by determining whether there are other target orders in the interval time period, and thus screening target orders with related orders, it is possible to avoid misjudging repeated purchase behaviors as related purchase behaviors and capture the user's most direct needs.

[0042] It should be noted that the unit of the interval time period is day.

[0043] Furthermore, based on the number of target orders with associated orders and the length and regularity of the intervals between the target orders with associated orders and their associated orders, the degree of association between the target category and the reference category in the historical orders of the relevant user is determined; the calculation formula for the degree of association between the target category and the reference category in the historical orders of the relevant user is:

[0044] ;

[0045] Where, is the degree of correlation between the target category and the reference category in the historical orders of the relevant user, is the average of the intervals between all target orders and their associated orders. is the number of target orders with associated orders, is the quantity of all target orders, is the number of all historical orders containing products belonging to the reference category, is the standard deviation of the interval time between all target orders and their associated orders. Characterizes the regularity of the interval length between the target order and its associated orders. is the Sigmoid function.

[0046] It should be noted that the calculation formula of the degree of correlation consists of three items: the time mean item , quantity proportion item and time regularity Among them, the purchase time intervals of commodities with associated relationships are shorter, so the smaller the time mean item is, the greater the degree of association between the target category and the reference category in the historical orders of the relevant user; commodities with associated relationships are not sporadic in orders, so the larger the quantity proportion item is, the greater the degree of association between the target category and the reference category in the historical orders of the relevant user; the same user has personal purchasing habits, so for commodities with associated relationships, the purchase time intervals are regular, and the time regularity of the target order and its associated orders is measured by the standard deviation of the time length. The smaller the standard deviation is, the stronger the time regularity is, and the greater the degree of association between the target category and the reference category in the historical orders of the relevant user.

[0047] S203: Determine whether the target category and the reference category belong to related categories.

[0048] Specifically, based on the degree of correlation between the target category and the reference category in the historical orders of all relevant users, determine whether the target category and the reference category belong to the associated class, including: arranging the degree of correlation between the target category and the reference category in the historical orders of all relevant users in order from small to large, and obtaining the first quartile after arrangement: if the first quartile is greater than a preset correlation threshold, the target category and the reference category belong to the associated class; if the first quartile is less than or equal to the preset correlation threshold, the target category and the reference category do not belong to the associated class.

[0049] The specific value of the association threshold can be set according to actual application scenarios and requirements, and the value range of the association threshold is [0.5, 0.75]. The present invention sets the association threshold to 0.6.

[0050] It should be noted that determining the correlation by the first quartile can prevent a small number of highly correlated users from dominating the results. At the same time, weakly correlated combinations are filtered out by a preset correlation threshold, and only strong correlated combinations with high frequency and temporal regularity are retained.

[0051] It should be further explained that association class determination effectively identifies short-term association needs through time proximity screening and multi-dimensional weight calculation (time mean, quantity ratio, regularity).

[0052] S204. If the target category and the reference category do not belong to associated categories, determine the stages of the target category and the reference category in historical orders.

[0053] Specifically, if the target category and the reference category are not associated categories: Based on the number of target orders with associated orders for all relevant users and the regularity of the intervals between target orders with associated orders and their associated orders, determine the stage of the target category and the reference category in historical orders, including:

[0054] 1. Obtain the interval time between the target order and its associated orders for all relevant users, arrange the duration of all interval time periods in ascending order, and calculate the duration in The standard deviation of the duration of all intervals in the range , Indicates the sorted percentile, Indicates the sorted percentile.

[0055] 2. Among them, for Integer in the range, get The smallest value Corresponding range , as the phased duration range of the target category and reference category in historical orders, and The standard deviation of duration corresponding to the stage duration range.

[0056] It should be noted that the initial interval As a benchmark, gradually expand the upper and lower limits, calculate the standard deviation of the interval time period in each expansion interval, and select the expansion interval that minimizes the difference in the standard deviation of the interval time period of adjacent expansion intervals as the staged duration range.

[0057] 3. The calculation formula for the target category and reference category in historical orders is as follows:

[0058] ;

[0059] Where, The stages of target categories and reference categories in historical orders, The duration of the interval between the target order and its associated orders. The number of related users in The duration range of the target category and reference category in historical orders. 、 They are the lower limit and upper limit of the duration range of the target category and reference category in historical orders, respectively. is the number of all relevant users, is the standard deviation of duration corresponding to the stage duration range, is the natural exponential function, is the Sigmoid function, Characterizes the regularity of the interval time between the target order and its associated orders for all relevant users.

[0060] It should be noted that the stage calculation formula consists of two items, namely the quantity proportion item and time regularity , where in all historical orders of each user, the products with advanced relationships will basically appear only once, so the interval time between the target order with associated orders and its associated orders is The number of related users within The closer the ratio to the number of all relevant users is to 1, the greater the stage-by-stage nature of the target category and the reference category in the historical orders of the relevant user; the stage-by-stage upgrades between products have regular time, so the time regularity is measured by the standard deviation of the duration corresponding to the stage-by-stage duration range. The smaller the standard deviation, the stronger the time regularity, and the greater the stage-by-stage nature of the target category and the reference category in the historical orders of the relevant user.

[0061] S205. Determine whether the target category and the reference category belong to the advanced category.

[0062] Specifically, if the stage of the target category and the reference category in historical orders is greater than a preset stage threshold, the target category and the reference category belong to the advanced category.

[0063] The specific value of the stage threshold can be set according to the actual application scenario and requirements, and the value range of the stage threshold is [0.6, 0.8]. The present invention sets the stage threshold to 0.66.

[0064] It should be noted that advanced category judgment captures the phased upgrade needs in the user life cycle through expanded percentile interval and standard deviation analysis, and supports long-term recommendation triggering.

[0065] It should be further explained that step S2 innovatively introduces time series analysis, and accurately distinguishes between the associated class and the advanced class through the correlation degree calculation formula and the stage calculation formula, combined with group user behavior statistics and dynamic threshold judgment.

[0066] S3. Recommend products to the user based on the associated categories and advanced categories of the major categories to which the products in the user's reference order belong.

[0067] It should be noted that the item-based collaborative filtering algorithm only focuses on static co-occurrence and lacks temporal dynamic analysis, resulting in mismatched recommendation timing.

[0068] Therefore, the user's reference order is determined based on the user's advanced reference time and associated reference time, and then products are recommended to the user based on the associated categories and advanced categories of the major categories to which the products in the user's reference order belong.

[0069] The specific steps are:

[0070] 1. The maximum value of the upper limit of the phased duration range of all two categories with advanced relationships in historical orders is recorded as the user's advanced reference duration. ; Get all the historical orders of the user, the order time is The historical orders within the range are recorded as the user's first reference order. For the current moment.

[0071] 2. Obtain the category to which the product in the user's first reference order belongs: If there is an advanced category in this category, further determine the advanced reference duration Is it Within the range, if the advanced reference time exist If the category is within the specified range, the advanced category recommendation will be triggered, and the advanced category of the category will be used as the recommended category for the user. It is the periodic duration range of this category and its advanced categories in historical orders.

[0072] 3. The average of the intervals between the target order and its associated orders in all the user's historical orders is used as the user's associated reference duration. ; Get all the historical orders of the user, the order time is Historical orders within the range and recorded as the user's second reference order, For the current moment.

[0073] 4. Obtain the category to which the product in the user's second reference order belongs: If the category has an associated category, trigger the associated category recommendation, and use the associated category of the category as the user's recommended category.

[0074] 5. Recommend recommended products to users.

[0075] It should be noted that step S3 generates dynamic recommendations based on the time window constraints of the user's recent behavior. Among them, advanced recommendations use the upper limit of the group's stage duration as a reference, combined with extended interval matching to ensure that the timing of recommendations conforms to the user's stage needs; associated recommendations use the individual average interval duration as a window, focusing on immediate associated needs to avoid recommendations too early or too late.

[0076] An embodiment of the present invention also discloses an AI search recommendation system for procurement, including a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, an AI search recommendation method for procurement according to the present invention is implemented.

[0077] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

Claims

1. An AI search and recommendation method for procurement, characterized in that: include: Obtain the major category to which each commodity belongs through the purchase item classification catalog; Take any category as the target category, and any category other than the target category as the reference category; For all historical orders of a relevant user: Record historical orders containing products belonging to the target category as target orders. Based on the intervals between historical orders, determine whether the target order has any associated orders among all historical orders containing products belonging to the reference category. Based on the number of target orders with associated orders and the length and regularity of the intervals between the target orders with associated orders and their associated orders, determine the degree of association between the target category and the reference category in the historical orders of the relevant user. Based on the degree of association between the target category and the reference category in the historical orders of all relevant users, determine whether the target category and the reference category belong to associated categories. If the target category and the reference category do not belong to the associated category: determine the stage of the target category and the reference category in the historical orders based on the number of target orders with associated orders of all relevant users and the regularity of the interval time between the target orders with associated orders and their associated orders. , The stages of target categories and reference categories in historical orders, The duration of the interval between the target order and its associated orders. The number of related users in The duration range of the target category and reference category in historical orders. is the number of all relevant users, is the standard deviation of duration corresponding to the stage duration range, is the natural exponential function, is the Sigmoid function, Characterizes the regularity of the interval length between the target order and its associated orders for all relevant users; and then determines whether the target category and the reference category belong to the advanced category, including: if the stage of the target category and the reference category in historical orders is greater than a preset stage threshold, then the target category and the reference category belong to the advanced category; Recommend products to users based on the associated categories and advanced categories of the major categories to which the products in the user's reference order belong.

2. The AI ​​search and recommendation method for procurement according to claim 1, characterized in that: The related users refer to users who have historical orders containing products belonging to the target category and historical orders containing products belonging to the reference category.

3. The AI ​​search and recommendation method for procurement according to claim 1, characterized in that: Said, including: The step of determining whether the target order has an associated order among all historical orders containing products belonging to a reference category based on the interval time between the historical orders includes: Record the target category as the target category , record the reference category as the reference category ; For the Target orders , among all historical orders containing products belonging to the reference category: If all historical orders containing products belonging to the reference category and the target order There are other target orders in the interval time period, then the target order There is no associated order; otherwise, it will be associated with the target order There are multiple historical orders with no other target orders in the interval time period. The historical order with the shortest interval time is recorded as the target order Related orders.

4. The AI ​​search and recommendation method for procurement according to claim 1, characterized in that: Determining the degree of correlation between the target category and the reference category in the historical orders of the relevant user includes: ； Where, is the degree of correlation between the target category and the reference category in the historical orders of the relevant user, is the average of the intervals between all target orders and their associated orders. is the number of target orders with associated orders, is the quantity of all target orders, is the number of all historical orders containing products belonging to the reference category, is the standard deviation of the interval time between all target orders and their associated orders. Characterizes the regularity of the interval length between the target order and its associated orders. is the Sigmoid function.

5. The AI ​​search and recommendation method for procurement according to claim 1, characterized in that: Determining whether the target category and the reference category belong to an associated category based on the degree of association between the target category and the reference category in historical orders of all relevant users includes: Arrange the degree of correlation between the target category and the reference category in the historical orders of all relevant users in ascending order, and obtain the first quartile after arrangement: if the first quartile is greater than the preset correlation threshold, the target category and the reference category belong to the correlation class; if the first quartile is less than or equal to the preset correlation threshold, the target category and the reference category do not belong to the correlation class.

6. The AI ​​search and recommendation method for procurement according to claim 1, characterized in that: The phased duration range and the duration standard deviation corresponding to the phased duration range include: Get the interval time between the target order and its associated orders for all relevant users, arrange the duration of all interval time periods in ascending order, and calculate the duration in The standard deviation of the duration of all intervals in the range , Indicates the sorted percentile, Indicates the sorted percentile; for Integer in the range, get The smallest value Corresponding range , as the phased duration range of the target category and reference category in historical orders, and The standard deviation of the duration corresponding to the stage duration range.

7. The AI ​​search and recommendation method for procurement according to claim 1, characterized in that: The method for obtaining the user's reference order includes: The maximum value of the upper limit of the phased duration range of all two categories with advanced relationships in historical orders is recorded as the user's advanced reference duration ; Get all the historical orders of the user, the order time is The historical orders within the range are recorded as the user's first reference order. For the current moment; The average of the intervals between the target order and its associated orders in all the user's historical orders is used as the user's associated reference duration. ; Get all the historical orders of the user, the order time is The historical orders within the range are recorded as the user's second reference order.

8. An AI search recommendation system for procurement, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an AI search recommendation method for procurement according to any one of claims 1 to 7 is implemented.

Images