Data mining method based on dynamic graph neural network

Abstract

The invention discloses a data mining method based on a dynamic graph neural network, and relates to the technical field of data mining, and the method comprises the steps: obtaining the purchase information of each user in a platform; performing data mining on the browsing information of the user, and determining behavior properties of the user during purchase; performing historical matching on the careful deliberate commodities, performing commodity information analysis on commodities with repeated characteristics in the careful deliberate commodities, and determining purchase influence factors; marshalling commodities purchased at the same time based on the time information to obtain commodity group data, matching and analyzing the commodities according to the commodity group data, and determining the correlation degree and the purchase cycle between the commodities; the retrieval information of the user is read, commodity recommendation is conducted on the user according to the retrieval information and the purchase influence factors, purchase adding information of the user is obtained, association degree matching is conducted on the purchased commodities according to the purchase adding information, and the associated commodities are recommended. The method has the effect of improving the commodity recommendation accuracy.

Description

Technical Field

[0001] The present application relates to the field of data mining technology, and in particular to a data mining method based on dynamic graph neural network. Background Art

[0002] Dynamic graph neural network is a machine learning model used to process dynamic graphs. It solves the time dependency problem that traditional static graph models cannot handle by capturing the temporal evolution of graph structure and the dynamic relationship between nodes. Data mining refers to the process of automatically discovering hidden, valuable information or patterns from large amounts of data through algorithms or statistical methods.

[0003] In the existing technology, when e-commerce platforms recommend products to users, they usually judge and analyze a large amount of user purchase data to determine the user's tags and purchasing habits, and then divide users with the same tags. In this way, products are recommended to users based on the divided tags and purchasing habits, so as to achieve the effect of accurately recommending products and increase the user's purchase probability.

[0004] Regarding the above-mentioned related technologies, in the process of recommending products to users, although the products purchased by users with the same label are recommended to each other through the user's label, which effectively solves the problem of the single category of recommended products, due to the different needs of different users, the correlation between the recommended products and the products actually needed by the users is low, which in turn reduces the possibility of users purchasing the recommended products. There is room for improvement. Summary of the Invention

[0005] In order to improve the accuracy of product recommendations, this application provides a data mining method based on dynamic graph neural network.

[0006] This application provides a data mining method based on dynamic graph neural network, which adopts the following technical solutions:

[0007] A data mining method based on a dynamic graph neural network, comprising:

[0008] Step S1, obtaining purchase information of each user on the platform; the purchase information includes time information when the product is purchased and corresponding product information, as well as browsing information when the product is purchased;

[0009] Step S2, data mining is performed on the user's browsing information to determine the nature of the user's behavior when purchasing each product; the behavior nature includes careful consideration and quick decision;

[0010] Step S3: Perform historical matching on carefully considered products to determine whether there are any products that meet the repetitive characteristics among the carefully considered products, and analyze the product information of the products to determine the factors that influence purchase;

[0011] Step S4: Based on the time information, the products purchased at the same time are grouped to obtain product group data with a time sequence, and the products are matched and analyzed according to the product group data to determine the correlation between the products and the purchase cycle;

[0012] Step S5: Read the user's search information. If the user's search information is not read, determine the recommended products for the current day based on the product's purchase cycle and purchase information and make recommendations.

[0013] In step S6, if the user's search information is read, product recommendations are made to the user based on the search information and purchase influencing factors, the user's additional purchase information is obtained, and the additional purchase products are matched with the correlation degree based on the additional purchase information, and the products related to the additional purchase products are determined, and the related products are recommended.

[0014] Preferably, the purchase time data of each product is obtained, and the browsing information is matched according to the purchase time data to determine whether there is browsing information in the corresponding time period, and the corresponding browsing information is matched with the corresponding product information;

[0015] If there is corresponding browsing information before purchasing the product, the nature of the behavior of purchasing the product is determined to be a careful consideration type; otherwise, the nature of the behavior of purchasing the product is determined to be a quick decision type.

[0016] Preferably, product information matching is performed on quick-decision products to determine whether the same product exists in historical purchase information and whether the behavior of the product in the historical purchase information is carefully considered;

[0017] If it is determined that the same product exists in the historical purchase information and the nature of the purchase behavior is carefully considered, the behavior nature of the product is re-marked and marked as carefully considered.

[0018] Preferably, the repetitive characteristic is that there are multiple purchase behaviors in the historical purchase process.

[0019] Preferably, product information matching is performed on products with a carefully considered behavior nature, and all purchase information of the user when purchasing the same product in historical purchase information is determined to obtain purchase information of the same product;

[0020] Determine the proportion of purchase behavior for the same product, and determine the proportion of careful consideration and quick decision-making.

[0021] If the percentage is greater than the preset benchmark value, it is determined that the user's purchase behavior preference for this product is a carefully considered purchase;

[0022] Reading the product information of the carefully considered purchase, and performing product evaluation index judgment on the read product information, determining product evaluation index items with the same product evaluation index data in the read multiple product information, obtaining a first common index item, and screening to obtain a first product;

[0023] Reading browsing information of the first product, comparing product evaluation indicators with the browsing information of the corresponding product, determining that the first product is superior to product evaluation indicators of the products in the corresponding browsing information, and obtaining a first indicator item set;

[0024] The product evaluation index items of the first index item set are compared to determine the common index items of each purchase in the first index item set, and the common index items are matched with the first common index items, and the same product evaluation index items in the common index items and the first common index items are used as purchase influencing factors for the user's product purchase.

[0025] Preferably, based on the purchase influencing factors, a big data analysis is performed on the first product and its corresponding browsing information to determine the data change amount of the corresponding purchase influencing factors.

[0026] Preferably, step S41 is to match commodities among commodity group data based on time sequence, determine the purchase time interval of commodities of the same category, and obtain purchase cycle data of each category of commodities;

[0027] Step S42, permuting and combining the quantities of different commodity categories in the commodity group data to obtain combination data corresponding to the commodity group data;

[0028] Step S43: Perform data matching on the product group data of different groups, using the product group data as a unit, to determine the overlap ratio of various permutations and combinations in the combined data, and then determine the degree of association between the products in the permutations and combinations based on the overlap ratio. The overlap ratio is the ratio of the number of overlapping groupings to the product group data containing the product.

[0029] Preferably, the commodities are classified based on the commodity information to obtain first category data;

[0030] Based on the built-in distinguishing items, the products of the same category are divided according to function, appearance, and usage adjustment to obtain the second category of product data.

[0031] Preferably, the method for calculating the overlap rate of goods is:

[0032] Selecting a product from the historical purchase record to obtain the original product, and using the product as the subject, matching the product group data to determine the product group data containing the original product, and counting the number of the matched product group data to obtain a first data volume;

[0033] Based on the matched product group data, the original products are randomly combined to obtain combination data corresponding to the product group data, and the number of the same combination data of each matched product group data is counted to obtain a second data volume;

[0034] The ratio of the second data volume to the first data volume is calculated to obtain the overlap rate of the corresponding product combination.

[0035] In summary, this application includes at least one of the following beneficial technical effects:

[0036] 1. By acquiring the purchase information of users on the platform to establish a basic database, and by conducting data mining on the browsing information of users before purchasing products, the nature of the user's purchasing behavior when purchasing each product is determined. By performing dynamic graph neural network analysis on carefully considered products, the factors that influence the user's purchase of such products are determined. By grouping the products purchased at the same time and then conducting data mining on the products in each group, the user's combination purchase habits of other products after purchasing a product are determined. At the same time, by matching the products in each group, the user's time purchase habits for a certain product are determined. In this way, the purchase cycle, purchase influencing factors and correlation are used to prioritize the products on the platform according to different user needs, improve the accuracy of recommendations, and thus increase the user's purchase rate on the platform;

[0037] 2. By determining the proportion of purchase behavior in historical purchase records for each product, the user's purchasing behavior preferences when purchasing each product are determined. This allows the user to filter through a large number of historical purchase records, thereby reducing the amount of data to be analyzed. By reading and jointly analyzing carefully considered purchases, the amount of data to be analyzed is further reduced while also determining the common purchase characteristics of users when purchasing that type of product. By reading and judging the browsing information of the first product, the unique characteristics of the product selected by the user in each purchase are determined. By comparing the product evaluation indicators obtained from the two judgments and filters, the user's purchase influencing factors are further determined. This allows the recommended products to better meet the user's needs when making purchase recommendations based on these factors, thereby increasing the platform's product transaction rate.

[0038] 3. By arranging and combining product group data, the combination categories that can be included in a product group data are determined. Then, when making a combination judgment on a certain product, the product group data containing the product is matched with other products to determine the correlation between the product and other products. Then, when the user purchases the product, the recommended products can better meet the user's needs, thereby improving the transaction rate of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flowchart of the steps of the data mining method based on the dynamic graph neural network in this embodiment; DETAILED DESCRIPTION

[0040] The following is combined with Figure 1 This application is described in further detail.

[0041] The embodiment of the present application discloses a data mining method based on a dynamic graph neural network.

[0042] Example: Figure 1 As shown, the present invention provides a data mining method based on a dynamic graph neural network, comprising:

[0043] Step S1, obtaining purchase information of each user on the platform; the purchase information includes time information when the product is purchased and corresponding product information, as well as browsing information when the product is purchased;

[0044] Step S2, data mining is performed on the user's browsing information to determine the nature of the user's behavior when purchasing each product; the behavior nature includes careful consideration and quick decision;

[0045] Specifically, they are:

[0046] S21, obtaining the purchase time data of each product, matching the browsing information according to the purchase time data, determining whether there is browsing information in the corresponding time period, and matching the corresponding browsing information with the corresponding product information;

[0047] S22: If there is corresponding browsing information before purchasing the product, then the nature of the behavior of purchasing the product is determined to be a careful consideration type; otherwise, then the nature of the behavior of purchasing the product is determined to be a quick decision type.

[0048] S23, obtaining the purchase time data of each product, matching the browsing information according to the purchase time data, determining whether there is browsing information in the corresponding time period, and matching the corresponding browsing information with the corresponding product information;

[0049] S24, if there is corresponding browsing information before purchasing the product, then the nature of the behavior of purchasing the product is determined to be a careful consideration type; otherwise, then the nature of the behavior of purchasing the product is determined to be a quick decision type.

[0050] Step S3: Perform a historical match on carefully considered products to determine whether there are any products that meet the repeat feature. Then, analyze the product information to determine the factors influencing purchases. The repeat feature is that there are multiple purchases in the purchase history, for example, paper towels.

[0051] Step S4: Based on the time information, the products purchased at the same time are grouped to obtain product group data with a time sequence, and the products are matched and analyzed according to the product group data to determine the correlation between the products and the purchase cycle;

[0052] Step S5: Read the user's search information. If the user's search information is not read, determine the recommended products for the current day based on the product's purchase cycle and purchase information and make recommendations.

[0053] In step S6, if the user's search information is read, product recommendations are made to the user based on the search information and purchase influencing factors, the user's additional purchase information is obtained, and the additional purchase products are matched with the correlation degree based on the additional purchase information, and the products related to the additional purchase products are determined, and the related products are recommended.

[0054] In this embodiment, the purchase information of users on the platform is obtained to establish a basic database. By performing data mining on the browsing information of users before purchasing products, the behavioral nature of the user's purchasing behavior when purchasing each product is determined. By performing dynamic graph neural network analysis on carefully considered products, the purchase influencing factors that affect the user's choice when purchasing such products are determined. By grouping products purchased at the same time and then performing data mining on the products in each group as a unit, the user's combination purchase habits of other products after purchasing a product are determined. At the same time, by matching products in each group, the user's time purchase habits for a certain product are determined. This allows the purchase cycle, purchase influencing factors and correlation to be used to prioritize recommendations for products on the platform according to different user needs, thereby improving the accuracy of recommendations and thereby increasing the user's purchase rate on the platform.

[0055] For example, in the process of using dynamic graph neural networks to conduct data mining on the purchase information of platform users to determine the user's purchasing habits, the user's purchase information is first obtained to build a database, and then the user's browsing information, product information and purchase time information are mined to determine the user's purchasing habits. The next time the user uses the platform, more accurate product push is made to the user, thereby maximizing the platform's transaction volume.

[0056] By mining user browsing data, if a user repeatedly compares and filters products before purchasing, it indicates that the user carefully considers the product before purchasing. Conversely, if the user purchases the product directly without comparing or filtering, this purchase behavior is a quick decision. By judging the user's purchase behavior of a large number of the same products, the user's purchase behavior is determined. However, since quick-decision products are not recommended, the recommendation algorithm only recommends carefully considered products.

[0057] When recommending carefully considered products, since users often search and compare the same product before purchasing, it's important to determine the factors that influence each purchase. For example, if a user buys Brand A's product a today and Brand B's product b tomorrow, analyzing and judging large amounts of data can more accurately and clearly determine the factors that lead to these multiple purchases of the same product. For example, factors such as price, craftsmanship, brand, store, and reviews can be used.

[0058] After determining the user's purchasing habits for the same product over time, the correlation between multiple products in the same purchase record is re-evaluated to determine which related products the user will purchase after purchasing one product. For example, if a user purchases product A and product B, they will most likely also purchase product A. In this case, a correlation between products A and B is determined. By utilizing the time information of each time a user purchases the same product, the user's purchase cycle for the product is determined, thereby determining the purchase cycle of the product. For example, if a user purchases a computer, they will also purchase corresponding extension products such as a keyboard and mouse. Therefore, when the user purchases a computer, extension product recommendations will be made.

[0059] After determining the user's past purchase habits for the same product and the user's interrelated purchasing habits when purchasing products, if the user only browses products on the platform, the platform cannot detect the user's search information. It then matches the recommended cycle of products purchased in the history based on the purchase cycle and recommends products that have reached the recommendation cycle. For example, if a user purchases a lotion and lotion set once a month, when recommending the lotion and lotion set, it first determines the date of the most recent purchase of the lotion and lotion set in the historical purchase record, and then determines whether the purchase period of the lotion and lotion set has reached one month based on the current date. If so, it is recommended; otherwise, it is not recommended. Not recommending means not making a priority recommendation.

[0060] If a user searches for a product, we recommend products based on factors influencing purchases, ensuring they can quickly identify the products they need, thereby increasing the platform's transaction rate. Furthermore, after a user adds a product to their cart or purchases it, we identify related products based on the added or purchased item and recommend those products. For example, if a user is purchasing a computer, we prioritize recommendations for keyboards, mice, and other items.

[0061] In step S3, historical matching is performed on carefully considered products to determine whether there are products that meet the repetitive characteristics among the carefully considered products. Product information of the products is analyzed to determine purchase influencing factors, including the following steps:

[0062] S31, matching product information for products with a behavior attribute of "careful consideration" to determine all purchase information of the user in the past purchase information of the same product, and obtaining purchase information of the same product;

[0063] S32, determining the proportion of behavior types in purchasing information of the same product, and determining the proportion of careful consideration and quick decision-making;

[0064] S33, if the proportion data is greater than the preset reference value, it is determined that the user's purchase behavior preference for the product is a carefully considered purchase;

[0065] S34, reading the commodity information of the carefully considered purchase, and judging the commodity evaluation index of the read commodity information, determining the commodity evaluation index items with the same commodity evaluation index data in the multiple commodity information read, obtaining the first common index item, and screening to obtain the first commodity; wherein the commodity evaluation index item can be multiple collectible index items such as store, price, brand, quality, etc.

[0066] S35, reading browsing information of the first product, comparing product evaluation indicators with the browsing information of the corresponding product, determining that the first product is superior to the product evaluation indicators of the products in the corresponding browsing information, and obtaining a first indicator item set;

[0067] S36, compare the product evaluation index items of the first index item set, determine the index items common to each purchase in the first index item set, match the common index items with the first common index items, and use the same product evaluation index items in the common index items and the first common index items as purchase influencing factors for the user's product purchase.

[0068] S37: Based on the purchase influencing factors, a big data analysis is performed on the first product and its corresponding browsing information to determine the data variation of the corresponding purchase influencing factors. By utilizing a big data analysis algorithm to determine the range of influence of each purchase influencing factor, when the purchase influencing factor of a product falls within the data variation range, it is determined that the product can be given priority recommendation; otherwise, it is determined that the product cannot be given priority recommendation. For example, if the purchase influencing factor is price, when a user selects a product based on price, a phenomenon may occur: the higher the price, the more likely the user is to choose the product, and the lower the price, the more likely the user is to choose the product. Instead, within a certain data range, the higher / lower the price, the more likely the user is to choose the product.

[0069] In this embodiment, the proportion of the nature of the purchase behavior in the historical purchase records of each commodity is judged to determine the user's purchase behavior preference when purchasing each commodity, and then a large number of historical purchase records are screened to reduce the amount of data to be analyzed. Then, by reading and jointly analyzing the carefully purchased commodities, not only the amount of data to be analyzed is further reduced, but also the common purchase characteristics of the user when purchasing this type of commodity are determined. Then, by reading and judging the browsing information of the first commodity, the unique nature of the commodity selected by the user in each purchase is determined. Then, by comparing the commodity evaluation indicators obtained from the two judgments and screenings, the purchase influencing factors of the user when purchasing this type of commodity are further determined, so that when recommending commodities to users based on the purchase influencing factors, the recommended commodities are more likely to meet the needs of users, thereby improving the platform commodity transaction rate.

[0070] For example, since carefully considered products are products purchased after users have screened them, product analysis on carefully considered products is to analyze the user's screening behavior, thereby determining the user's reference conditions when purchasing the product, and thereby enabling the user's purchase conditions to be met more quickly when products are recommended to the user based on the reference conditions, thereby achieving the greatest possible transaction.

[0071] First, by utilizing the user's behavior when purchasing goods, the goods that need to be analyzed are divided, thereby reducing the amount of data to be analyzed and improving the analysis rate.

[0072] By matching the analyzed items with historical purchase records, we can determine whether users generally carefully consider their purchases of the item. If so, this indicates that users need to carefully consider their purchases, further reducing the amount of data required for analysis of purchase influencing factors and improving the accuracy of the analysis. We then analyze the purchased product information and browsing information for the items after the two rounds of screening to identify the common characteristics of the same item purchased by the user, as well as the unique characteristics between the items purchased and the items browsed by the user. Specifically, we can determine why users chose a particular item among a large number of items based on browsing information, and determine the user's selection patterns based on purchased product information, ultimately leading to the factors influencing their purchases.

[0073] For example, if there are 10 purchase records for product A, 8 of which involved careful consideration and 2 did not, the percentage is 0.8. Assuming a preset baseline value of 0.6, the user's purchase preference for product A is determined to be careful consideration. For example, if the 8 careful considerations are numbered 1 to 8, the information for products 1 to 8 is compared to identify common indicators, such as price, quality, and brand. Browsing information is then evaluated for product 1. This involves comparing its information with the information in its corresponding browsing information to identify indicators unique to product 1, such as price and quality. This is repeated for indicators for products 2 to 8. The common indicators for products 1 to 8 are then matched with the unique indicators for products 1 to 8 to identify the indicators shared by products 1 to 8 in both the first common indicator set and the first indicator set. These indicators are then marked as purchase influencing factors.

[0074] In step S4, based on the time information, the products purchased at the same time are grouped to obtain product group data with a time sequence, and the products are matched and analyzed according to the product group data to determine the correlation between the products and the purchase cycle, including the following steps:

[0075] In step S41, based on the time sequence, the product group data is matched to determine the purchase time interval of the same type of products, and the purchase cycle data of each type of products is obtained; wherein, when the same type of products are divided, they include major categories and minor categories, major categories refer to products of the same type, and minor categories refer to products of the same type with the same functions and appearance.

[0076] Specifically:

[0077] S411, classify the products based on the product information to obtain the first category of data; that is, the major category.

[0078] S412: Based on the built-in distinguishing factors, products in the same category are divided by function, appearance, and usage adjustment to obtain the second category of product data. This is called a subcategory. For example, facial masks are a large category in themselves, but if they are further divided by function, they can be divided into hydrating masks, whitening masks, etc.

[0079] Step S42: Permutations and combinations are performed on the quantity of different product categories in the product group data to obtain combination data corresponding to the product group data; the combination data is a combination of categories of other products with one product as the main component. For example, if product group data A contains three products, a, b, and c, and product a is the main component, the combination data obtained are (b), (c), and (b, c); if product b is the main component, the combination data obtained are (a), (c), and (a, c); if product c is the main component, the combination data obtained are (a), (b), and (a, b). The combination data of product group data A is the combination data of product a as the main component, the combination data of product b as the main component, and the combination data of product c as the main component.

[0080] Step S43: Perform data matching on the product group data of different groups, using the product group data as a unit, to determine the overlap ratio of various permutations and combinations in the combined data, and then determine the degree of association between the products in the permutations and combinations based on the overlap ratio. The overlap ratio is the ratio of the number of overlapping groupings to the product group data containing the product.

[0081] In this embodiment, by reading the information of historically purchased goods and classifying the goods according to the goods information, the category of the goods is determined, and the goods are matched according to the category of the goods to determine the user's purchase cycle on the goods. By arranging and combining the goods group data, the combination categories that can be included in a goods group data are determined. Then, when making a combination judgment on a certain goods, the remaining goods are matched with the goods group data containing the goods to determine the correlation between the goods and other goods. Then, when the user purchases the goods, the recommended goods can better meet the user's needs, thereby improving the transaction rate of the platform.

[0082] For example, when performing product association data mining on purchased product records, the products in the historical purchase records are first classified to determine the categories of the products, so that product information of the same category can be mined during data mining, thereby improving data quality.

[0083] For example, there are products a, b, c, and d in the historical purchase records. By classifying the products, for example, products in category AB include a and b; products in category C include c, and products in category D include d, then when judging the purchase cycle of products in the same category, for example, category AB, the time of all purchased records of a and b in the historical purchase records is read to obtain time points a1, a2, a3, b1, b2, and b3. The time intervals of time points a1, a2, a3, b1, b2, and b3 are analyzed and judged to determine the time span when the user purchased products in category AB.

[0084] When combining product categories in product group data, for example, there is group 1: product a, product b, product c;

[0085] Group 2: Product a, Product b, Product d;

[0086] Group 3: Product a, Product d, Product e.

[0087] The permutations and combinations for group 1 are: for product a: (b), (c), (b, c); for product b: (a), (c), (a, c); for product c: (a), (b), (a, b); the permutations and combinations for group 2 are: for product a: (b), (d), (b, d), for product b: ...; the permutations and combinations for group 3 are: for product a: (e), (d), (e, d), ...;

[0088] When judging the overlap rate, product a is matched, and it is determined that the product group data containing product a includes group 1, group 2, and group 3. By comparing the pairs of product a in group 1: (b), (c), (b, c) with the pairs of product a in group 2: (b), (d), (b, d), and comparing group 1 with group 3, it is found that when users purchase product a, they will mostly purchase product b at the same time. Therefore, it is determined that there is an association between product a and product b, and the association degree is calculated: there are 3 groups that purchase product a, of which two meet the (a, b) product requirement at the same time, so the association degree of a to b is two-thirds.

[0089] In step S43, combining data matching is performed on the product group data of different groups in units of product group data, the overlap rate of various permutations and combinations in the combined data is determined, and the association degree between the products in the permutations and combinations is determined based on the overlap rate, including the following steps:

[0090] S431: Select a product from the historical purchase record to obtain the original product. Use the product as the subject to match the product group data to determine the product group data that includes the original product. Count the number of the matched product group data to obtain a first data volume.

[0091] S432: Based on the matched product group data, the original products are randomly combined to obtain combination data corresponding to the product group data, and the number of the same combination data of each matched product group data is counted to obtain a second data volume;

[0092] S433: Calculate the ratio of the second data volume to the first data volume to obtain the overlap rate of the corresponding product combination.

[0093] For example, when calculating the overlap rate of goods, first select the main product in a group of goods, then determine the situations in which the goods in the group can be combined with other goods, and match and filter the historical purchase records by selecting the main product to determine the group that contains the product in other product group data. For example, group 1 contains products a, b, and c. Product a is selected as the main body, and other groups are matched according to product a to obtain groups with product a. For example, there is product a in group 2 and group 3. Then the products in group 1 that can be combined with product a are, for example, (a, b) and (a, c), the combination results of products that can be combined with product a in group 2 are (a, b) and (a, d), and the combination results of products that can be combined with product a in group 3 are (a, c) and (a, d). Through statistics, it is determined that there are 3 product groups containing product a, two of which contain product b, two contain product c, and two contain product d. Therefore, it can be judged that the correlation degree of product a to product b is two-thirds, the correlation degree of product a to product c is two-thirds, and the correlation degree of product a to product d is two-thirds. However, if b is taken as the main body, the groups matched with product b are group 4 and group 5. The combination results of products that can be combined with product b in group 4 are (b, c) and (b, d), and the combination results of products that can be combined with product b in group 5 are (b, d) and (b, e). Then, through statistics, it is determined that there are 4 product groups containing product b, of which two contain product a, two contain product c, and three contain product d. Therefore, it can be judged that the correlation degree of product b to product a is one-half.

[0094] Compared with the existing data mining method based on dynamic graph neural network, the present invention improves the accuracy of product recommendation.

[0095] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A data mining method based on dynamic graph neural network, characterized in that: include: Step S1, obtaining the purchase information of each user on the platform; The purchase information includes the time information when the product was purchased and the corresponding product information, as well as the browsing information when the product was purchased; Step S2, data mining is performed on the user's browsing information to determine the nature of the user's behavior when purchasing each product; the behavior nature includes careful consideration and quick decision; Step S3: Perform historical matching on carefully considered products to determine whether there are any products that meet the repetitive characteristics among the carefully considered products, and analyze the product information of the products to determine the factors that influence purchase; Step S4: Based on the time information, the products purchased at the same time are grouped to obtain product group data with a time sequence, and the products are matched and analyzed according to the product group data to determine the correlation between the products and the purchase cycle; Step S5: Read the user's search information. If the user's search information is not read, determine the recommended products for the current day based on the product's purchase cycle and purchase information and make recommendations. In step S6, if the user's search information is read, product recommendations are made to the user based on the search information and purchase influencing factors, the user's additional purchase information is obtained, and the additional purchase products are matched with the correlation degree based on the additional purchase information, and the products related to the additional purchase products are determined, and the related products are recommended.

2. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: Step S2 is specifically as follows: Obtain the purchase time data of each product, match the browsing information according to the purchase time data, determine whether there is browsing information in the corresponding time period, and match the corresponding browsing information with the corresponding product information; If there is corresponding browsing information before purchasing the product, the nature of the behavior of purchasing the product is determined to be a careful consideration type; otherwise, the nature of the behavior of purchasing the product is determined to be a quick decision type.

3. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: Step S2 further includes: For quick-decision products, product information matching is performed to determine whether the same product exists in historical purchase information and whether the behavior of the product in the historical purchase information is carefully considered; If it is determined that the same product exists in the historical purchase information and the nature of the purchase behavior is carefully considered, the behavior nature of the product is re-marked and marked as carefully considered.

4. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: The repetitive feature is that there are multiple purchase behaviors in the historical purchase process.

5. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: Step S3 is specifically as follows: Match product information for products with a carefully considered behavior, determine all purchase information of the user in the user's historical purchase history for the same product, and obtain purchase information for the same product; Determine the proportion of purchase behavior for the same product, and determine the proportion of careful consideration and quick decision-making. If the percentage is greater than the preset benchmark value, it is determined that the user's purchase behavior preference for this product is a carefully considered purchase; Reading the product information of the carefully considered purchase, and performing product evaluation index judgment on the read product information, determining product evaluation index items with the same product evaluation index data in the read multiple product information, obtaining a first common index item, and screening to obtain a first product; Reading browsing information of the first product, comparing product evaluation indicators with the browsing information of the corresponding product, determining that the first product is superior to product evaluation indicators of the products in the corresponding browsing information, and obtaining a first indicator item set; The product evaluation index items of the first index item set are compared to determine the common index items of each purchase in the first index item set, and the common index items are matched with the first common index items, and the same product evaluation index items in the common index items and the first common index items are used as purchase influencing factors for the user's product purchase.

6. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: Step S3 further includes: Based on the purchase influencing factors, a big data analysis is performed on the first product and its corresponding browsing information to determine the data change amount of the corresponding purchase influencing factors.

7. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: Step S4 is specifically as follows: Step S41, matching products between product group data based on time sequence, determining the purchase time interval of the same type of products, and obtaining purchase cycle data of each type of products; Step S42, permuting and combining the quantities of different commodity categories in the commodity group data to obtain combination data corresponding to the commodity group data; Step S43 , performing combination data matching on the commodity group data of different groups based on the commodity group data, determining the overlap rate of various permutations and combinations in the combination data, and determining the association degree between the commodities in the permutations and combinations based on the overlap rate.

8. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: Step S41 further includes: Classify the products based on the product information to obtain the first category of data; Based on the built-in distinguishing items, the products of the same category are divided according to function, appearance, and usage adjustment to obtain the second category of product data.

9. The data mining method based on dynamic graph neural network according to claim 1, characterized in that: In step S43, the product overlap rate is calculated as follows: Selecting a product from the historical purchase record to obtain the original product, and using the product as the subject, matching the product group data to determine the product group data containing the original product, and counting the number of the matched product group data to obtain a first data volume; Based on the matched product group data, the original products are randomly combined to obtain combination data corresponding to the product group data, and the number of the same combination data of each matched product group data is counted to obtain a second data volume; The ratio of the second data volume to the first data volume is calculated to obtain the overlap rate of the corresponding product combination.

Images