Recommendation method and system based on behavior analysis

Abstract

The invention relates to the field of data processing, in particular to a recommendation method and system based on behavior analysis, and the method comprises the steps: obtaining historical selling data of an unmanned vending machine, and carrying out the preprocessing of the data; according to the single item frequency of commodity selling in the preprocessed historical selling data and the combination frequency of the modes in the Apriori algorithm, the binding degree of each mode is evaluated, a preset minimum support degree threshold value is adjusted based on the binding degree, the binding degree and the preset minimum support degree threshold value are in an inverse relation, a corrected frequent item set is obtained, and the corrected frequent item set is stored in a database; formulating a selling scheme of the commodity combination according to the corrected frequent item set; according to the selling frequency of the set meal and the selling frequency of each commodity in the set meal, the optimal degree of each set meal is obtained, and the set meal is adjusted based on the priority degree of the set meal; according to the method, the binding degree between the commodities is evaluated and quantified, and the minimum support degree threshold value is dynamically adjusted, so that the package demand which better meets the customer is formulated.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a recommendation method and system based on behavior analysis. Background Art

[0002] With the deepening of the new retail concept and the popularization of Internet of Things technology, the unmanned vending machine industry is transforming from the traditional "mechanized delivery" model to "intelligent service". However, the rapid expansion of this industry has also exposed some significant pain points. Traditional unmanned vending machines rely on fixed product combinations and empirical replenishment strategies, which are difficult to adapt to the dynamic changes in consumer demand, resulting in a low average sell-out rate of products. In this context, product combination recommendation technology based on big data has become the key to solving the industry's difficulties.

[0003] "Behavior analysis" is a technology or method that observes, records, analyzes and predicts the behavior patterns of individuals or groups, which can obtain valuable information and provide support for decision-making. Behavior analysis is widely used in many fields such as psychology, marketing, computer science and sociology. In the unmanned vending machine industry, behavior analysis can improve sales efficiency and customer satisfaction by optimizing product mix and recommendation strategies through in-depth research on consumer purchasing behavior.

[0004] Among the existing product recommendation technologies, the Apriori algorithm is often used for product pairing recommendations in unmanned vending machines. Currently, the Apriori algorithm has obvious limitations. For products that are often purchased together with other specific products in the purchase behavior, but have a low sales frequency themselves, it is difficult for the fixed minimum support threshold to identify them as frequent item sets. This results in these products being unable to be effectively sold together with other products, thus missing some potentially useful patterns and reducing the overall efficiency of product sales. Therefore, it seems powerless to cope with the diverse and dynamic needs of consumers. Summary of the invention

[0005] In order to solve the problem that the Apriori algorithm has difficulty in identifying commodities that are sold infrequently but are often purchased together with other specific commodities, resulting in ineffective joint sales and thus reducing overall sales efficiency, the present invention provides solutions in the following aspects.

[0006] In a first aspect, a recommendation method based on behavioral analysis includes: obtaining historical sales data of unmanned vending machines and preprocessing the data; evaluating the degree of binding of each pattern based on the single item frequency of commodity sales in the preprocessed historical sales data and the combination frequency of patterns in the Apriori algorithm, adjusting the preset minimum support threshold based on the degree of binding, wherein the degree of binding is inversely proportional to the preset minimum support threshold, obtaining a modified frequent item set, and formulating a sales plan for a commodity combination based on the modified frequent item set; obtaining the degree of preference of each package based on the sales frequency of the package and the sales frequency of each commodity in the package, and adjusting the package based on the priority of the package; wherein the combination frequency refers to the frequency with which a certain commodity or pattern is purchased in combination with other commodities, the degree of binding refers to the intensity of the frequency with which a certain commodity is purchased together with other specific commodities relative to the frequency with which it is purchased alone, and a sales plan is a package.

[0007] By analyzing the historical sales data of unmanned vending machines, the aim is to solve the problem that the traditional Apriori algorithm is difficult to identify low-frequency and high-binding product combinations. First, the sales data is preprocessed to ensure the accuracy of the data, the binding degree of each product combination pattern is evaluated, and the minimum support threshold is dynamically adjusted according to this binding degree to ensure that those product combinations that are often purchased with other products despite the low frequency of individual sales can be identified as frequent item sets. The degree of preference of each package is calculated based on the actual sales performance of the package, and the package is adjusted accordingly. It is conducive to improving sales efficiency, optimizing inventory management, and enhancing customer satisfaction, thereby promoting the overall performance and market competitiveness of unmanned vending machines. Product combination sales plans can be formulated more accurately to adapt to the diversification and dynamics of consumer needs.

[0008] Preferably, the preprocessing of data includes: According to the historical sales data, obtain all the types of goods sold and sales records, where the sales records include: sales time, sold goods and transaction amount; Unify the selling time into a standard time format, assign a unique code to each product, and convert the transaction amount into a unified currency unit.

[0009] Preprocessing the historical sales data of unmanned vending machines is a key step in implementing the behavior analysis recommendation method. The effect of preprocessing is reflected in the following aspects: first, by unifying the sales time into a standard time format, the consistency and comparability of time data are ensured; second, a unique code is assigned to each product, which improves the accuracy and tracking efficiency of data processing; finally, the transaction amount is converted into a unified currency unit to facilitate financial analysis and comparison. Together, the quality and consistency of the data are ensured, laying a solid foundation for subsequent data analysis and pattern recognition, so that the recommendation system can more accurately capture consumer behavior patterns, formulate more effective product combination sales strategies, and ultimately improve sales efficiency and customer satisfaction.

[0010] Preferably, the combined frequency sequence includes: The combination frequencies of individual products and various combination sales are obtained according to the preprocessed historical sales data, and the combination frequencies are arranged in order from large to small to form a combination frequency sequence.

[0011] It provides a clear perspective to observe and understand consumers' purchasing behavior patterns. It reveals which product combinations are the most popular by combining frequency sequences, thus helping us identify potential high-efficiency sales combinations. It is beneficial for unmanned vending machines to more accurately customize product combinations and packages, optimize inventory management, reduce excess inventory, and increase average order value through bundling sales.

[0012] Preferably, obtaining the binding degree of each mode includes: Taking any product or pattern as a marker product or pattern, the ratio between the maximum combination frequency of the marker product or pattern and the occurrence frequency of the marker product is calculated to obtain the association strength; Calculate the difference between adjacent frequency values ​​in the combination frequency sequence to obtain the degree of frequency difference, and calculate the ratio between the number of times the marked product or pattern appears alone in the historical sales data and the total number of all different combinations in which the marked product or pattern participates minus one to obtain the average distribution value; The maximum value between the average distribution value and 1 is selected as the weighted value, and the product of the weighted value and the degree of association is used as the binding degree of the target product or model.

[0013] It can accurately identify and strengthen the correlation between products, thereby optimizing inventory management, improving sales efficiency, and enhancing customer satisfaction. It also supports adjusting sales strategies according to market dynamics and provides data-driven product combinations and recommendation strategies for vending machines to improve operational efficiency and market competitiveness.

[0014] Preferably, the adjusting the preset minimum support threshold based on the binding degree includes: The ratio between the preset minimum support threshold and the binding degree is used as the significant minimum support threshold. The transaction data is rescanned using the significant minimum support threshold to identify significant frequent itemsets.

[0015] It is conducive to adapting to the diversity and dynamism of consumer needs, making the recommendation system more flexible and accurate, and ultimately promoting sales performance growth and improving market competitiveness.

[0016] Preferably, the significant frequent itemsets include: The historical sales data is sorted out, and the frequency of each product appearing separately is counted to determine the frequent 1-item sets. Based on the frequent 1-item sets, larger candidate item sets are gradually generated, and the frequency of occurrence of these candidate item sets is counted by scanning the data set. The frequent item sets that meet the requirements are screened out according to the set minimum support threshold. Larger item sets are continuously generated and screened until no new frequent item sets can be found, and the significant frequent item sets are obtained.

[0017] It is possible to extract valuable patterns from the raw data, namely, those combinations of products that are often purchased together. The identification of significant frequent itemsets helps optimize inventory management, highlighting which products are often purchased together by customers, so that replenishment and promotion can be more targeted. In addition, these frequent itemsets can also be used to develop more effective marketing strategies and improve customer experience, such as recommending product combinations that may be of interest to customers through bundling or recommendation systems.

[0018] Preferably, obtaining the preference level of each package includes: Taking any package as the target package, calculate the ratio between the sales frequency of the target package in the package sales data and the minimum individual sales frequency of all products in the package to obtain the package performance strength ratio; The package performance intensity ratio is used as the preference level of each package.

[0019] Preferably, obtaining the preference level of each package also includes: Taking any package as the target package, calculate the ratio between the sales frequency of the target package in the package sales data and the minimum individual sales frequency of all products in the package to obtain the package performance strength ratio; Calculate the sum of the historical sales frequency of each product in the target package and the package sales frequency, and then divide it by the average sum of the individual sales frequencies of the corresponding products to obtain the package sales performance index; The product of the package performance intensity ratio and the package sales performance index is used as the preference level of the target package.

[0020] Preferably, the package adjustment based on the package priority includes: A preference level threshold for the preset package is set. If the preference level of the package is less than the preference level threshold, the package is considered unsuitable for combined sale and should be cancelled. Otherwise, if the preference level is greater than or equal to the preference level threshold, the package is considered suitable for combined sale.

[0021] In a second aspect, a recommendation system based on behavior analysis includes: 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 recommendation method based on behavior analysis is implemented.

[0022] The present invention has the following effects: 1. The present invention evaluates and quantifies the degree of binding between commodities by analyzing the single item frequency and combination frequency of commodities in historical sales data, and then dynamically adjusts the minimum support threshold. It can identify commodities that are often purchased together with other specific commodities despite their low individual sales frequency. By incorporating these commodities into the combined sales plan, the average customer price and total sales can be more effectively increased, thereby improving overall sales efficiency.

[0023] 2. The present invention formulates a product combination sales plan through a significant frequent item set, which can reduce inventory backlogs and ensure that the products in the inventory are more in line with consumers' purchasing habits and preferences, which is conducive to reducing inventory costs and improving inventory turnover, so that unmanned vending machines can respond to market changes more flexibly.

[0024] 3. The recommendation method of the present invention through behavioral analysis can provide a product combination that better meets customer needs, thereby improving the customer's shopping experience. By accurately recommending product packages that customers may be interested in, it can increase customer trust and satisfaction with unmanned vending machines, promote customer loyalty, and bring more potential customers to the vending machines. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a method flow chart of steps S1 to S3 in a recommendation method based on behavior analysis in an embodiment of the present invention.

[0026] Figure 2 The figure is a structural block diagram of a recommendation system based on behavior analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0028] Reference Figure 1 , a recommendation method based on behavior analysis includes steps S1 to S3, which are as follows: S1: Obtain historical sales data of the unmanned vending machine and pre-process the data; Collect historical sales data within one month or the last 200 times.

[0029] Based on historical sales data, obtain all types of goods sold and detailed sales records, including but not limited to: sales time, sales goods and transaction amount; Unify the sales time into a standard time format, assign a unique code to each product, and convert the transaction amount into a unified currency unit to ensure data consistency and comparability.

[0030] Further preprocessing of historical sales data is also required, including: removing duplicate records, filling or deleting missing values, and correcting erroneous data to improve data quality, and aggregating data according to time, product and transaction dimensions to better analyze product sales trends and combination sales opportunities.

[0031] Through these preprocessing steps, we can ensure the accuracy and completeness of the data, laying a solid foundation for the subsequent product combination sales analysis. Based on high-quality data, we can further analyze the correlation between products and design effective product combination sales plans, thereby increasing the average order value, reducing inventory, and ultimately improving the overall sales efficiency of the vending machine.

[0032] S2: According to the single item frequency of commodity sales in the preprocessed historical sales data and the combination frequency of patterns in the Apriori algorithm, the binding degree of each pattern is evaluated, and the preset minimum support threshold is adjusted based on the binding degree, where the binding degree is inversely proportional to the preset minimum support threshold, to obtain the modified frequent item set, and formulate a sales plan for the commodity combination based on the modified frequent item set.

[0033] Among them, the combination frequency sequence refers to the frequency with which a certain product or model is purchased in combination with other products. The degree of binding refers to the intensity of the frequency with which a certain product is purchased together with other specific products relative to the frequency with which it is purchased alone. A sales plan is a package.

[0034] Get the combined frequency sequence: Obtain the combined sales frequencies of individual products and various combinations based on historical sales data; Exemplary, commodity The frequency of individual sales is 100 times. With goods The frequency of selling together is 50 times, the product With goods The frequency of selling together is 30 times, the product With goods and products The frequency of selling together is 20 times, and the product is obtained different combination frequencies.

[0035] Arrange the combination frequencies in descending order to form a combination frequency sequence.

[0036] In addition, another embodiment further includes performing cluster analysis on historical sales data: Convert each user's purchase behavior into a feature vector. For example, use the number of purchases of each product as a feature. For example, product type: product ,commodity ,commodity ,commodity ; User 1 purchased 2 items , 1 product , 3 times product , 0 products , the eigenvector is ; User 2 purchased 1 item , 2nd product , 0 products , 1 product , the eigenvector is .

[0037] In addition, you can also add purchase time features, such as dividing the time period into working days and non-working days (Monday to Friday, weekends), setting the purchase feature on working days to 1 and on non-working days to 0.

[0038] Use clustering algorithms (such as K-Means, hierarchical clustering, etc.) to cluster users' purchasing behaviors and obtain several clusters. For each cluster, calculate the frequency of the product combinations purchased by all users in it, extract the purchase records of all users from each cluster, generate all possible product combinations, count the number of times each product combination appears in the cluster, and arrange the combination frequencies in descending order to form a combination frequency sequence.

[0039] It should be noted that if a product, when sold in combination with many other products, only shows a significantly higher sales frequency with a specific product than with other combinations, it can be considered that the product has a high degree of binding (Binding Degree). This phenomenon indicates that the product has a strong dependence in the sales scenario and is more suitable for sale together with the bound products rather than being sold as an independent item.

[0040] Get the binding degree of each mode, including: Taking any product or pattern as a marker product or pattern, the ratio between the maximum combination frequency of the marker product or pattern and the occurrence frequency of the marker product is calculated to obtain the association strength; Calculate the difference between adjacent frequency values ​​in the combination frequency sequence to obtain the degree of frequency difference, and calculate the ratio between the number of times the marked product or pattern appears alone in the historical sales data and the total number of all different combinations in which the marked product or pattern participates minus one to obtain the average distribution value; The maximum value between the average distribution value and 1 is selected as the weighted value, and the product of the weighted value and the degree of association is used as the binding degree of the target product or model.

[0041] Specifically, if a product or a pattern appears in historical sales data much more frequently after being combined with another product than when combined with other products, then the product or pattern is considered to have a high degree of binding. The first product or model among all other products After combining the products ( This product does not belong to This mode is not the first The frequency of each product) is sorted from large to small to get the first Combination frequency sequence of commodities or patterns , among which The frequency value is recorded as , according to The frequency of occurrence of a product or pattern in combination with each other product is used to obtain the degree of binding.

[0042] Specifically, the binding degree satisfies the following relationship: ; In the formula, Indicates The degree of binding of products or models, express, Indicates The frequency of occurrence of a product or pattern, , Respectively represent The first in the frequency sequence of the combination of commodities or patterns and The frequency value of the combination, Indicates The number of product types contained in a product or pattern.

[0043] It should be noted that Represents all possible frequency combinations Take the maximum value, The larger the value of The first The frequency of occurrence of a commodity or pattern is much higher than its individual occurrence frequency, which indicates that the The commodity and other commodities in The combination has a stronger correlation or binding relationship; on the contrary, a smaller value indicates that the The commodity and other commodities in The correlation in the combination is weak.

[0044] For example, if the product With goods The combination frequency Far higher than commodities Frequency of single occurrence , indicating that consumers are buying goods When buying products, they often buy them at the same time .

[0045] Reflect the The frequency of occurrence of a product or pattern Evenly distribute it to all possible combinations, and each combination is assigned a frequency value. This is because when calculating the combined frequency difference, we usually do not consider the In the case of a single product or mode, only in a combination.

[0046] , ensuring that the weighted value of the difference will not be less than 1, avoiding the underestimation of the binding degree due to the small difference, which helps to more accurately identify the correlation between products and optimize product recommendations, inventory management and marketing strategies.

[0047] It means that the calculation starts from the second combination frequency, because It is the highest combination frequency. There is no previous combination frequency to compare with it, so the difference is calculated starting from the second combination frequency.

[0048] By calculating the weighted sum of the combination frequency differences, you can more accurately identify the associations between products and optimize product recommendations. Understanding the weighted sum of the combination frequency differences can help optimize inventory management and ensure that products that are often purchased together have sufficient inventory. Based on the weighted sum of the combination frequency differences, more effective promotions can be designed, such as bundled sales or discounted packages.

[0049] The ratio between the preset minimum support threshold and the binding degree is used as the significant minimum support threshold. The transaction data is rescanned using the significant minimum support threshold to identify significant frequent itemsets.

[0050] Get significant frequent itemsets, including: The historical sales data is sorted out, and the frequency of each product appearing separately is counted to determine the frequent 1-item sets. Based on the frequent 1-item sets, larger candidate item sets are gradually generated, and the frequency of occurrence of these candidate item sets is counted by scanning the data set. The frequent item sets that meet the requirements are screened out according to the set minimum support threshold. Larger item sets are continuously generated and screened until no new frequent item sets can be found, and the significant frequent item sets are obtained.

[0051] For example, the transaction data is: Transaction 1: Transaction 2: Transaction 3: Transaction 4: and transaction 5: ; Set the minimum support threshold to 0.4 (i.e., it appears in at least 2 transactions). Generate frequent 1-item sets as follows: commodity Appeared in transactions 1, 2, 3, and 5, a total of 4 times, with a support of , greater than 0.4, so the product It is frequent.

[0052] commodity Appeared in transactions 1, 2, and 4, a total of 3 times, with a support of , greater than 0.4, so the product It is frequent.

[0053] commodity Appeared in transactions 1, 3, and 4, a total of 3 times, with a support of , greater than 0.4, so the product It is frequent.

[0054] commodity Appeared in transaction 5, appeared 1 time in total, with support of , less than 0.4, so the product Not frequently.

[0055] Therefore, the frequent 1-item set is: , , .

[0056] Generate frequent 2-item sets as: Product portfolio It appears 2 times in total, with a support of , equal to 0.4, so product B is frequent.

[0057] Product portfolio It appears 2 times in total, with a support of , equal to 0.4, so product B is frequent.

[0058] Product portfolio It appears 2 times in total, with a support of , equal to 0.4, so product B is frequent.

[0059] Product portfolio Appeared 1 times in total, with support of , is less than 0.4, so product D is not frequent.

[0060] Therefore, the frequent 2-itemset is , , .

[0061] Generate frequent 3-item sets as: Product portfolio : 1 time, which is less than the minimum support threshold, so there is no frequent item set in the frequent 3-item set; As a result, the frequent itemsets are: , , , , , .

[0062] In this implementation, the preset minimum support threshold is , which can be adjusted according to specific circumstances.

[0063] By extracting frequent item sets from historical sales data, calculating the binding degree of each product or pattern, and forming a combined frequency sequence, this information can be used to optimize product recommendations, inventory management, and marketing strategies, thereby improving the sales efficiency and customer satisfaction of unmanned vending machines.

[0064] S3: According to the selling frequency of the package and the selling frequency of each product in the package, the priority of each package is obtained, and the package is adjusted based on the priority of the package.

[0065] It should be noted that if after the combination package is sold, the package sales frequency is too low or the package sales frequency seriously affects the individual sales frequency of each product therein, then the package option is considered to have a low degree of preference and such a package should be cancelled by increasing the minimum support threshold.

[0066] Get the best of each plan, including: Taking any package as the target package, calculate the ratio between the sales frequency of the target package in the package sales data and the minimum individual sales frequency of all products in the package to obtain the package performance strength ratio; The package performance intensity ratio is used as the preference level of each package.

[0067] Specifically, after launching the product package sale, the sales records collected within one month or one hundred times are recorded as the package sales data, and the number of frequent item sets obtained is recorded as , each frequent item set corresponds to a product package combination selling plan, The package is recorded as Product collection .

[0068] Specifically, the preference of the set meal satisfies the following relationship: ; In the formula, Indicates The preference of the package, Indicates The sales frequency of this package in the package sales data, Indicates The minimum value of the individual sales frequency of all products in the package. Indicates The sales frequency of a product in historical sales data.

[0069] That is to say, This reflects the individual sales of the “least popular” item in the package. Indicates The strength of the sales frequency of the first set meal relative to the sales frequency of the "least popular" item in the set meal. The sales frequency of this package is much higher than the sales frequency of any product in the package alone, indicating that the This type of package has a better market performance, and the products in the package may not perform well when sold separately.

[0070] In addition, another embodiment further includes: Taking any package as the target package, calculate the ratio between the sales frequency of the target package in the package sales data and the minimum individual sales frequency of all products in the package to obtain the package performance strength ratio; Calculate the sum of the historical sales frequency of each product in the target package and the package sales frequency, and then divide it by the average sum of the individual sales frequencies of the corresponding products to obtain the package sales performance index; The product of the package performance intensity ratio and the package sales performance index is taken as the preference degree of the target package.

[0071] Get the sales frequency of each package and each single item The sales frequency of a commodity is recorded as , will The selling frequency of the package is recorded as , will The number of types of goods included in the package sales is recorded as .

[0072] Specifically, the preference of the set meal satisfies the following relationship: ; In the formula, Indicates The preference of the package. Indicates The sales frequency of this package in the package sales data, Indicates The minimum value of the individual sales frequency of all products in the package. Indicates The sales frequency of the product in the historical sales data, Indicates The number of products included in the package. Indicates The product collection corresponding to the package. Indicates The sales frequency of a product in historical sales data.

[0073] That is to say, Indicates The strength of the combined performance of a product in historical sales and package sales relative to its individual sales frequency. The larger the value, the greater the help of the package sales method to the sales efficiency of each product in the package, that is, the higher the preference of the package, and vice versa. The smaller the value, the less helpful the package sales method is to the sales efficiency of each product in the package, that is, the lower the preference of the package.

[0074] By comparing the individual sales frequency of each item in the package, we can understand which items in the package have poor market performance and need to further analyze the reasons or adjust the marketing strategy.

[0075] A preference level threshold for the preset package is set. If the preference level of the package is less than the preference level threshold, the package is considered unsuitable for combined sale and should be cancelled. Otherwise, if the preference level is greater than or equal to the preference level threshold, the package is considered suitable for combined sale.

[0076] Exemplarily, the preference threshold is 0.7, which can be adjusted according to specific circumstances.

[0077] The present invention also provides a recommendation system based on behavior analysis. Figure 2As shown, the system includes a processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a recommendation method based on behavior analysis according to the first aspect of the present invention is implemented. The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, whose settings and functions are known in the art, and therefore will not be described in detail here.

[0078] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these modifications and improvements all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A recommendation method based on behavior analysis, characterized in that: include: Obtain historical sales data of unmanned vending machines and pre-process the data; According to the single item frequency of commodity sales in the preprocessed historical sales data and the combination frequency of patterns in the Apriori algorithm, the binding degree of each pattern is evaluated, and the preset minimum support threshold is adjusted based on the binding degree, where the binding degree is inversely proportional to the preset minimum support threshold, to obtain the modified frequent item set, and formulate a sales plan for the commodity combination based on the modified frequent item set; According to the sales frequency of the package and the sales frequency of each product in the package, the priority of each package is obtained, and the package is adjusted based on the priority of the package; Among them, combination frequency refers to the frequency with which a certain product or model is purchased in combination with other products; the degree of binding refers to the strength of the frequency with which a certain product is purchased together with other specific products relative to the frequency with which it is purchased alone; a sales plan is a package.

2. The recommendation method based on behavior analysis according to claim 1, characterized in that: The preprocessing of data includes: According to the historical sales data, obtain all the types of goods sold and sales records, where the sales records include: sales time, sold goods and transaction amount; Unify the selling time into a standard time format, assign a unique code to each product, and convert the transaction amount into a unified currency unit.

3. The recommendation method based on behavior analysis according to claim 1, characterized in that: The combined frequency sequence comprises: The combination frequencies of individual products and various combination sales are obtained according to the preprocessed historical sales data, and the combination frequencies are arranged in order from large to small to form a combination frequency sequence.

4. The recommendation method based on behavior analysis according to claim 1, characterized in that: The obtaining of the binding degree of each mode includes: Taking any product or pattern as a marker product or pattern, the ratio between the maximum combination frequency of the marker product or pattern and the occurrence frequency of the marker product is calculated to obtain the association strength; Calculate the difference between adjacent frequency values ​​in the combination frequency sequence to obtain the degree of frequency difference, and calculate the ratio between the number of times the marked product or pattern appears alone in the historical sales data and the total number of all different combinations in which the marked product or pattern participates minus one to obtain the average distribution value; The maximum value between the average distribution value and 1 is selected as the weighted value, and the product of the weighted value and the degree of association is used as the binding degree of the target product or model.

5. The recommendation method based on behavior analysis according to claim 1, characterized in that: The adjusting the preset minimum support threshold based on the binding degree includes: The ratio between the preset minimum support threshold and the binding degree is used as the significant minimum support threshold. The transaction data is rescanned using the significant minimum support threshold to identify significant frequent itemsets.

6. The recommendation method based on behavior analysis according to claim 1, characterized in that: The significant frequent item sets include: The historical sales data is sorted out, and the frequency of each product appearing separately is counted to determine the frequent 1-item sets. Based on the frequent 1-item sets, larger candidate item sets are gradually generated, and the frequency of occurrence of these candidate item sets is counted by scanning the data set. The frequent item sets that meet the requirements are screened out according to the set minimum support threshold. Larger item sets are continuously generated and screened until no new frequent item sets can be found, and the significant frequent item sets are obtained.

7. The recommendation method based on behavior analysis according to claim 1, characterized in that: The obtaining of the preference level of each package includes: Taking any package as the target package, calculate the ratio between the sales frequency of the target package in the package sales data and the minimum individual sales frequency of all products in the package to obtain the package performance strength ratio; The package performance intensity ratio is used as the preference level of each package.

8. The recommendation method based on behavior analysis according to claim 1, characterized in that: The obtaining of the preference level of each package further includes: Taking any package as the target package, calculate the ratio between the sales frequency of the target package in the package sales data and the minimum individual sales frequency of all products in the package to obtain the package performance strength ratio; Calculate the sum of the historical sales frequency of each product in the target package and the package sales frequency, and then divide it by the average sum of the individual sales frequencies of the corresponding products to obtain the package sales performance index; The product of the package performance intensity ratio and the package sales performance index is used as the preference level of the target package.

9. The recommendation method based on behavior analysis according to claim 1, characterized in that: The package adjustment based on the package priority includes: A preference level threshold for the preset package is set. If the preference level of the package is less than the preference level threshold, the package is considered unsuitable for combined sale and should be cancelled. Otherwise, if the preference level is greater than or equal to the preference level threshold, the package is considered suitable for combined sale.

10. A recommendation system based on behavioral analysis, 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, the recommendation method based on behavior analysis according to any one of claims 1 to 9 is implemented.

Images