HK1166412B

HK1166412B - Method and apparatus for determining a linked list of candidate products

Info

Publication number: HK1166412B
Application number: HK12107086.6A
Authority: HK
Inventors: 张伟
Original assignee: 阿里巴巴集团控股有限公司
Filing date: 2012-07-19
Publication date: 2016-06-30

Description

Method for determining candidate product linked list and related device

Technical Field

The present application relates to the field of computer network technologies, and in particular, to a method for determining a linked list of candidate products, a method for determining similarity between upper values of nominal attributes of two products, an apparatus for determining a linked list of candidate products, and a system for providing a linked list of candidate products.

Background

As a new shopping mode, online shopping has the advantages of complete types, convenience, rapidness, time and labor conservation and low price compared with the traditional physical shopping mode, and becomes a common shopping mode.

The online store operator uploads the information of each product sold by the online store operator on an electronic commerce website in advance, and the product information comprises product identification, pictures and attribute values of various attributes of the product. According to the value characteristics of the attribute values, the product attributes can be divided into the following two categories: a nominal class attribute and a non-nominal class attribute. The non-nominal class attributes include numerical type attributes, ordinal type attributes, aggregate class attributes, and the like.

The nominal attribute is characterized in that the attribute value is an unordered character string, for example, a product brand belongs to the nominal attribute, the value of the attribute value is the unordered character string, and the value range of the product brand attribute is elegance, magnolia oil, violet, bie-quan, lancome, and the like, taking cosmetics as an example. And the attribute value of the non-nominal class attribute is a natural number or may correspond to a numeric ordered string of natural numbers. For example, the product price attribute belongs to a numerical attribute, the value range of the attribute value is a real number greater than zero, and taking a shoe with a certain style as an example, the price of the shoe is 59.99 yuan. If the product sales attribute belongs to the ordinal type attribute, the value range of the attribute value is a natural number, or the attribute value can be mapped to other forms of natural numbers, such as "high", "medium", "low", and other numeric sequential character strings, and taking a certain style of shoe as an example, the sales of the shoe is 100 pairs. The product color belongs to a set type attribute, and the value range of the attribute value is a set formed by one or more elements in a preset enumeration set, such as the color value of a shoe is { purple, red and yellow }.

When a user purchases an online product, a common process is as follows: the method includes logging in an e-commerce website through a browser client, obtaining information of various products through a search function provided by the e-commerce website, a recommended product list or a sales product list of an online store operator collected by the user before, and the like, selecting one product based on the obtained information, and issuing a product order after confirming purchase.

In the above process, the user selects a product and confirms the purchase according to various product information, which is a key step. In order to provide more information about related products for users, so that the users can more easily compare with three products, the e-commerce website generally provides the users with information about some other candidate products similar or similar to the selected product after the users select one product.

In the prior art, a basic principle of providing a candidate product chain table similar or similar to a specified product to a user is shown in fig. 1, assuming that a product selected by the user is a product t, the specific steps are as follows:

step 101, acquiring attribute values of each attribute of each product from a product attribute information database, wherein the attribute values comprise the attribute values of each attribute of a product t and other products;

102, calculating similarity values of the products t and each other product one by one according to the obtained attribute values of the products;

taking the product c as an example, calculating Similarity values (t, c) of the product c and the product t according to the obtained values of the attributes of the product c and the corresponding attributes of the product t,

wherein i is an attribute identifier, and assuming that each product has n attributes, the value of i is 1 to n; t is t_iIs the value of the ith attribute, c, of the product t_iIs the value of the ith attribute of the product c; w is a_iThe weight value of the ith attribute; sim_iThe similarity value of the ith attribute for product t and product c.

103, selecting Similarity value exceeding predetermined threshold d based on Similarity value (t, c) of each product and product t selected by user calculated in step 102_sA similar product set C of product composition products t_A；

104, collecting similar products C_AThe products are arranged according to the sequence of similarity values with the products t from top to bottom, and a linked list of preset N products which are ranked at the top is selected as a similar product linked list of the products t;

and step 105, providing the related information of each product in the similar product chain table determined in the step 104, such as product identification, picture, brief introduction, evaluation and the like, to the user.

In addition, before the user selects the product, the identifier of each product and the similar product linked list corresponding to the product can be stored in advance, so that after the user selects the product, the similar product linked list of the product is inquired according to the identifier of the selected product and provided for the user.

In the above step 102, for different types of attributes of the product, the prior art adopts the following scheme to calculate the similarity value Sim of the attribute_i：

1. If the attribute i is a numerical attribute, the similarity value of the product t and the product c on the attribute i is as follows:

wherein D (,) is a distance measure, D (t)_i，c_i)＝|t_i-c_iL, |; min D is the minimum value of the distance measurement values of all the products on the attribute i in pairs;

2. if the attribute i is an ordinal type attribute, the similarity value of the product t and the product c on the attribute i is as follows:

wherein n is the value upper limit of the ordinal number;

3. if the attribute i is an aggregate type attribute, the similarity value of the product t and the product c on the attribute i is as follows:

4. if the attribute i is a nominal type attribute, the similarity value of the product t and the product c on the attribute i is as follows:

wherein:n is the total number of products;

wherein f (t)_i) And f (c)_i) Respectively represent t_iAnd c_iThe number of occurrences in the attribute value of the attribute of a product of the homogeneous product set.

s (-) with attribute value t_iAnd attribute value c_iThe occurrence times of the two products in the value of the nominal attribute are related, and if the occurrence times of the two products are close to each other, for example, the two products have more occurrence times or have fewer occurrence times, the similarity values of the two products are higher; otherwise, the similarity value of the two is lower.

The log (N/f (-) function) is used to measure the specificity, or distinctive feature, of an attribute value, when the attribute value t is_iWhen the frequency of the attribute value of the nominal attribute of each product is low, the function value is large; otherwise, when the frequency of occurrence is high, the function value is small.

t_iAnd c_iSimilarity value Sim between_i(t_i，c_i) And s: (Has similar meaning, but the value range of s (-) is normalized for calculating the similarity value between two products, even if the value range is adjusted to [0, 1 ]]。

At present, the nominal class attribute of a product accounts for a large proportion of all attributes of the product, such as a product brand attribute, a product name attribute, and the like, and many nominal class attributes, such as the product brand attribute, are important considerations when a user selects a product, so when calculating the Similarity value simiarity (t, c) between two products, the weight value of the nominal class attribute, that is, the importance of the nominal class attribute, tends to be very high, whereas the prior art is substantially implemented based on a string matching technique when calculating the Similarity value of the nominal class attribute value between two products, that is, when the string of attribute values of the nominal class attribute of two products is the same, the Similarity value is 1; otherwise, calculating the similarity value according to the statistical distribution of the character strings; the processing scheme in the prior art cannot deeply find the semantic meaning of the attribute value, and cannot well calculate the similarity value of the important attribute, namely the nominal class attribute, so that the candidate product of the product selected by the user cannot be accurately provided for the user.

Disclosure of Invention

The embodiment of the application provides a method for determining a candidate product linked list, which is used for solving the problem that the prior art cannot accurately determine the similarity value between products, so that the candidate product linked list cannot be accurately provided.

Correspondingly, the embodiment of the application also provides a system for providing the candidate product linked list and a device for determining the similarity value on the product nominal class attribute.

The technical scheme provided by the embodiment of the application is as follows:

a method of determining a linked list of candidate products, comprising: providing a collection of homogeneous products comprising a first product and a plurality of second products, performing for each second product in the collection of homogeneous products: calculating a similarity value of the values of the first product and the second product over each non-nominal class attribute; when calculating the similarity value of the first product and the second product on each nominal class attribute, executing: when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product; calculating the product similarity values of the first product and the second product according to the similarity values of the non-nominal class attributes of the first product and the second product and the similarity values of the nominal class attributes; and taking at least one second product which is ranked at the top with the similarity value of the first product as a candidate product linked list of the first product.

A method of determining a similarity value of values on a nominal class attribute between two products, comprising: when calculating the similarity value of the values of the first product and the second product on the nominal class attribute, executing: and when the value of the nominal class attribute of the first product is different from the value of the nominal class attribute of the second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in the purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product.

An apparatus for determining a linked list of candidate products, comprising: a providing unit for providing a homogeneous product set comprising a first product and a plurality of second products; a first similarity determining unit, configured to calculate, for each second product in a set of similar products, a similarity value of values of the first product and the second product on each non-nominal class attribute; a second similarity determining unit, configured to, when calculating, for each second product in the set of similar products, a similarity value of the first product and the second product with respect to each nominal class attribute, execute: when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product; the product similarity determining unit is used for calculating the product similarity values of the first product and the second product according to the similarity value of the non-nominal class attribute values of the first product and the second product determined by the first similarity determining unit and the similarity value of the nominal class attribute values of the first product and the second product determined by the second similarity determining unit; and the linked list determining unit is used for taking at least one second product which is ranked at the front with the first product similarity value as a candidate product linked list of the first product according to the product similarity values of the first product and each second product determined by the third similarity determining unit.

A system for providing a linked list of candidate products, comprising: the web server is used for sending a candidate product query request to the device for determining the candidate product linked list, wherein the request comprises the identification of the specified product; the transaction record database is used for storing a user purchase record set; the product attribute database is used for storing the value of each attribute of each product; a device for determining a candidate product linked list, configured to determine a candidate product linked list corresponding to each product according to a user purchase record set stored in the transaction record database and a value of each attribute of each product stored in the product attribute database, and store an identifier of each product and a corresponding relationship of the candidate product linked list of the product; after receiving a candidate product query request sent by a web server, selecting a candidate product linked list corresponding to the identifier of the specified product contained in the product query request from the stored identifiers of the products and the corresponding relation of the candidate product linked lists, and providing the selected candidate product linked list to the web server.

The method comprises the steps of determining similarity values of two products on a nominal class attribute in the process of determining the similarity values of the two products, determining whether tendency characteristics of an integral user group on purchasing the products with the nominal class attribute value consistent with the attribute value of a first product and the products with the nominal class attribute value consistent with the attribute value of a second product are similar or not according to the value of the nominal class attribute of each product and records of purchasing the products by the integral user, and if the tendency characteristics are similar, determining that the similarity values of the attribute value of the nominal class attribute of the first product and the attribute value of the nominal class attribute of the second product are higher; otherwise, the similarity value is low, so that the similarity value can be determined according to the inherent semantic meaning of the attribute value, and the accuracy of calculating the similarity value of the nominal attribute value is improved.

Drawings

FIG. 1 is a flow diagram of a prior art method for providing a linked list of candidate products associated with a given product;

FIG. 2 is a flow chart of the main implementation principle of the embodiment of the present application;

FIG. 3a is a schematic block diagram of a system for providing a linked list of candidate products in an embodiment of the present application;

FIG. 3b is a flowchart illustrating a process of providing a candidate product linked list to a user according to a first embodiment of the present disclosure;

FIG. 4 is a schematic structural diagram of an apparatus for determining a candidate product linked list according to a first embodiment of the present application;

FIG. 5 is a flowchart illustrating a method for calculating a correlation value between attribute values of a product A and a product B in a nominal class attribute I according to a first embodiment of the present application;

FIG. 6a is a diagram illustrating a user attribute value matrix according to an embodiment of the present application;

fig. 6B is a schematic diagram of a column vector corresponding to an attribute value of a product a nominal class attribute I and a column vector corresponding to an attribute value of a product B nominal class attribute I in a user attribute value matrix in an embodiment of the present application;

FIG. 7 is a flowchart illustrating a method for determining similarity values of nominal class attribute values in a conditional probability manner according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of an apparatus for determining a candidate product linked list in the second embodiment of the present application.

Detailed Description

The applicant finds that the reason why the prior art cannot provide the candidate products which are more related to the product selected by the user to the user preferentially is that: when calculating the similarity value of the same nominal class attribute value between two products, the similarity value is determined based on a hard calculation paradigm, namely based on the occurrence frequency of the nominal class attribute values of the two products in the nominal class attribute values of all products respectively; and the inherent semantic meaning of the attribute value cannot be deeply discovered. For example, for a cosmetic category of products, each product has a brand attribute that is a nominal class attribute, and the value of the attribute is an unordered character string, for example, suppose there are three brands "brand one, brand two, brand three, brand four, brand five, and brand six", and suppose that the three brands "brand one, brand two, and brand three" belong to high-grade brands, and "brand four, brand five, and brand six" belong to low-grade brands. At this time, it cannot be said that the similarity between the two brands is high because "brand two" and "brand five" appear with close frequencies in the attribute values of the nominal class attribute, that is, the brand names of all products. When calculating the similarity of the brand attributes, the products of the same high-grade brand should have a higher similarity value, and the products of the high-grade brand and the products of the low-grade brand should have a lower similarity value.

The basic idea of the application is: in the process of calculating the similarity value between two products a (first product) and B (second product), the step of calculating the similarity value of the nominal class attribute value is improved, specifically:

for each nominal class attribute, the tendency characteristics of the products purchased by the whole user group on the value of the nominal class attribute of product A (i.e. which users tend to purchase products with the value of the nominal class attribute consistent with the value of the nominal class attribute of product A, such as which users tend to purchase products of brand two) and the tendency characteristics of the products purchased by the whole user group on the value of the nominal class attribute of product B (i.e. which users tend to purchase products with the value of the nominal class attribute consistent with the value of the nominal class attribute of product B, such as which users tend to purchase products of brand five), are determined according to the records of the products purchased by the users respectively with the value of the nominal class attribute consistent with the value of the nominal class attribute of product A, and the records of the products purchased by the users with the value of the nominal class attribute consistent with the value of product B, such as which users tend to purchase products of brand five), if the products purchased by the whole user group exhibit the tendency characteristics of the products purchased on the value of the nominal class attribute of product A The same tendency is shown in the value of the nominal category attribute of the product B, that is, a user who purchases a product of the brand "brand two" also typically purchases a product of the brand "brand five", and then the similarity between the brand "brand two" and the brand "brand five" is considered to be high in the nominal category attribute of the product brand, otherwise the similarity between the brand "brand two" and the brand "brand five" is considered to be low.

On the basis, determining similarity values of the product currently selected by the user and other products by combining a correlation value calculation method of attribute values on other category attributes; and determining a candidate product linked list preferentially provided for the user according to the calculated similarity value.

As shown in fig. 2, the main implementation principle flow of the embodiment of the present application is as follows:

step 10, providing a homogeneous product set including a first product and a plurality of second products, and executing steps 20 to 50 for each second product in the homogeneous product set:

step 20, calculating a similarity value of the values of the first product and the second product on each non-nominal class attribute;

step 30, when calculating the similarity value of the value of each nominal class attribute of the first product and the second product, executing: when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product;

step 40, calculating the product similarity values of the first product and the second product according to the similarity value of the non-nominal class attribute values of the first product and the second product calculated in the step 20 and the similarity value of the nominal class attribute value calculated in the step 30;

and step 50, taking at least one second product which is ranked at the top with the similarity value of the first product as a candidate product linked list of the first product.

In the calculating of the similarity values of the nominal class attributes of the two products in step 30, when determining the similarity values of the two products on each nominal class attribute, the similarity of the values of the nominal class attributes of the first product and the second product may be set to a highest value, for example, 1, when the values of the nominal class attributes of the first product and the second product are the same.

Two embodiments will be described in detail below to illustrate and explain the main implementation principles of the method of the present application, based on the above inventive principles of the present application.

Example one

Referring to fig. 3a, a schematic structure diagram of a system for providing a candidate product linked list in the embodiment of the present application is shown. The system comprises a browser client, a webpage Web server, a transaction record database, a product attribute database and a device for determining a candidate product linked list.

The method comprises the following steps that a user logs in a webpage server through a browser client to check various product information, selects interested products, confirms to purchase the products and the like;

the web server is used for sending a candidate product query request to the device for determining the candidate product linked list, wherein the request comprises the identification of the specified product;

the transaction record database is used for storing order record data of products purchased by the user, and each order record comprises order generation time, user identification, identification of the products purchased by the user, the quantity of the purchased products and the like;

the product attribute database is used for storing the attribute values of all the attributes of each product;

a device for determining a candidate product linked list, configured to determine, for each product in a set of similar products, a similarity value between the product and each of the other products in the set of similar products according to the similarity value of each attribute value of the product and each of the set attribute weight values, determine a candidate product linked list corresponding to each product according to each attribute value of each product stored in a user purchase record set and a product attribute database stored in a transaction record database, and store an identifier of each product and a corresponding relationship of the candidate product linked list of the product; after receiving a candidate product query request sent by a web server, selecting a candidate product linked list corresponding to the identifier of the specified product contained in the product query request from the stored identifiers of the products and the corresponding relation of the candidate product linked lists, and providing the selected candidate product linked list to the web server;

when the similarity value of the first product and the second product in the homogeneous product set is determined, executing the following steps for each second product in the homogeneous product set: calculating a similarity value of the values of the first product and the second product over each non-nominal class attribute; when calculating the similarity value of the first product and the second product on each nominal class attribute, executing: when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product; calculating the product similarity values of the first product and the second product according to the similarity values of the non-nominal class attributes of the first product and the second product and the similarity values of the nominal class attributes; and taking at least one second product which is ranked at the top with the similarity value of the first product as a candidate product linked list of the first product.

A flowchart of a process for providing a candidate product linked list for a user in an embodiment of the present application is shown in fig. 3 b.

Step 301, a user interacts with a web server through a browser client, and sends a request for browsing products, where the request for browsing products may be a variety of messages, such as a request for viewing a recommended product list, a request for querying product information according to a keyword, or a request for viewing a product list sold in a favorite web store after logging in through a user name and a password;

step 302, after receiving a request for browsing a product, a web server correspondingly provides a product list to a user, where the product list may not only include a product identifier, but also additionally provide a thumbnail, a brief introduction, and the like of the product;

step 303, the user selects one of the products a based on the product list provided by the web server, for example, holding a mouse on a thumbnail of the product, or clicking an identifier of the product, etc.;

step 304, the web server sends a candidate product query request to the device for determining the candidate product linked list, wherein the request comprises the identifier of the product A selected by the user;

step 305, after the device for determining the candidate product linked list receives the candidate product query request, the candidate product linked list corresponding to the identifier of the product a included in the candidate product query request is searched for as ListA from the correspondence (shown in table 1) between the identifiers of the products and the candidate product linked list stored in advance: H-I-J, H, I, J, K, L, M, R, S, T, U in Table 1 are the identifications of other products, respectively; table 1 gives an example of the correspondence of stored product identifications to a linked list of candidate products.

TABLE 1

Product identification	Corresponding candidate product linked list
		A	ListA：H-I-J
B	ListB：K-L-M
		C	ListC：R-S-T-U

The candidate product linked list in table 1 is a list in which the means for determining the candidate product linked list calculates the similarity between product a and other products in advance based on the data stored in the transaction record database and the product attribute database, and adds the product whose similarity with product a exceeds a predetermined threshold to the candidate product set C_AAnd C is_AThe products in (1) are sorted according to the similarity value from high to low, and a predetermined number of products ranked at the top are selected from the sorted products to generate a candidate product linked list, in this embodiment, the candidate product linked list of the product a is ListA: H-I-J;

in order to improve the efficiency of calculating the similarity values of the two products, the product categories may be classified in advance, and the similarity values of the other products in the same product set to which the product a belongs and the product a are only calculated, for example, if the product a selected by the user is "brand five" cream, then the similarity values between the other products in the cosmetic product set and the product a "brand five" cream only need to be calculated.

Step 306, determining the candidate product linked list ListA to be found by the candidate product linked list determining device: H-I-J is sent to a webpage server;

step 307, the web server displays the candidate product linked list sent by the device for determining the candidate product linked list to the user through the browser client;

step 308, the user determines the product to be purchased according to the displayed candidate product linked list, and after confirming the purchase, sends a confirmation purchase notification to the web server, wherein the confirmation purchase notification comprises the identifier of the product confirmed to be purchased;

in step 309, the web server generates an order after receiving the purchase confirmation notification, and stores the user purchase record in the transaction record database.

In the step 306, the information can also be directly sent to the browser client to be displayed to the user; i.e. not relayed through the web server, step 307 is omitted.

In the above step 305, a schematic structural diagram of the apparatus for determining the candidate product linked list is shown in fig. 4, when determining the candidate product linked list, the apparatus first calculates a similarity value between attribute values of each product (taking product a as an example) in a similar product set and each attribute value of each other product (taking product B as an example) in the similar product set one by one, and calculates the similarity value between product a and product B according to the correlation value between the attribute values of product a and product B, because the present application mainly describes in detail the improvement in calculating the correlation values of the nominal class attribute values of two products through the steps in fig. 5, the method for calculating the correlation values of the attribute values of the non-nominal class attributes, such as the correlation values of the numerical attribute, the ordinal attribute, and the set class attribute, is similar to the prior art, and is not described herein again.

Here, the calculation of the correlation value of the values of the nominal class attributes of product a and product B, which are identified as Attribute _ I, is taken as an example.

Step 501, determining a value set ITEM of a nominal class Attribute Attribute _ I according to the values of the nominal class Attribute Attribute _ I of each product stored in a product Attribute database;

the data Table _ P in the product attribute database stores values of each attribute of each product, the storage structure of the data Table _ P is shown in Table 2, each row contains values of each attribute of the same product, each column contains values of the same attribute of each product, and it can also be understood that values of different attributes of one product are respectively stored in different fields of the same row.

TABLE 2

By querying the SQL statement "Select distinguint Attribute _ I from Table _ P" through the database, different Attribute values in a field corresponding to the Attribute _ I can be obtained from Table _ P, so as to obtain the Attribute value set value _ Attribute _ I containing N elements of the Attribute _ I ═ ITEM ═ value_iN, in this embodiment, the Attribute value set value _ Attribute _ I specifically includes 3 different values: ITEM₁、ITEM₂And ITEM₃。

Step 502, obtaining a user Set _ U of purchased products from a transaction record database;

assuming that the storage structure of the order data Table _ T in the transaction record database is as shown in Table 2, different fields of each row respectively store various related data of one order record, including order generation time, user identification, identification of products purchased by the user, number of purchased products, and the like. Using Table 3 wherein action 3, the order record with serial number 55 indicates the identification u₁₀₀The user of (1), purchased 1 product identified as a at 00 o 18/4/1/2007.

TABLE 3

The field storing the user identifier in the Table _ T is the user, and different user identifiers u in the field user can be obtained from the Table _ T by querying the SQL statement "Select different user from Table _ T" through the database₁₀₀、u₁₀₁Thereby obtaining the user identification Set _ U ═ { U ═₁₀₀，u₁₀₁}；

Preferably, considering that the purchasing behavior of the user is continuous, that is, the user who is used to do online shopping often needs to do online shopping every week or every month, the behavior of the user has certain habitual or tendency characteristics; some users only make an online purchase once in 2 or 3 years, and the behavior of such users has great contingency and is difficult to find tendentiousness, so that the order records of the former users are more useful in comparison, and users who make online transactions within a predetermined time period, such as users who make online transactions within one month, one quarter, half a year, one year and the like, can be further screened from the user identification Set _ U in view of reducing data volume and improving processing efficiency.

Step 503, determining each user identifier U according to the Attribute value Set value _ Attribute _ I of the Attribute _ I obtained in step 501 and the user identifier Set _ U obtained in step 502_iThe triple with the Attribute value of Attribute Attribute _ I<u_i，item_j，1/0>If the user identification is u_iThe user purchased the Attribute value of the Attribute Attribute _ IIs item_jThe third vector value is 1 (or other first predetermined value), i.e.<u_i，item_j，1>(ii) a Otherwise, the third vector value in the feature triplet is 0 (or other second predetermined value), i.e., the third vector value is set to 0<u_i，item_j，0>；

Sequentially extracting each user identifier in the user identifier Set _ U, and constructing a triple between the user identifier and the Attribute value of the Attribute _ I, wherein two construction methods of the triple are provided below, and the following steps are executed:

the first scheme is as follows: the user identifier u is taken out from the order data Table Table _ T_iAccording to the product identification field in the taken-out record, the user identification u can be obtained_iOf the user purchased product identificationObtaining collections from a product attribute databaseThe value of each product Attribute Attribute _ I in (1) is defined as the first vector u_iThe second vector isSetting the third vector value of the triple of the values of the Attribute Attribute _ I of each product to be 1; let the first vector be u_iThe second vector is the Attribute value Set _ Attribute _ I except the user identifier u_iThe third vector value in the triple of the Attribute value other than the value of the product Attribute Attribute _ I that the user has purchased is set to 0; thereby obtaining each user identification u_iN triples corresponding to the number of Attribute values contained in the Set of Attribute values Set _ Attribute _ I.

Scheme II: sequentially acquiring each Attribute value item in the Attribute value Set _ Attribute _ I_kAnd k ranges from 0 to N (the number of elements included in the Attribute value Set _ Attribute _ I), executing the following SQL statement:

Select*

From Table_T and Table_P

Where T.user＝”u_i”and T.product＝P.product and P.Attribute_I＝”item_k”

if the return value after the execution of the statement is not null, the user u is indicated_iThe value purchased for the nominal class Attribute Attribute _ I is item_kThe first vector is u_iAnd the second vector is item_kIs set to 1, i.e. the third vector in the triplet<u_i，item_k，1>(ii) a Otherwise, the first vector is u_iAnd the second vector is item_kIs set to 0, i.e. the third vector in the triplet of<u_i，item_k，0>。

Step 504, determining a user Attribute value matrix of the nominal class Attribute _ I according to the N triples of the nominal class Attribute _ I corresponding to each user determined in step 503, wherein each row in the user Attribute value matrix contains information whether a product purchased by the same user has each Attribute value in the Attribute value set value _ Attribute _ I, and each column in the matrix contains information whether a product purchased by each user has the same Attribute value in the Attribute value set value _ Attribute _ I;

as shown in fig. 6a, according to the arrangement order of the Attribute values in the Set of Attribute values Set _ Attribute _ I, the third vectors in the N triples corresponding to the same user are sequentially filled into different positions in the same row of the feature matrix.

Step 505, extracting the Attribute value item of the product A nominal class Attribute Attribute _ I from the Attribute value matrix obtained in step 504_iCorresponding column vectorAnd the Attribute value item of the product B nominal class Attribute Attribute _ I_jCorresponding column vectorAs shown by the black bold line box in fig. 6b, in the present embodiment

Wherein the column vectorThe Attribute value of the product nominal class Attribute Attribute _ I indicating the purchase of the entire user group is item_iWhich users tend to purchase item for the Attribute value of the nominal class Attribute Attribute _ I_iThe product of (1); column vectorThe Attribute value of the product nominal class Attribute Attribute _ I indicating the purchase of the entire user group is item_jWhich users tend to purchase item for the Attribute value of the nominal class Attribute Attribute _ I_jThe product of (1).

Step 506, calculating the data extracted in step 505Andthe calculated cross-correlation value is taken as the similarity value sim of the value of the nominal class Attribute Attribute _ I of the product A and the product B_i(item_i，item_j)；

It should be noted that, instead of calculating the cross-correlation degree in step 506, a conditional probability method may be used to determine the nominal class Attribute _ I of product a and product BSimilarity value sim of attribute values_i(item_i，item_j) The detailed process is shown in fig. 7:

step 701, determining a first user set for purchasing products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of a first product, a second user set for purchasing products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of a second product, a third user set for simultaneously purchasing products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the first product and a third user set for purchasing products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the second product according to records that the products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the first product and the products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the second product are purchased by each user;

in this embodiment, according to the ith and jth columns in FIG. 6b, the Attribute value for purchasing the nominal class Attribute Attribute _ I can be obtained as item_iUser set U of products_A(i.e. theA set of users corresponding to an element with an element value of 1), an Attribute value of purchasing a nominal class Attribute Attribute _ I is item_jUser set U of products_B(i.e. theA set of users corresponding to an element with a middle element value of 1), and a concurrent purchase attribute value of item_iProduct and attribute value of item_jUser set U of products_AB；

Step 702, according to the first user set, the second user set and the third user set, determining a first conditional probability that a user purchases a product with an attribute value consistent with the nominal attribute value of the second product when purchasing a product with an attribute value consistent with the nominal attribute value of the first product, and a second conditional probability that a user purchases a product with an attribute value consistent with the nominal attribute value of the first product when purchasing a product with an attribute value consistent with the nominal attribute value of the second product;

according to user set U_AUser set U_ABDetermining that the user is purchasing an attribute value of item_iPurchase attribute value of item_jThe conditional probability of the product of (a) is:

P(B|A)＝|U_AB|/|U_Awhere | U | is the number of elements contained in the set U;

the purchase attribute value item can be obtained by the same way_jPurchase attribute value of item in case of product of_iThe conditional probability of the product of (a) is:

P(A|B)＝|U_AB|/|U_B|；

and 703, taking the arithmetic mean value of the first conditional probability and the second conditional probability as the similarity value for determining the nominal class attribute values of the first product and the second product.

Determining an Attribute value item of a nominal class Attribute Attribute _ I of product A_iAnd the Attribute value item of the nominal class Attribute Attribute _ I of product B_jSimilarity value sim of_i(item_i，item_j) Comprises the following steps:

sim_i(item_i，item_j)＝(P(B|A)+P(A|B))/2

that is, if the Attribute value of the purchase nominal class Attribute Attribute _ I is item_iWhile the user is also more likely to purchase the Attribute value item of the nominal class Attribute Attribute _ I_jThe product of (2) is illustrated in the attribute value item_iAnd attribute value item_jHas higher similarity.

Of course, fig. 7 only shows a scheme for calculating similarity values of the two product nominal class attributes with higher accuracy, and the problem that the semantic meaning of the attribute values cannot be found can be solved by directly using the first conditional probability and the second conditional probability as the similarity values.

As shown in fig. 4, the apparatus for determining a candidate product linked list specifically includes a providing unit 401, a first similarity determining unit 402, a second similarity determining unit 403, a product similarity determining unit 404, and a linked list determining unit 405; it may further preferably include a storage unit 406, a receiving unit 407, a selecting unit 408, and a transmitting unit 409, wherein:

a providing unit 401 for providing a homogeneous product set comprising a first product and a plurality of second products;

a first similarity determining unit 402, configured to calculate, for each second product in the set of similar products, a similarity value of values of the first product and the second product on each non-nominal class attribute;

a second similarity determining unit 403, configured to, when calculating, for each second product in the set of similar products, a similarity value between the first product and the second product on each nominal class attribute, perform: when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product;

a product similarity determining unit 404, configured to calculate a product similarity value of the first product and the second product according to the similarity value of the non-nominal class attribute values of the first product and the second product determined by the first similarity determining unit 402 and the similarity value of the nominal class attribute values of the first product and the second product determined by the second similarity determining unit 403;

a linked list determining unit 405, configured to, according to the product similarity value of the first product and each second product determined by the product similarity determining unit 404, take at least one second product ranked earlier than the first product similarity value as a candidate product linked list of the first product;

a storage unit 406, configured to store the identifier of the first product and the corresponding relationship of the candidate product linked list determined by the linked list determining unit 405;

a receiving unit 407, configured to receive a candidate product query request;

the selecting unit 408 is configured to select, after the receiving unit 407 receives the candidate product query request, a candidate product linked list corresponding to the identifier of the product included in the candidate product query request from the correspondence between the identifier of the product and the candidate product linked list stored in the storage unit 406;

a sending unit 409, configured to send the candidate product linked list obtained by the selecting unit 408 to the web server.

For a specific process of calculating the similarity value between the nominal class attribute values of the product a and the nominal class attribute values of the other products B, please refer to fig. 5 to 7, which are not described herein again.

When determining the correlation value of the attribute values of a certain nominal class of attributes of a product a and a product B, determining whether the tendency characteristics of the nominal class attribute value of the products purchased by the whole user group on the attribute value of the product A and the attribute value of the product B are similar or not according to the attribute value of the nominal class attribute of each product and the record of each user for purchasing the products, if so, the similarity of the attribute values of the nominal class attributes of the product A and the product B is higher, otherwise, the similarity of the attribute values of the nominal class attributes of the product A and the product B is determined to be lower, therefore, the semantic similarity between the attribute value of the nominal attribute of the product A and the attribute value of the nominal attribute of the product B can be found, the problems in the prior art are solved, the accuracy of calculating the similarity value of the nominal attribute values is improved, and the accuracy of calculating the candidate product linked list is further improved.

Example two

The first embodiment provides a scheme for providing a subsequent product linked list of a selected product A in an off-line mode: calculating the similarity value of the product A and other products, providing a candidate product linked list based on the calculated similarity value of the selected product A and other products, storing the corresponding relation between the identifier of each product and the candidate product linked list of the product, selecting the candidate product linked list corresponding to the product identifier contained in the candidate product query request from the corresponding relation between the stored product identifier and the candidate product linked list of the product according to the product identifier contained in the candidate product query request when receiving the candidate product query request, and sending the selected candidate product linked list to a webpage server. The corresponding relation between the product identification pre-stored in the scheme and the candidate product linked list of the product occupies certain storage resources, and the probability that the corresponding relation between part of the product identification and the candidate product linked list of the product is retrieved is lower. Therefore, the present application further provides a way to determine the candidate product linked list online, that is, the device for determining the candidate product linked list in step 305 does not select the candidate product linked list corresponding to the product identifier included in the candidate product query request from the stored correspondence between each product identifier and the candidate product linked list of the product, but calculates the candidate product linked list corresponding to the product identifier included in the candidate product query request in real time according to the data in the transaction record database and the product attribute database.

Preferably, in consideration of the scheme of calculating the candidate product linked list in real time, when the number of other products in the same product set is large, the calculation of the similarity value takes much time, and if the device for determining the candidate product linked list in step 305 receives a large number of candidate product query requests sent by the web server in a short time, the processing pressure will increase, so that the candidate product query requests can be screened in advance according to other attribute values except the nominal class attribute, and if the product identifier in the candidate product query request corresponds to the designated product a and other products B except the nominal class attribute, the designated product a and the other products B except the nominal class attributeIf the similarity of the attribute values of other types of attributes (such as numerical type and the like) except the nominal attributes is lower than the set threshold, the similarity of the attribute values on the nominal attributes of the product A and the product B does not need to be calculated, and the product B is directly excluded from the candidate product set C_AAnd (c) out.

Fig. 8 is a schematic structural diagram of an apparatus for determining a candidate product linked list according to an embodiment of the present application. The device for determining the candidate product linked list specifically comprises: a receiving unit 801, a product similarity value determining unit 802, a linked list determining unit 803 and a sending unit 804, wherein:

a receiving unit 801, configured to receive a candidate product query request sent by a web server;

a product similarity value determining unit 802, configured to determine, for a specified product corresponding to a product identifier included in the query request, a similarity value between the specified product and each other product in the same-class product set as well as a similarity value between the specified product and the other product according to the similarity value of each attribute value of each other product and each set attribute weight value, where when determining the similarity value of each attribute value on each nominal class attribute of the specified product and each other product, the following steps are performed: when the attribute values of the nominal class attributes of the specified product and each other product are different, determining similarity values of the nominal class attributes of the specified product and the values of the nominal class attributes of the other products according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the specified product in the purchase record set to purchase the product corresponding to the value of the nominal class attribute of the other products;

a linked list determining unit 803, configured to use at least one other product ranked with the similarity value of the specified product as a candidate product linked list of the specified product;

a sending unit 804, configured to send the candidate product linked list obtained by the linked list determining unit 803 to the web server.

Those skilled in the art will appreciate that all or part of the steps in the method for implementing the above embodiments may be implemented by relevant hardware instructed by a program, and the program may be stored in a computer readable storage medium, such as: ROM/RAM, magnetic disk, optical disk, etc.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.

Claims

1. A method for determining a linked list of candidate products, comprising:

providing a collection of homogeneous products comprising a first product and a plurality of second products, performing for each second product in the collection of homogeneous products:

calculating a similarity value of the values of the first product and the second product over each non-nominal class attribute;

when calculating the similarity value of the first product and the second product on each nominal class attribute, executing: when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product;

calculating the product similarity values of the first product and the second product according to the similarity values of the non-nominal class attributes of the first product and the second product and the similarity values of the nominal class attributes;

taking at least one second product which is ranked at the top with the similarity value of the first product as a candidate product of the first product;

the determining, according to the tendency of a user who purchases a product corresponding to the value of the nominal class attribute of the first product in the purchase record set to purchase a product corresponding to the value of the nominal class attribute of the second product, a similarity value between the value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product includes:

determining a first user set, a second user set and a third user set according to the purchase record set;

the first user set is a set of products which are purchased from the purchase record set and have the value of the nominal class attribute consistent with the value of the nominal class attribute of the first product;

the second user set is a set of products which are purchased from the purchase record set and have the value of the nominal class attribute consistent with that of the second product;

the third user set is a set which purchases the products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the first product and purchases the products of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the second product in the purchase record set;

according to the first user set, the second user set and the third user set, determining a first conditional probability that a user purchases a product of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the first product under the condition that the user purchases the product of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the second product, and determining a second conditional probability that the user purchases the product of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the first product under the condition that the user purchases the product of which the value of the nominal class attribute is consistent with the value of the nominal class attribute of the second product;

taking the arithmetic mean value of the first conditional probability and the second conditional probability as the similarity value of the values of the first product and the second product on the nominal class attribute; or

determining a user attribute value relation matrix of the nominal class attribute according to a purchase record set, wherein each list in the user relation matrix indicates a record of whether a product with the same value of the nominal class attribute is purchased by each user;

selecting the column corresponding to the value of the nominal class attribute of the first product from the user attribute value relation matrixAs a record of the product having the value of the nominal class attribute identical to that of the first product and purchased by each user, a column corresponding to the value of the nominal class attribute of the second productA record of each user purchasing a product having the value of the nominal class attribute consistent with the value of the nominal class attribute of the second product;

similarity values sim (i, j) of the nominal class attribute values of said first product and said second product are

2. The method of claim 1, wherein determining a user attribute value relationship matrix for the nominal class attribute based on the set of purchase records comprises:

for each user purchase record in the set of user purchase records, performing:

determining the value of the nominal class attribute of the product purchased by the user according to the user purchase record and the value of the nominal class attribute of each product; and the number of the first and second electrodes,

determining the nominal class attribute vector of the user according to the determined value of the nominal class attribute of the product purchased by the userWhere m is the identity of the user for whichEach element R in (1)_m,iWhere i is an identification of a value in the set of values, and if the user purchases a product identified as the value of i, the element R is assigned to_m,iIs set to a first value; otherwise the element R_m,iIs set to a second value;

and taking the nominal class attribute vector of each user as a row in the matrix, and determining the user attribute value relation matrix of the nominal class attribute.

3. The method of claim 1, wherein calculating a similarity value for the first product and the second product for the value on each nominal class attribute, further comprises:

and when the value of the nominal class attribute of the first product is the same as that of the second product, determining that the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product is the set highest value.

4. The method of any one of claims 1 to 3, wherein after selecting at least one second product with a top ranking as the candidate product list for the first product, further comprising:

storing the identification of the first product and the corresponding relation of the determined candidate product linked list;

and after receiving the candidate product query request, selecting a candidate product linked list corresponding to the product identifier contained in the candidate product query request from the stored product identifier and the corresponding relation of the candidate product linked list.

5. A method for determining a similarity value between values on a nominal class attribute between two products, comprising:

when calculating the similarity value of the values of the first product and the second product on the nominal class attribute, executing:

when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product;

determining a similarity value between the value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to a tendency of a user who purchases a product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase a product corresponding to the value of the nominal class attribute of the second product, specifically comprising:

selecting the column corresponding to the value of the nominal class attribute of the first product from the user attribute value relation matrixAs products whose value of the nominal class attribute coincides with that of the first productThe record purchased by each user is a column corresponding to the value of the nominal class attribute of the second productA record of each user purchasing a product having the value of the nominal class attribute consistent with the value of the nominal class attribute of the second product;

Or

Determining a similarity value between the value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to a tendency of a user who purchases a product corresponding to the value of the nominal class attribute of the first product in the purchase record set to purchase a product corresponding to the value of the nominal class attribute of the second product, specifically comprising:

and taking the arithmetic mean value of the first conditional probability and the second conditional probability as the similarity value of the values of the first product and the second product on the nominal class attribute.

6. An apparatus for determining a linked list of candidate products, comprising:

a providing unit for providing a homogeneous product set comprising a first product and a plurality of second products;

a first similarity determining unit, configured to calculate, for each second product in a set of similar products, a similarity value of values of the first product and the second product on each non-nominal class attribute;

a second similarity determining unit, configured to, when calculating, for each second product in the set of similar products, a similarity value of the first product and the second product with respect to each nominal class attribute, execute: when the value of the nominal class attribute of a first product is different from the value of the nominal class attribute of a second product, determining the similarity value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product according to the tendency of a user who purchases the product corresponding to the value of the nominal class attribute of the first product in a purchase record set to purchase the product corresponding to the value of the nominal class attribute of the second product;

the product similarity determining unit is used for calculating the product similarity values of the first product and the second product according to the similarity value of the non-nominal class attribute values of the first product and the second product determined by the first similarity determining unit and the similarity value of the nominal class attribute values of the first product and the second product determined by the second similarity determining unit;

the linked list determining unit is used for taking at least one second product which is ranked at the front with the first product similarity value as a candidate product linked list of the first product according to the product similarity values of the first product and each second product determined by the third similarity determining unit;

wherein the second similarity determination unit is specifically configured to:

Or

The second similarity determination unit is specifically configured to:

7. The apparatus of claim 6, further comprising:

the storage unit is used for storing the identifier of the first product and the corresponding relation of the candidate product linked list determined by the linked list determining unit;

the receiving unit is used for receiving a candidate product query request;

and the selecting unit is used for selecting the candidate product linked list corresponding to the identifier of the product contained in the candidate product query request from the identifiers of the products stored in the storage unit and the corresponding relation of the candidate product linked lists after the receiving unit receives the candidate product query request.

8. The apparatus of claim 6, wherein the second similarity determining unit is further configured to determine the similarity value between the value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product to be the highest value when the value of the nominal class attribute of the first product and the value of the nominal class attribute of the second product are the same.