User profiling method, device, equipment and storage medium

Abstract

The present disclosure provides a user portrait method, device, equipment and storage medium, the method includes obtaining all commodities purchased by a target user, and based on all the obtained commodities, extracting the attribute information of each commodity according to the pre-constructed mapping relationship data between commodities and attribute information, wherein the attribute information includes attribute name and commodity attribute value; according to the extracted attribute name and commodity attribute value of each commodity, determining the label information of the target user, wherein the label information represents the portrait result of the target user. The attribute information of the commodity in the present disclosure is determined based on the purchase behavior information and comment information of all users who purchase the commodity, and the user portrait result is determined based on the attribute information of the commodity purchased by the target user. Therefore, the present disclosure actually portraits the target user by analyzing the purchase behavior information and comment information of all users on the platform, thereby making the portrait result more accurate.

Description

Technical Field

[0001] The present disclosure relates to the technical field of network data analysis and processing, and in particular to a user profiling method, apparatus, device and storage medium. Background Art

[0002] User portraits, also known as user roles, are an effective tool for outlining target users and linking user demands with design directions. User portraits have been widely used in various fields. In traditional user portrait methods, they are highly dependent on basic information such as age, gender, location, and occupation left by users when registering. However, about 80% of users do not reserve basic information at all, or the reserved basic information is not accurate, which leads to low accuracy of user portrait results. Therefore, how to improve the accuracy of user portraits is a technical problem that needs to be solved urgently by those skilled in the art. Summary of the invention

[0003] In view of this, the present disclosure proposes a user portrait method, apparatus, device and storage medium, which can improve the accuracy of user portraits.

[0004] According to a first aspect of the present disclosure, a user profiling method is provided, comprising:

[0005] Acquire all commodities purchased by the target user, and based on all the acquired commodities, extract attribute information of each of the commodities according to pre-built mapping relationship data between commodities and attribute information, wherein the attribute information includes attribute name and commodity attribute value;

[0006] Determine the tag information of the target user according to the extracted attribute name and the attribute value of each of the commodities, wherein the tag information represents the portrait result of the target user, including the tag name and the tag attribute value;

[0007] The attribute information of the commodity is determined based on the shopping data of all users who purchased the commodity, wherein the shopping data includes at least one of the user's purchasing behavior information and comment information, and the purchasing behavior information includes at least one of the purchased commodity name, commodity price, commodity category and purchase time.

[0008] In a possible implementation, when determining the attribute information of the commodity according to the shopping data of all users who purchased the commodity, it includes:

[0009] For each user who purchases the product, the shopping data of the user is obtained, and the initial tag information of the user is determined according to the shopping data of the user;

[0010] The attribute information of the commodity is determined according to the initial tag information of all users who purchased the commodity.

[0011] In a possible implementation, when determining the initial tag information of the user according to the shopping data of the user, the method includes:

[0012] Extracting the purchase behavior information from each of the shopping data, and labeling each of the purchase behavior information based on a pre-constructed first label generation rule to obtain a label corresponding to each of the purchase behavior information;

[0013] Extracting the review information from each of the shopping data, and labeling each of the review information based on a pre-built machine learning model for determining product review information labels, to obtain a label corresponding to each of the review information;

[0014] The tags corresponding to all the purchased behavior information and the tags corresponding to all the comment information are summarized and analyzed to obtain the initial tag information of the user.

[0015] In a possible implementation, when determining the tag information of the target user according to the extracted attribute name and the commodity attribute value of each commodity, it includes:

[0016] Determine all the tag names of the target user according to the extracted attribute names of the commodities;

[0017] Traverse the tag names;

[0018] For the currently traversed tag name, obtain the commodity attribute value that matches the tag name in each of the commodities, and calculate the tag attribute value corresponding to the tag name based on all the obtained commodity attribute values;

[0019] After the traversal is completed, all tag names of the target user and the tag attribute values ​​corresponding to each tag name can be obtained.

[0020] In a possible implementation, when calculating the tag attribute value of the tag name based on all the obtained commodity attribute values, it includes:

[0021] Filtering out target product attribute values ​​representing the overall distribution of all product attribute values ​​from all product attribute values;

[0022] Based on all the obtained target product attribute values, determine the initial tag attribute value of the tag name;

[0023] The tag attribute value of the tag name is calculated based on the initial tag attribute value of the tag name and a preset standard tag attribute value.

[0024] In a possible implementation, when selecting a target commodity attribute value representing the overall distribution of all commodity attribute values ​​from all commodity attribute values, the following steps are included:

[0025] Based on all commodity attribute values, construct a first commodity attribute value matrix;

[0026] Calculate the skewness value of each row of data in the first commodity attribute value matrix to obtain a first skewness matrix;

[0027] Sorting each row of data in the first skewness matrix to obtain a second skewness matrix;

[0028] A set number of quantiles are screened out from the second skewness matrix, and commodity attribute values ​​corresponding to the respective quantiles are extracted as target commodity attribute values.

[0029] In a possible implementation, when determining the initial tag attribute value of the tag name based on all the obtained target product attribute values, it includes:

[0030] Based on all the obtained target product attribute values, construct a second product attribute value matrix;

[0031] Aggregating the second product label attribute value matrix to obtain a factor matrix;

[0032] The factor matrix is ​​used as the initial label attribute value of the label name.

[0033] According to a second aspect of the present disclosure, a user portrait device is provided, comprising:

[0034] A commodity attribute information acquisition module is used to acquire all commodities purchased by the target user, and based on all acquired commodities, extract attribute information of each of the commodities according to pre-built mapping relationship data between commodities and attribute information, wherein the attribute information includes attribute name and commodity attribute value;

[0035] A user tag information generating module, used to determine the tag information of the target user according to the extracted attribute name and the commodity attribute value of each commodity, wherein the tag information represents the portrait result of the target user, including the tag name and the tag attribute value;

[0036] The attribute information of the commodity is determined based on the shopping data of all users who purchased the commodity, wherein the shopping data includes at least one of the user's purchasing behavior information and comment information, and the purchasing behavior information includes at least one of the purchased commodity name, commodity price, commodity category and purchase time.

[0037] According to a third aspect of the present disclosure, a user portrait device is provided, comprising: a processor; and a memory for storing processor executable instructions; wherein the processor is configured to execute the method described in the first aspect of the present disclosure.

[0038] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions, when executed by a processor, implement the method described in the first aspect of the present disclosure.

[0039] In the present disclosure, a user portrait method is provided, which includes obtaining all commodities purchased by a target user, and based on all the obtained commodities, extracting the attribute information of each commodity according to the pre-constructed mapping relationship data between the commodity and the attribute information, wherein the attribute information includes the attribute name and the commodity attribute value; determining the label information of the target user according to the extracted attribute name and the commodity attribute value of each commodity, wherein the label information represents the portrait result of the target user, including the label name and the label attribute value; the attribute information of the commodity is determined according to the shopping data of all users who purchase the commodity, wherein the shopping data includes at least one of the user's purchase behavior information and comment information, and the purchase behavior information includes at least one of the purchased commodity name, commodity price, commodity category and shopping time. It can be seen from the above steps that in the present disclosure, the attribute information of the commodity is actually determined according to the purchase behavior information and comment information of all users who purchase each commodity, and then the target user is profiled according to the commodity attribute information of the commodity purchased by the target user, that is, the present disclosure actually profiles the target user by analyzing the purchase behavior information and comment information of all users on the platform, so the result of the portrait can be made more accurate.

[0040] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0042] Figure 1 A flowchart showing a method for determining attribute information of a commodity based on shopping data of all users who purchased the commodity according to an embodiment of the present disclosure;

[0043] Figure 2 A flowchart showing a method for determining user initial tag information based on all shopping data of the user according to an embodiment of the present disclosure is shown;

[0044] Figure 3A flowchart showing a method for determining attribute information of a commodity based on initial tag information of all users who purchased the commodity according to an embodiment of the present disclosure;

[0045] Figure 4 A flowchart showing a user portrait method according to an embodiment of the present disclosure is shown;

[0046] Figure 5 A schematic block diagram of a user portrait device according to an embodiment of the present disclosure is shown.

[0047] Figure 6 A schematic block diagram of a user portrait device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0048] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0049] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0050] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0051] <Method Example>

[0052] First of all, it should be noted that before implementing the user portrait method disclosed in the present invention, it is necessary to first construct the attribute information of each commodity for all commodities sold on the shopping platform, and establish the mapping relationship data between the commodity and the attribute information. Among them, the attribute information of the commodity includes the attribute name and the commodity attribute value corresponding to the attribute name. The attribute name is used to characterize the characteristics of the user who purchased the commodity, and the commodity attribute value is used to characterize the probability that the user who purchased the commodity has the characteristic. For example, if a user purchases the commodity of baby milk powder, then there is an 80% probability that the user has a child, then the attribute name of the commodity of baby milk powder can be configured as child, which is used to characterize that the user who purchased the baby milk powder has a child. At the same time, the commodity attribute value corresponding to the attribute name of child is set to [0.8, 0.2], which is used to characterize that there is an 80% probability that the user who purchased the baby milk powder has a child, and a 20% probability that the user has no child.

[0053] The attribute information of each product on the shopping platform may be pre-set based on experience, or may be determined based on the shopping data of all users who purchased the product, which is not specifically limited here. The user's shopping data may include at least one of the user's purchase behavior information and comment information, and the purchase behavior information may include at least one of the purchased product name, product price, product category, and purchase time.

[0054] In a possible implementation, when determining the attribute information of a commodity based on the shopping data of all users who purchased the commodity, the following can be done: Figure 1 As shown, steps S1010-S1020 are included.

[0055] S1010, for each user who purchases a product, obtain the user's shopping data, and determine the user's initial label information based on the user's shopping data.

[0056] It should be noted here that if a user wants to shop on the shopping platform, he needs to register first. After successful registration, the shopping platform will generate a unique identity (such as an identity ID) for the user, and after each shopping, the mapping relationship between the user's identity and his / her shopping data (including the name of the purchased product, product price, product category, purchase time and other purchase behavior information and comment information) will be recorded and stored in the first mapping relationship table. That is, each piece of data in the first mapping relationship table records the purchase behavior information and comment information such as the product name, product price, product category and purchase time associated with the user's identity.

[0057] When a product is determined, all the identities associated with the product name can be extracted by querying the first mapping table, and the users with these identities are all the users who purchased the product. Further, each user who purchased the product is traversed; for the currently traversed user, all the shopping data associated with his identity is extracted by querying the first mapping table, and used as all the shopping data of the user; after the traversal is completed, all the users who purchased the product and all the shopping data of each user can be obtained.

[0058] Furthermore, for each user who purchases a product, the initial label information is determined based on all the user's shopping data. The following is an example of a user to describe the process of determining the initial label information based on all the user's shopping data. Specifically, Figure 2 As shown, steps S1011-S1013 are included.

[0059] S1011, extracting purchase behavior information from each shopping data, and labeling each purchase behavior information based on a pre-constructed first label generation rule to obtain a label corresponding to each purchase behavior information. The first label generation rule is pre-configured with labels corresponding to specific product names, product prices, product categories, purchase times, and at least one combination of the above.

[0060] Before executing step S1011, it is necessary to construct a first tag generation rule. In a possible implementation, the first tag generation rule is constructed based on a preconfigured tag library, in which tags representing various characteristics of the user are set. For example, the tag library may be provided with tags representing basic user information, such as gender, age, marital status, whether there are children, and place of residence; tags representing user interest preferences, such as love of sports, love of music, or love of art; tags representing user consumption characteristics, such as consumption amount or consumption frequency; tags representing user behavior characteristics, such as active time, active scene, etc. When constructing the first tag generation rule based on the tag library, the specific product name, product price, product category, purchase time and at least one combination of the above information can be determined first, and then the tags matching the specific product name, product price, product category, shopping time and at least one combination of the above information are selected from the tag library, thereby completing the construction of the first tag generation rule. For example, if a user has purchased baby formula within two years, then the user is likely to be married (marriage) and have a child (child). Therefore, the label [marriage, child] corresponding to the product name of baby formula and the purchase time within two years can be established in the first label generation rule. For another example, if a user has purchased lipstick, then the user is likely to be a female (gender_female). Therefore, the label [gender_female] corresponding to the product name of lipstick can be established in the first label generation rule.

[0061] After constructing the first label generation rule, you can label each purchase behavior information based on the first label generation rule to obtain the label corresponding to each purchase behavior information. Specifically, traverse all purchase behavior information, extract the product name, product price, product category and purchase time in the purchase behavior information for the current traversed purchase behavior information, determine the label that matches the product name, product price, product category, purchase time, and at least one of the above information combinations by querying the first label generation rule, aggregate and remove duplicates from all labels, and use the aggregated and deduplicated labels as the labels corresponding to the purchase behavior information. After the traversal is completed, you can get the labels corresponding to all purchase behavior information.

[0062] For example, when the label [marrriage, child] corresponding to the combination of the product name of baby formula and the purchase time within two years is established in the first label generation rule, when the product name in the traversed purchase behavior information is baby formula and the purchase time is within two years, the purchase behavior will be labeled [marrriage, child] based on the first label generation rule.

[0063] S1012, extracting the review information from each shopping data, and labeling each review information based on a pre-built machine learning model for determining the review information label, to obtain a label corresponding to each review information.

[0064] Before executing step S1012, it is necessary to first construct a machine learning model for determining the comment information label. The specific construction steps include: obtaining a set number of comment information, and labeling each comment information by manual labeling. For example, when the comment information is "This dress is nice, my husband likes it very much", the user is most likely a married woman, so the comment information will be labeled with [gender_female, marriage]. Among them, when labeling the comment information, the labels used are also selected from the label library described above. After manually labeling each comment information, the model parameters of the basic machine learning model are trained with each comment information as input data and the label corresponding to the comment information as output data, so that after the training is completed, the construction of the machine learning model for determining the comment information label is completed.

[0065] After building the machine learning model, you can label each comment based on the machine learning model to get the label corresponding to each comment. Specifically, traverse all the comments, and use the trained machine learning model to label the comments currently traversed. After the traversal is completed, you can get the labels corresponding to all the comments.

[0066] S1013, summarizing and analyzing the labels corresponding to all the purchased behavior information and the labels corresponding to all the comment information, and obtaining the user's initial label information. Specifically, the following steps are included:

[0067] First, the labels corresponding to all the purchased behavior information and the labels corresponding to all the comment information are summarized, and the number of times each label appears is calculated.

[0068] For example, the label corresponding to all purchase behavior information of user A is [gender_female, marriage], and the label corresponding to all comment information is [gender_female, child]. After processing in this step, the result shown in Table 1 will be obtained.

[0069]

[0070] Second, determine the user's initial tag information based on various tags and the number of times each tag appears. Specifically, traverse various tag names, obtain the second tag generation rule that matches the tag for the currently traversed tag, and determine the initial tag name corresponding to the tag and the tag attribute value corresponding to the initial tag name based on the second tag generation rule that matches the tag name. The initial tag name and the tag attribute value corresponding to the initial tag name constitute a piece of initial tag information for the user. After the traversal is completed, all the initial tag information of the user can be obtained.

[0071] It should be noted here that for each tag in the tag library, a second tag generation rule matching it is set, and the second tag generation rule specifies the initial tag name corresponding to the tag, and how to calculate the tag attribute value corresponding to the initial tag name according to the number of occurrences of the tag. In this way, after traversing to the tag, the second tag generation rule matching the tag can be obtained, and the initial tag name corresponding to the tag and the tag attribute value corresponding to the initial tag name can be determined based on the second tag generation rule matching the tag.

[0072] For example, for the label gender_female in Table 1, the second label generation rule that matches it is specified that: the initial label name corresponding to the label gender_female is gender, and the label attribute value of gender is [first attribute value, second attribute value], where the first attribute value represents the probability that the user is a man, and the second attribute value represents the probability that the user is a woman; when gender_female appears once, the second attribute value of gender is 0.8, and the first attribute value is 1-0.8=0.2. When gender_female appears n times, the second attribute value of gender is equal to 0.8 n / (0.8 n +0.2 n ), the difference between the first attribute value of 1 and the second attribute value. According to the second label generation rule, it can be determined that the initial label name corresponding to the label gender_female is gender, and the second attribute value of gender is equal to 0.8 2 / (0.8 2 +0.2 2 )=0.94, the first attribute value is equal to 1-0.94=0.06, that is, the label attribute value corresponding to gender is [0.06,0.94].

[0073] For the label marriage in Table 1, the second label generation rule that matches it is specified that: the initial label name corresponding to the label marriage is marriage, and the label attribute value of marriage is [first attribute value, second attribute value], where the first attribute value is used to represent the probability that the user is married, and the second attribute value represents the probability that the user is unmarried; when marriage appears once, the first attribute value of marriage is 0.8, and the second attribute value is 1-0.8=0.2. When marriage appears n times, the first attribute value of marriage is equal to 0.8 n / (0.8 n +0.2 n ), the second attribute value is the difference between 1 and the first attribute value. According to the second label generation rule, it can be determined that the initial label name corresponding to the label marriage is marriage, the first attribute value of marriage is equal to 0.8, and the second attribute value is equal to 1-0.8=0.2, that is, the label attribute value corresponding to marriage is [0.8,0.2].

[0074] For the label child in Table 1, the second label generation rule that matches it is specified that: the initial label name corresponding to the label child is child, and the label attribute value of child is [first attribute value, second attribute value]. The first attribute value represents the probability that the user has a child, and the second attribute value represents the probability that the user does not have a child. When child appears once, the first attribute value of child is 0.9, and the second attribute value is 1-0.9=0.1. When child appears n times, the first attribute value of child is equal to 0.9 n / (0.9 n +0.1 n ), the second attribute value is the difference between 1 and the first attribute value. According to the second label generation rule, it can be determined that the initial label name corresponding to the label child is child, the first attribute value of child is equal to 0.9, and the second attribute value is equal to 1-0.9=0.1, that is, the label attribute value of child is [0.9,0.1].

[0075] At this point, all the initial label information corresponding to user A is obtained as shown in Table 2.

[0076]

[0077] By referring to the above method, the initial tag information of all users who purchase the product can be obtained. After the initial tag information of all users who purchase the product is obtained, step S1020 can be executed.

[0078] S1020, determining the attribute information of the product based on the initial tag information of all users who purchased the product. Specifically, Figure 3 As shown, steps S1021-S1023 may be included.

[0079] S1021, extracting the initial tag names from the initial tag information of all users, and summarizing and removing duplicates of all the obtained initial tag names to obtain the attribute names of the products.

[0080] For example, all users who purchase product A include user A and user B, and the initial tag information of user A and user B is shown in Table 3. The initial tag name in the initial tag information of user A is extracted to obtain the three initial tag names of gender, marriage, and child. The initial tag name in the initial tag information of user B is extracted to obtain the initial tag name of gender. All the obtained initial tag names are aggregated and deduplicated to obtain the three initial tag names of gender, marriage, and child. The attribute names of product A include the three attribute names of gender, marriage, and child.

[0081]

[0082] S1022: For each attribute name of the product, extract initial label information corresponding to the attribute name from the initial label information of each user who purchases the product.

[0083] For example, the users who purchase product A include 6 persons, A, B, D, E, G, and S. For the attribute name gender of product A, initial label information with label name gender is extracted from the initial label information of the 6 persons, A, B, D, E, G, and S, respectively, as the initial label information corresponding to the attribute name gender. The extracted results can be shown in Table 4.

[0084]

[0085] Referring to the generation process of the initial label information corresponding to the attribute name gender, the initial label information corresponding to the attribute name marriage and the initial label information corresponding to the attribute name child can be obtained.

[0086] S1023, calculate the commodity attribute value corresponding to each attribute name according to the initial label information corresponding to each attribute name, so that the commodity attribute value corresponding to each attribute name of the commodity can be obtained, and each attribute name of the commodity and the corresponding commodity attribute value constitute the attribute information of the commodity. Specifically, extract the label attribute value in each initial label information corresponding to the attribute name, average the first attribute value in all label attribute values, obtain the first attribute value corresponding to the attribute name, average the second attribute value in all label attribute values, obtain the second attribute value corresponding to the attribute name, and obtain the commodity attribute value corresponding to the attribute name according to the first attribute value and the second attribute value corresponding to the attribute name. Referring to the above method, the commodity attribute value corresponding to each attribute name can be obtained, and then the attribute information of the commodity is constituted by each attribute name of the commodity and the corresponding commodity attribute value.

[0087] Continuing with the above embodiment, when calculating the commodity attribute value corresponding to the attribute name gender of commodity A based on Table 4, the first attribute value and the second attribute value corresponding to the attribute name gender can be calculated with reference to Table 5, so that the commodity attribute value corresponding to the attribute name gender of commodity A is [0.15, 0.85].

[0088]

[0089] Referring to the calculation process of the product attribute value corresponding to the attribute name gender, you can get the product attribute value corresponding to the attribute name gender, and the product attribute value corresponding to the attribute name child. Finally, all the attribute names of product A and the product attribute values ​​corresponding to each attribute name are used as the attribute information of product A, that is, the attribute information of product A includes the attribute name gender and the corresponding product attribute value, the attribute name marriage and the corresponding product attribute value, the attribute name child and the corresponding product attribute value.

[0090] By referring to the above method, the attribute information of all commodities on the shopping platform can be obtained. Since the attribute information of each commodity on the platform is determined based on the shopping data of all users who purchase the commodity, the accuracy of the attribute information of each commodity can be improved. Furthermore, as the shopping data of users on the shopping platform is continuously updated, the attribute information of each commodity on the platform will also be updated synchronously, so that the attribute information of the commodity is continuously iterated and updated, and becomes more accurate.

[0091] It should be noted here that if only the product level is summarized, there may be such a possibility that all consumers who purchase the product do not have the initial label information corresponding to the product's attribute information. In this way, it is impossible to calculate the product's attribute information. In order to make all products have attribute information, in a possible implementation, when the product attribute value in the calculated product attribute information is empty, it also includes: calculating the product attribute value corresponding to the small category to which the product belongs, and using the product attribute value corresponding to the small category as the product attribute value in the product attribute information.

[0092] When calculating the product attribute value corresponding to the small category to which the product belongs, it includes: obtaining other products included in the small category to which the product belongs, obtaining the product attribute values ​​corresponding to the other products and the sample size used to calculate the product attribute values, and calculating the product attribute value corresponding to the small category to which the product belongs based on the product attribute values ​​corresponding to the other products and the sample size used to calculate the product attribute values.

[0093] For example, the products under the S2 small category include D, E, F, and G. The first attribute value of the product attribute value corresponding to the attribute name gender of each product and the sample size for calculating the first attribute value are shown in Table 6.

[0094]

[0095] As shown in Table 6, the first attribute value of gender of product D cannot be calculated from the initial label information of all users who purchased product D. Therefore, it is necessary to determine the subcategory S2 to which product D belongs, and further calculate the first attribute value of gender corresponding to subcategory S2 based on the first attribute values ​​of gender of all other products belonging to subcategory S2 and the sample size, and use the first attribute value of gender corresponding to subcategory S2 as the first attribute value of gender of product D. See Table 7 for the specific calculation process.

[0096]

[0097] That is, in the end, the first attribute value of the gender of product D is 0.218. Furthermore, the second attribute value of the gender of product D = 1-0.218 = 0.782, that is, the product attribute value of the gender of product D is [0.15, 0.85].

[0098] Furthermore, if the calculated commodity attribute value corresponding to the small category to which the commodity belongs is still empty, continue to obtain the medium category to which the small category belongs, and calculate the commodity attribute value corresponding to the medium category to which the commodity belongs, and use the commodity attribute value corresponding to the medium category to which the commodity belongs as the commodity attribute value in the commodity attribute information. Specifically, obtain other small categories included in the medium category to which the small category belongs, obtain the commodity attribute values ​​corresponding to the other small categories and the sample size used to calculate the commodity attribute values, and calculate the commodity attribute value corresponding to the medium category to which the commodity belongs based on the commodity attribute values ​​corresponding to the other small categories and the sample size used to calculate the commodity attribute values. For the specific calculation process, refer to the calculation process of the commodity attribute value corresponding to the small category, which will not be repeated here.

[0099] If the calculated product attribute value corresponding to the medium category to which the product belongs is still empty, the product attribute value corresponding to the large category to which the medium category belongs is calculated, and the product attribute value corresponding to the large category to which the product belongs is used as the product attribute value in the product attribute information.

[0100] If the calculated product attribute value corresponding to the major category to which the product belongs is still empty, the product attribute values ​​corresponding to all categories to which the major category belongs are calculated, and the product attribute values ​​corresponding to all categories to which the product belongs are used as the product attribute values ​​in the product attribute information.

[0101] Through the above method, all attribute information of all products on the shopping platform can be calculated.

[0102] It should be further explained that if the sample size used to calculate the commodity attribute value is small, the accuracy of the calculated commodity attribute value is low. Therefore, in a possible implementation method, when the sample size used to calculate the commodity attribute value of the commodity is lower than the preset quantity threshold, it also includes: obtaining the commodities included in the small category to which the commodity belongs, obtaining the commodity attribute values ​​corresponding to all commodities and the sample size used to calculate the commodity attribute values, and calculating the commodity attribute value corresponding to the small category to which the commodity belongs based on the commodity attribute values ​​corresponding to all commodities and the sample size used to calculate the commodity attribute values.

[0103] For example, the commodities under the S1 small category include A, B, and C. Table 8 shows the first attribute value of the commodity attribute value corresponding to the attribute name gender of each commodity and the sample size for calculating the first attribute value.

[0104]

[0105] As shown in Table 8, the sample size used to calculate the first attribute value of gender of commodity A is less than the preset sample size threshold of 10. Therefore, it is necessary to determine the subcategory S1 to which commodity A belongs, and further calculate the first attribute value of gender corresponding to subcategory S1 based on the first attribute values ​​and sample sizes of gender of all commodities belonging to subcategory S1, and use the first attribute value of gender corresponding to subcategory S1 as the first attribute value of gender of commodity A. See Table 9 for the specific calculation process.

[0106]

[0107] That is, in the end, the first attribute value of gender of product A is 0.145. Furthermore, the second attribute value of gender of product A = 1-0.145 = 0.855, that is, the final product attribute value of gender of product A is [0.145, 0.855]. Similarly, the product attribute values ​​of gender of products B and C are finally updated to [0.145, 0.855].

[0108] Furthermore, if the commodity attribute value corresponding to the small category of the commodity used for calculation is still less than the set quantity threshold, the commodity attribute values ​​corresponding to the medium category, large category and annual category are calculated in turn until the sample size used to calculate the commodity attribute value is greater than the set quantity threshold, and the commodity attribute value corresponding to the category finally calculated is used as the commodity attribute value in the commodity attribute information.

[0109] Through the above method, all attribute information of all products on the shopping platform can be revised to make the attribute information of the products more accurate.

[0110] After completing the construction of the product attribute values ​​of each product on the shopping platform, it also includes establishing mapping relationship data between the product and the attribute information. Specifically, mapping relationship data between the product name and the attribute information is established. In this way, user profiling can be performed based on the mapping relationship data later.

[0111] After establishing the mapping relationship data between each product on the shopping platform and its attribute information, the user profiling method disclosed in the present invention can be executed.

[0112] Figure 4 A flow chart of a user portrait method according to an embodiment of the present disclosure is shown. Figure 4 As shown, the method includes steps S1100-S1200.

[0113] S1100, obtain all commodities purchased by the target user, and based on all the obtained commodities, extract the attribute information of each commodity according to the pre-built mapping relationship data between commodities and attributes. Specifically, after determining the target user, according to the identity identifier of the target user, all shopping data associated with the identity identifier are found from the first mapping relationship table, and the commodity names of all commodities purchased by the target user are extracted from all shopping data. Furthermore, based on the commodity name of each commodity, the attribute information associated with the commodity name can be extracted from the mapping relationship data between the commodity and the attribute information, and finally the attribute information of all commodities purchased by the target user is obtained, wherein the attribute information of the commodity includes the attribute name and the commodity attribute value corresponding to the attribute name.

[0114] S1200: Determine tag information of the target user based on the extracted attribute names and attribute values ​​of each product, wherein the tag information represents a portrait result of the target user, including a tag name and a tag attribute value.

[0115] In a possible implementation, when determining the tag information of the target user according to the extracted attribute names and commodity attribute values ​​of each commodity, the following steps may be included.

[0116] First, all the tag names of the target user are determined based on the extracted attribute names of each product. Specifically, the attribute names of each product are aggregated and deduplicated, and the aggregated and deduplicated attribute names are used as all the tag names of the user.

[0117] For example, the target user is user A, and user A purchased product A and product B. The attribute names of product A include gender, marriage, and child, and the attribute name of product B includes gender. After removing duplicate attribute names of product A and product B, the attribute names gender, marriage, and child are obtained. Then, all label names of user A are gender, marriage, and child.

[0118] Second, traverse the tag names, and for the currently traversed tag name, obtain the product attribute value that matches the tag name in each product, and based on all the obtained product attribute values, calculate the tag attribute value corresponding to the tag name. After the traversal is completed, all the tag names of the target user and the tag attribute values ​​corresponding to each tag name can be obtained.

[0119] For example, if the tag name currently traversed is gender, then the product attribute values ​​of all products purchased by user A with the attribute name of gender are obtained as the product attribute values ​​matching the tag name of gender.

[0120] In a possible implementation, when calculating the tag attribute value corresponding to the tag name based on all the obtained commodity attribute values, the following steps may be included.

[0121] First, select the target product attribute value that represents the overall distribution of all product attribute values ​​from all product attribute values. Specifically:

[0122] Step 1: First, construct a first commodity attribute value matrix based on all commodity attribute values. For example, the first commodity attribute value matrix A constructed by all commodity attribute values ​​is as follows:

[0123]

[0124] in, is the first attribute value of the nth product attribute value, is the second attribute value of the nth product attribute value, and n is the number of all product attribute values.

[0125] Step 2, calculate the skewness value of each row of data in the first commodity attribute value matrix to obtain a first skewness matrix. Specifically, the calculation formula of the first skewness matrix is ​​as follows:

[0126]

[0127] in, (j) is the standard distribution of product attribute values ​​(i.e. the actual distribution of current product attribute values, which can be roughly considered to obey the normal distribution), (j) is specifically equal to the standard deviation of all attribute values ​​in column j. For example, when j=1, (1) is equal to to The standard deviation of all values ​​of . When j=2, (2) is equal to to The standard deviation of all values ​​in .

[0128] Step 3, sorting each row of data in the first skewness matrix to obtain a second skewness matrix. Specifically, each row of elements in the first skewness matrix can be sorted from small to large to obtain the second skewness matrix.

[0129] Step 4, select a set number of quantiles from the second skewness matrix, and extract the commodity attribute values ​​corresponding to each quantile as the target commodity attribute values. Specifically, 5 quintiles can be selected from the second skewness matrix, and the commodity attribute values ​​corresponding to each quintile are extracted as the target commodity attribute values. It should be noted here that if 5 quintiles cannot be extracted from the second skewness matrix, the commodity attribute values ​​corresponding to all elements in the second skewness matrix are used as the target commodity attribute values.

[0130] Secondly, based on all the obtained target product attribute values, determine the initial tag attribute value of the tag name. Specifically, the following steps may be included:

[0131] Step 1: construct a second commodity attribute value matrix based on all obtained target commodity attribute values. Specifically, the construction process of the second commodity attribute value matrix B refers to the construction process of the first commodity attribute value matrix A, and will not be repeated here.

[0132] Step 2: Aggregate the second product label attribute value matrix to obtain a factor matrix. Specifically, how to calculate the factor matrix of the second product label attribute value matrix is ​​common knowledge in the art and will not be described in detail here.

[0133] Step 3: Use the factor matrix as the initial label attribute value of the label name. Specifically, the calculated factor matrix is ​​a matrix with one row and two columns. The first element value of the first row is used as the first attribute value in the initial label attribute value of the current label name, and the first element value of the second row is used as the second attribute value in the initial label attribute value of the current label name. For example, if the factor matrix is ​​[0.14, 0.86], the initial label attribute value of the current label name is [0.14, 0.86].

[0134] Next, based on the initial label attribute value of the label name and the preset standard label attribute value, the final label attribute value of the label name is calculated. Among them, the system sets a standard label attribute value for each label name, and the standard label attribute value is set based on the actual distribution of the label attribute value corresponding to the label name. For example, the current label name is gender, and the actual gender ratio of men and women in the current population is 0.48:0.52, then the standard label attribute value of gender can be set to [0.48,0.52]. When the initial label attribute value of gender is calculated to be [0.14,0.86], the corresponding data in the initial label attribute value and the standard label attribute value can be multiplied, and the multiplication result can be normalized to obtain the final label attribute value of the current label name.

[0135] By referring to the calculation process of the tag attribute value of the current tag name, the tag attribute values ​​corresponding to all the tag names of the user can be obtained, and then all the tag names of the user and the tag attribute values ​​corresponding to each tag name are obtained as the tag information of the target user.

[0136] The present disclosure provides a user portrait method, including obtaining all commodities purchased by a target user, and based on all the obtained commodities, extracting the attribute information of each commodity according to the pre-constructed mapping relationship data between the commodity and the attribute information, wherein the attribute information includes the attribute name and the commodity attribute value; determining the label information of the target user according to the extracted attribute name and the commodity attribute value of each commodity, wherein the label information represents the portrait result of the target user, including the label name and the label attribute value; the attribute information of the commodity is determined according to the shopping data of all users who purchase the commodity, wherein the shopping data includes at least one of the user's purchase behavior information and comment information, and the purchase behavior information includes at least one of the purchased commodity name, commodity price, commodity category and purchase time. It can be seen from the above steps that in the present disclosure, the attribute information of the commodity is actually determined according to the purchase behavior information and comment information of all users who purchase each commodity, and then the target user is profiled according to the commodity attribute information of the commodities purchased by the target user, that is, the present disclosure actually profiles the target user by analyzing the purchase behavior information and comment information of all users on the platform, thus making the portrait result more accurate.

[0137] <Device Example>

[0138] Figure 5 FIG. 1 is a schematic block diagram of a user portrait device according to an embodiment of the present disclosure. Figure 5 As shown, the user portrait device 100 includes:

[0139] The commodity attribute information acquisition module 110 is used to acquire all commodities purchased by the target user, and based on all the acquired commodities, extract the attribute information of each of the commodities according to the pre-built mapping relationship data between commodities and attribute information, wherein the attribute information includes the attribute name and the commodity attribute value;

[0140] A user tag information generating module 120 is used to determine the tag information of the target user according to the extracted attribute name and the commodity attribute value of each commodity, wherein the tag information represents the portrait result of the target user, including the tag name and the tag attribute value;

[0141] The attribute information of the commodity is determined based on the shopping data of all users who purchased the commodity, wherein the shopping data includes at least one of the user's purchasing behavior information and comment information, and the purchasing behavior information includes at least one of the purchased commodity name, commodity price, commodity category and purchase time.

[0142] <Equipment Embodiment>

[0143] Figure 6FIG. 1 is a schematic block diagram of a user portrait device according to an embodiment of the present disclosure. Figure 6 As shown, the user portrait device 200 includes: a processor 210 and a memory 220 for storing executable instructions of the processor 210. The processor 210 is configured to implement any of the above-mentioned user portrait methods when executing the executable instructions.

[0144] Here, it should be noted that the number of processors 210 may be one or more. At the same time, in the user portrait device 200 of the embodiment of the present disclosure, an input device 230 and an output device 240 may also be included. Among them, the processor 210, the memory 220, the input device 230 and the output device 240 may be connected through a bus or in other ways, which are not specifically limited here.

[0145] The memory 220 is a computer-readable storage medium that can be used to store software programs, computer executable programs, and various modules, such as the program or module corresponding to the user portrait method of the embodiment of the present disclosure. The processor 210 executes various functional applications and data processing of the user portrait device 200 by running the software programs or modules stored in the memory 220.

[0146] The input device 230 may be used to receive input numbers or signals. The signals may be key signals related to user settings and function control of the device / terminal / server. The output device 240 may include display devices such as display screens.

[0147] <Storage Medium Embodiment>

[0148] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is also provided, on which computer program instructions are stored. When the computer program instructions are executed by the processor 210, any of the user portrait methods described above is implemented.

[0149] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A user profiling method, characterized in that: include: Acquire all commodities purchased by the target user, and based on all the acquired commodities, extract attribute information of each of the commodities according to pre-built mapping relationship data between commodities and attribute information, wherein the attribute information includes an attribute name and a commodity attribute value, the attribute name is used to characterize the characteristics of the user who purchased the commodity, and the commodity attribute value is used to characterize the probability that the user who purchased the commodity has the characteristics of the user; Determine the tag information of the target user according to the extracted attribute names and the attribute values ​​of the commodities, wherein the tag information represents the portrait result of the target user, including the tag name and the tag attribute value. The specific steps of determining the tag information of the target user include: determining all the tag names of the target user according to the extracted attribute names of the commodities; traversing the tag names; obtaining the commodity attribute values ​​matching the tag names in the commodities for the currently traversed tag names, and calculating the tag attribute values ​​corresponding to the tag names based on all the obtained commodity attribute values; when the traversal is completed, all the tag names of the target user and the tag attribute values ​​corresponding to the tag names are obtained; The attribute information of the commodity is determined based on the shopping data of all users who purchased the commodity, wherein the shopping data includes at least one of the user's purchase behavior information and comment information, and the purchase behavior information includes at least one of the commodity name, commodity price, and commodity category; When determining the attribute information of the commodity according to the shopping data of all users who purchased the commodity, it includes: For each user who purchases the product, the shopping data of the user is obtained, and the initial tag information of the user is determined according to the shopping data of the user; Determining attribute information of the commodity according to initial label information of all users who purchased the commodity; When determining the initial tag information of the user according to the shopping data of the user, it includes: Extracting the purchase behavior information from each of the shopping data, and labeling each of the purchase behavior information based on a pre-constructed first label generation rule to obtain a label corresponding to each of the purchase behavior information; Extracting the review information from each shopping data, and labeling each review information based on a pre-built machine learning model for determining a review information label, to obtain a label corresponding to each review information; Summarizing and analyzing the labels corresponding to all the purchased behavior information and the labels corresponding to all the comment information to obtain the initial label information of the user, wherein the initial label information includes an initial label name and a label attribute value corresponding to the initial label name, and the label attribute value corresponding to the initial label name is composed of a first attribute value and a second attribute value, and the probability that the user belongs to the corresponding initial label name is represented by the first attribute value and the second attribute value; Determining the attribute information of the commodity according to the initial tag information of all users who purchased the commodity includes: Extract the initial tag names from the initial tag information of all users, and aggregate and remove duplicates of all the obtained initial tag names to obtain the attribute names of the products; For each attribute name of the product, extract the initial label information corresponding to the attribute name from the initial label information of each user who purchases the product; According to the initial label information corresponding to each attribute name, the product attribute value corresponding to each attribute name is calculated: the label attribute value corresponding to each initial label name is extracted from the initial label information corresponding to each attribute name, and based on the extracted label attribute value corresponding to each initial label name, the product attribute value corresponding to each attribute name is calculated.

2. The method according to claim 1, characterized in that When calculating the tag attribute value corresponding to the tag name based on all the obtained commodity attribute values, it includes: Filtering out target product attribute values ​​representing the overall distribution of all product attribute values ​​from all product attribute values; Based on all the obtained target product attribute values, determine the initial tag attribute value of the tag name; Based on the initial tag attribute value of the tag name and a preset standard tag attribute value, a tag attribute value corresponding to the tag name is calculated.

3. The method according to claim 2, characterized in that When selecting a target product attribute value representing the overall distribution of all product attribute values ​​from all product attribute values, it includes: Based on all commodity attribute values, construct a first commodity attribute value matrix; Calculate the skewness value of each row of data in the first commodity attribute value matrix to obtain a first skewness matrix; Sorting each row of data in the first skewness matrix to obtain a second skewness matrix; A set number of quantiles are screened out from the second skewness matrix, and commodity attribute values ​​corresponding to the respective quantiles are extracted as target commodity attribute values.

4. The method according to claim 2, characterized in that: When determining the initial tag attribute value of the tag name based on all the obtained target product attribute values, it includes: Based on all the obtained target product attribute values, construct a second product attribute value matrix; Aggregating the second commodity attribute value matrix to obtain a factor matrix; The factor matrix is ​​used as the initial label attribute value of the label name.

5. A user portrait device, characterized in that: include: A commodity attribute information acquisition module is used to acquire all commodities purchased by the target user, and based on all the acquired commodities, extract the attribute information of each of the commodities according to the pre-built mapping relationship data between commodities and attribute information, wherein the attribute information includes an attribute name and a commodity attribute value, the attribute name is used to characterize the characteristics of the user who purchased the commodity, and the commodity attribute value is used to characterize the probability that the user who purchased the commodity has the characteristics of the user; The user tag information generation module is used to determine the tag information of the target user according to the extracted attribute names and the commodity attribute values ​​of each of the commodities, wherein the tag information represents the portrait result of the target user, including the tag name and the tag attribute value, and the specific steps of determining the tag information of the target user include: determining all the tag names of the target user according to the extracted attribute names of each of the commodities; traversing the tag names; for the currently traversed tag name, obtaining the commodity attribute value of each of the commodities that matches the tag name, and calculating the tag attribute value corresponding to the tag name based on all the obtained commodity attribute values; when the traversal is completed, all the tag names of the target user and the tag attribute values ​​corresponding to each of the tag names are obtained; The attribute information of the commodity is determined based on the shopping data of all users who purchased the commodity, wherein the shopping data includes at least one of the user's purchase behavior information and comment information, and the purchase behavior information includes at least one of the commodity name, commodity price, and commodity category; When determining the attribute information of the commodity according to the shopping data of all users who purchased the commodity, it includes: For each user who purchases the product, the shopping data of the user is obtained, and the initial tag information of the user is determined according to the shopping data of the user; Determining attribute information of the commodity according to initial label information of all users who purchased the commodity; When determining the initial tag information of the user according to the shopping data of the user, it includes: Extracting the purchase behavior information from each of the shopping data, and labeling each of the purchase behavior information based on a pre-constructed first label generation rule to obtain a label corresponding to each of the purchase behavior information; Extracting the review information from each shopping data, and labeling each review information based on a pre-built machine learning model for determining a review information label, to obtain a label corresponding to each review information; Summarizing and analyzing the labels corresponding to all the purchased behavior information and the labels corresponding to all the comment information to obtain the initial label information of the user, wherein the initial label information includes an initial label name and a label attribute value corresponding to the initial label name, and the label attribute value corresponding to the initial label name is composed of a first attribute value and a second attribute value, and the probability that the user belongs to the corresponding initial label name is represented by the first attribute value and the second attribute value; Determining the attribute information of the commodity according to the initial tag information of all users who purchased the commodity includes: Extract the initial tag names from the initial tag information of all users, and aggregate and remove duplicates of all the obtained initial tag names to obtain the attribute names of the products; For each attribute name of the product, extract the initial label information corresponding to the attribute name from the initial label information of each user who purchases the product; According to the initial label information corresponding to each attribute name, the product attribute value corresponding to each attribute name is calculated: the label attribute value corresponding to each initial label name is extracted from the initial label information corresponding to each attribute name, and based on the extracted label attribute value corresponding to each initial label name, the product attribute value corresponding to each attribute name is calculated.

6. A user profiling device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the method described in any one of claims 1 to 4 when executing the executable instructions.

7. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images