New and old commodity mixed recommendation method and device, electronic equipment and storage medium

Abstract

The invention provides a mixed recommendation method for new and old commodities, and the method comprises the steps: calculating the commodity similarity between the new commodity and each old commodity based on the commodity information of the new commodity and the commodity information of each old commodity, and obtaining a similarity matrix between the new commodity and the old commodity, the similarity matrix and the first scoring matrix have the same old commodity arrangement sequence; adding new commodities and initial score values of the new commodities into the first score matrix to obtain a second score matrix; on the basis of the similarity matrix and the first scoring matrix, the scoring values of the new commodities and the old commodities in the second scoring matrix are updated, and a third scoring matrix is obtained; and carrying out commodity recommendation based on the third scoring matrix. According to the invention, new commodities and old commodities can be recommended in a mixed manner, and the shopping experience of the user is improved.

Description

Technical Field

[0001] The present invention relates to the field of commodity recommendation, and in particular to a method, device, electronic device and storage medium for mixed recommendation of new and old commodities. Background Art

[0002] As people's personalized needs for items increase, various item recommendation methods have emerged. Most item recommendations are based on the user's behavior towards items, and some mathematical algorithms are used to predict the items that the user likes, and then recommend the obtained items to the user. At present, the existing recommendation methods mainly use the user's behavior towards the goods to make recommendations, but for new items, there is no user's behavior towards new items. Therefore, in most cases, new items can only be recommended violently, regardless of whether the user wants to buy the new item, the new item will be recommended, which reduces the user's shopping experience. Summary of the invention

[0003] The embodiment of the present invention provides a mixed recommendation method for new and old goods, which can realize the mixed recommendation of new and old goods, and improve the personalization and accuracy of the recommendation. By obtaining a first scoring matrix, obtaining the product information of the new product and the product information of each old product, and calculating the product similarity between the new product and each old product based on the product information of the new product and the product information of the old product, a similarity matrix between the new product and the old product is obtained, and the new product and the initial score value of the new product are added to the first scoring matrix to obtain a second scoring matrix, and the score values ​​of the new product and each old product in the second scoring matrix are updated through the similarity matrix and the first scoring matrix to obtain a third scoring matrix, and the third scoring matrix is ​​used to recommend products, so that new and old products can be mixed and recommended to improve the user's shopping experience.

[0004] In a first aspect, an embodiment of the present invention provides a method for mixed recommendation of new and old goods, the method comprising the following steps: Obtaining a first rating matrix, where the first rating matrix includes ratings of different users for each used product; Acquire product information of the new product and product information of each of the old products, and calculate product similarity between the new product and each of the old products based on the product information of the new product and the product information of the old products, to obtain a similarity matrix between the new product and the old products, wherein the similarity matrix has the same arrangement order of the old products as the first scoring matrix; Adding the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix; Based on the similarity matrix and the first scoring matrix, the scoring values ​​of the new products and each of the old products in the second scoring matrix are updated to obtain a third scoring matrix; Product recommendations are made based on the third rating matrix.

[0005] Optionally, adding the new product and the initial score value of the new product to the first score matrix to obtain a second score matrix includes: Determine whether there is a replacement product for the new product among each of the old products; If there is a replacement product for the new product, the replacement product is replaced by the new product in the first rating matrix, and the rating value of the replacement product is replaced by the initial rating of the new product to obtain the second rating matrix; If there is no replacement product for the new product, then in the first scoring matrix, an adjacent product of the new product is determined, where the adjacent product is the old product with the greatest similarity to the new product; The new product and the initial rating value of the new product are inserted into adjacent positions of the adjacent products to obtain the second rating matrix.

[0006] Optionally, inserting the new product and the initial score value of the new product into the adjacent position of the adjacent product to obtain the second score matrix includes: Taking the adjacent product as a boundary, dividing the similarity matrix into a first-side similarity matrix and a second-side similarity matrix, wherein the first-side similarity matrix is ​​located on one side of the adjacent product, and the second-side similarity matrix is ​​located on the other side of the adjacent product; Based on a preset first product similarity threshold, filtering product similarities in the first-side similarity matrix and the second-side similarity matrix that are less than the first product similarity threshold to obtain effective product similarities that are greater than or equal to the first product similarity threshold; If the sum of similarities of valid products in the similarity matrix on the first side is greater than the sum of similarities of valid products in the similarity matrix on the second side, determining the side where the similarity matrix on the first side is located as the insertion side of the adjacent product; If the sum of similarities of valid commodities in the first-side similarity matrix is ​​less than the sum of similarities of valid commodities in the second-side similarity matrix, determining the other side where the second-side similarity matrix is ​​located as the insertion side of the adjacent commodity; The new product and the initial score value of the new product are inserted into the adjacent position on the insertion side of the adjacent product to obtain the second score matrix.

[0007] Optionally, based on the similarity matrix and the first scoring matrix, updating the scoring values ​​of the new products and each of the old products in the second scoring matrix to obtain a third scoring matrix includes: Performing matrix decomposition on the first rating matrix to obtain a first user matrix and a first product matrix, wherein the first user matrix includes first user feature values ​​corresponding to each of the users, and the first product matrix includes first product feature values ​​corresponding to each of the old products; Performing matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix, wherein the second user matrix includes second user feature values ​​corresponding to each of the users, and the second product matrix includes second product feature values ​​corresponding to the new products and each of the old products, and the similarity between the second user matrix and the first user matrix is ​​greater than a preset similarity; Based on the first commodity matrix, the second commodity matrix and the similarity matrix, a third commodity matrix is ​​calculated, wherein the third commodity matrix includes the third commodity feature values ​​corresponding to the new commodity and each of the old commodities; Perform matrix multiplication on the second user matrix and the third product matrix to obtain a third rating matrix.

[0008] Optionally, performing matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix includes: By performing matrix decomposition on the second rating matrix, a second user matrix and a second product matrix are obtained; Calculating a metric distance between the second user matrix and the first user matrix; Taking the minimum metric distance as the optimization goal, the matrix decomposition process is iterated, and when the metric distance is less than a preset metric distance, the iteration is stopped to obtain the final second user matrix and the final second product matrix.

[0009] Optionally, the calculating a third product matrix based on the first product matrix, the second product matrix and the similarity matrix includes: By interpolation, the similarity matrix is ​​dimensionally expanded in a non-product dimension to obtain a similarity expansion matrix; Performing matrix multiplication calculation on the first product matrix and the similarity extension matrix to obtain an intermediate matrix; Perform matrix multiplication calculation on the intermediate matrix and the second product matrix to obtain a third product matrix.

[0010] Optionally, before adding the new product and the initial score value of the new product to the first score matrix to obtain the second score matrix, the method further includes: Based on a preset second product similarity threshold, selecting the old products whose product similarity is greater than the second product similarity threshold from among all the old products as reference products; Acquire historical sales data of each of the reference products within a preset time period, where the starting time of the preset time period is the time when the reference product was launched on the market; Based on the score value-sales data relationship table of each reference commodity, determine the historical score value corresponding to the historical sales data, each old commodity corresponds to one score value-sales data relationship table; According to the historical rating values ​​of the respective commodities, an average historical rating value is calculated as the initial rating value of the new commodity.

[0011] In a second aspect, an embodiment of the present invention further provides a mixed recommendation device for new and old goods, the mixed recommendation device for new and old goods comprising: A first acquisition module is used to acquire a first rating matrix, wherein the first rating matrix includes ratings of different users for each used product; A second acquisition module is used to acquire product information of the new product and product information of each of the old products, and based on the product information of the new product and the product information of the old products, calculate the product similarity between the new product and each of the old products to obtain a similarity threshold between the new product and the old products, wherein the similarity matrix has the same arrangement order of the old products as the first scoring matrix; An adding module, configured to add the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix; An updating module, configured to update the score values ​​of the new products and each of the old products in the second scoring matrix based on the similarity matrix and the first scoring matrix, to obtain a third scoring matrix; A product recommendation module is used to make product recommendations based on the third rating matrix.

[0012] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the mixed recommendation method for new and used goods provided in an embodiment of the present invention when executing the computer program.

[0013] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the mixed recommendation method for new and used goods provided in the embodiment of the invention.

[0014] In an embodiment of the present invention, a first rating matrix is ​​obtained, the first rating matrix includes ratings of different users for each old product; product information of a new product and product information of each old product are obtained, and based on the product information of the new product and the product information of the old product, the product similarity between the new product and each old product is calculated to obtain a similarity matrix between the new product and the old products, and the similarity matrix has the same arrangement order of the old products as the first rating matrix; new products and initial rating values ​​of the new products are added to the first rating matrix to obtain a second rating matrix; based on the similarity matrix and the first rating matrix, the rating values ​​of the new product and each old product in the second rating matrix are updated to obtain a third rating matrix; and product recommendations are performed based on the third rating matrix. The present invention obtains the product information of the new product and the product information of each old product by obtaining a first scoring matrix, and calculates the product similarity between the new product and each old product according to the product information of the new product and the product information of the old product to obtain a similarity matrix between the new product and the old product, adds the new product and the initial scoring value of the new product to the first scoring matrix to obtain a second scoring matrix, and updates the scoring values ​​of the new product and each old product in the second scoring matrix through the similarity matrix and the first scoring matrix to obtain a third scoring matrix, and uses the third scoring matrix to perform product recommendation, so as to perform mixed recommendation of new products and old products and improve the shopping experience of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 is a flow chart of a mixed recommendation method for new and old goods provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a mixed recommendation device for new and used goods provided by an embodiment of the present invention; Figure 3 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, Figure 1 1 is a flow chart of a mixed recommendation method for new and old goods provided by an embodiment of the present invention. The mixed recommendation method for new and old goods includes the following steps: 101. Obtain a first scoring matrix.

[0019] In an embodiment of the present invention, the above-mentioned mixed recommendation method for new and old goods can be applied to a goods recommendation platform. The above-mentioned goods recommendation platform can be constructed based on a server or a distributed server. The above-mentioned goods recommendation platform includes a device interface (for data acquisition devices to upload collected data), an algorithm container interface, a database, and a goods recommendation program. The above-mentioned device interface can be used to obtain data from a data acquisition device (including a device identifier), the above-mentioned algorithm container interface is used to obtain data from an algorithm container (including an identifier of an already connected device), the above-mentioned database is used to store the association between the device identifier and the task reasoning coefficient, and the above-mentioned goods recommendation program is used for task creation, goods recommendation, and connecting the data acquisition device corresponding to the device identifier to the corresponding algorithm container. The above-mentioned database can be a relational database, such as a MySQL database. The above-mentioned data acquisition device can be a data acquisition device such as a camera or a camera.

[0020] The first rating matrix includes ratings of different users for each old product. The old products can be understood as existing products, which means products that already have rating data.

[0021] The above rating matrix can be represented in a table format, where each row represents a user, each column represents a product, and each cell in the matrix contains the user's rating of the corresponding product. Each row in the rating matrix records the ratings of different users for each product.

[0022] For example, there are M users and N items, and the first rating matrix is ​​as follows:

[0023] Among them, user 1's ratings for product 1, product 2, product 3, ... product N are 4.0, 4.3, 3.5, ..., 4.3 respectively; user 2's ratings for product 1, product 2, product 3, ... product N are 3.5, 4.7, 4.8, ..., 4.2 respectively; user 3's ratings for product 1, product 2, product 3, ... product N are 3.6, 4.3, 4.6, ..., 3.9 respectively; user M's ratings for product 1, product 2, product 3, ... product N are 4.5, 4.5, 3.6, ..., 4.5 respectively. The above ratings can be set in the range of 1-5. The ratings indicate the user's evaluation of the product. A higher rating indicates that the user is more satisfied with the product, while a lower rating indicates that the user is not very satisfied with the product. Specifically, by analyzing the rating matrix, we can understand the evaluation trends of users on different products.

[0024] 102. Obtain product information of the new product and product information of each old product, and calculate product similarity between the new product and each old product based on the product information of the new product and the product information of the old products, to obtain a similarity matrix between the new product and the old products.

[0025] In the embodiment of the present invention, the similarity matrix and the first scoring matrix have the same old commodity arrangement order, which means that the similarity matrix between the new commodity and the old commodity is arranged according to the arrangement order of the old commodity, that is, it is arranged in the order of commodity 1, commodity 2, commodity 3... commodity N, which can ensure that the similarity matrix and the first scoring matrix can correspond to each other, thereby ensuring the consistency and comparability of the data, making the subsequent analysis or processing more accurate and reliable.

[0026] The above product information includes product name, brand, price, category, description, pictures, etc.

[0027] Furthermore, a similarity calculation method may be used to calculate the similarity between the new product and each old product. The similarity calculation method may be a text similarity calculation method, a cosine similarity calculation method, or the like.

[0028] The similarity matrix can show the similarity between the new product and each old product, and the similarity is a value between 0 and 1. If two products show high similarity in content, category or user behavior, their corresponding values ​​in the similarity matrix will be high.

[0029] The above similarity matrix can be shown as follows:

[0030] 103. Add the new product and the initial score value of the new product to the first score matrix to obtain a second score matrix.

[0031] In the embodiment of the present invention, the initial score value of the new product can be understood as the score set for the new product when the new product is added to the score matrix. The initial score can be any reasonable value. For example, if you are building a product recommendation system and want to add a new product to the system. You can set an initial score for the new product, which can be based on factors such as the name, brand, price, category, etc. of the product, or it can be set based on historical sales data of old products similar to the new product.

[0032] The second rating matrix includes ratings of different users for each old product and new product, as well as initial ratings of different users for the new product.

[0033] For example, suppose there is a first rating matrix containing several products and their ratings. Now you need to add new products and set initial ratings for the new products. In order to take the new products and their ratings into consideration, you need to expand the original first rating matrix to form a second rating matrix. The number of new products can be 1 or more.

[0034] For example, if there is a first rating matrix containing 10 items and their ratings, and then 5 new items are discovered and initial ratings are set for them, then these 5 new items and their ratings can be added to the original first rating matrix to obtain a second rating matrix containing 15 items and their ratings.

[0035] 104. Based on the similarity matrix and the first scoring matrix, the scoring values ​​of the new products and each old product in the second scoring matrix are updated to obtain a third scoring matrix.

[0036] In an embodiment of the present invention, it is considered that the addition of new products will affect the user's ratings of different old products. This impact may be large or small. Specifically, the more similar the old products are to the new products, the greater the impact. Therefore, the ratings of the new products and each old product in the second rating matrix may be updated according to the similarity matrix and the first rating matrix, so that the ratings of the new products and the old products are re-predicted.

[0037] The first rating matrix can be decomposed to obtain a first user matrix and a first product matrix, the second rating matrix can be decomposed to obtain a second user matrix and a second product matrix, and the third product matrix can be calculated by a similarity calculation method based on the first product matrix, the second product matrix and the similarity matrix, and matrix multiplication can be performed on the second user matrix and the third product matrix to obtain a third rating matrix.

[0038] The above matrix decomposition can be understood as the process of splitting a complex matrix into the product of multiple simple matrices. The purpose of matrix decomposition is to reduce the original high-dimensional rating matrix to a low-rank matrix, so that it is easier to discover the potential characteristics of users and products, which helps to improve the accuracy and efficiency of the recommendation system.

[0039] Through matrix decomposition and matrix multiplication, the third rating matrix combines the product feature information of the second product matrix and the third product matrix, so that the new products and the old products are integrated in the dimension of product features, and the rating values ​​of each user for the new products are predicted. The third rating matrix can reflect the user's preference or satisfaction with the new products and the old products.

[0040] In a possible embodiment, the first scoring matrix is ​​an M×N matrix, the similarity matrix is ​​a T×N matrix, and the second scoring matrix is ​​an M×(N+T) matrix. The transposed matrix N×T of the similarity matrix can be linearly transformed through a T×(N+T) linear matrix to obtain a linearly transformed similarity matrix N×(N+T). The value of each matrix unit in the linear matrix is ​​1. The linear transformation is performed on the transposed matrix N×T of the similarity matrix and the linear matrix T×(N+T) for matrix multiplication. The linearly transformed similarity matrix N×(N+T) is a matrix affected by the similarity of the products. The linearly transformed similarity matrix N×(N+T) is normalized to obtain a normalized similarity matrix N×(N+T), and the first scoring matrix M×N is multiplied by the transposed matrix N×(N+T) of the normalized similarity matrix to obtain a scoring update matrix M×(N+T). The scoring values ​​in the third scoring matrix M×(N+T) are subtracted from the scoring update values ​​corresponding to the scoring update matrix M×(N+T) to obtain the third scoring matrix M×(N+T).

[0041] 105. Make product recommendations based on the third rating matrix.

[0042] In the embodiment of the present invention, product recommendations may be made based on the third rating matrix.

[0043] Specifically, the third rating matrix is ​​used to predict the user's possible preference for the new product, so that when making product recommendations, the new product can be recommended together with the old product, so that the user has more choices and the user's shopping experience is improved.

[0044] In the embodiment of the present invention, the present invention can analyze the user's historical purchase records, evaluated products and ratings, combine the product's own attributes and other users' rating data, and use the recommendation algorithm to calculate which products are more in line with the current user's interests and needs, so as to make personalized recommendations. For example, if a user has frequently purchased and given good reviews of a certain type of product before, the system can infer that the user may also be interested in similar or related products, and then select products that meet the user's characteristics from a large number of products for recommendation, which can improve the accuracy of product recommendations and user satisfaction.

[0045] In an embodiment of the present invention, a first rating matrix is ​​taken, the first rating matrix includes ratings of different users for each old product; product information of the new product and product information of each old product are obtained, and based on the product information of the new product and the product information of the old product, the product similarity between the new product and each old product is calculated to obtain a similarity matrix between the new product and the old product, and the similarity matrix has the same arrangement order of the old products as the first rating matrix; the new product and the initial rating value of the new product are added to the first rating matrix to obtain a second rating matrix; based on the similarity matrix and the first rating matrix, the rating values ​​of the new product and each old product in the second rating matrix are updated to obtain a third rating matrix; and product recommendations are performed based on the third rating matrix. The present invention obtains the product information of the new product and the product information of each old product by obtaining a first scoring matrix, and calculates the product similarity between the new product and each old product according to the product information of the new product and the product information of the old product to obtain a similarity matrix between the new product and the old product, adds the new product and the initial scoring value of the new product to the first scoring matrix to obtain a second scoring matrix, and updates the scoring values ​​of the new product and each old product in the second scoring matrix through the similarity matrix and the first scoring matrix to obtain a third scoring matrix, and uses the third scoring matrix to perform product recommendation, so as to perform mixed recommendation of new products and old products and improve the shopping experience of users.

[0046] It is understandable that in the specific implementation of this application, related data such as product data, user data, sales data, etc. are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data, as well as the training, deployment and calling of algorithm models, must comply with relevant laws, regulations and standards of relevant countries and regions.

[0047] Optionally, in the step of adding new products and initial ratings of the new products to the first rating matrix to obtain the second rating matrix, it can be determined whether there are replacement products for the new products among the old products; if there are replacement products for the new products, the replacement products are replaced by the new products in the first rating matrix, and the rating values ​​of the replacement products are replaced by the initial ratings of the new products to obtain the second rating matrix; if there are no replacement products for the new products, the adjacent products of the new products are determined in the first rating matrix; the new products and the initial ratings of the new products are inserted into adjacent positions of the adjacent products to obtain the second rating matrix.

[0048] In the embodiment of the present invention, the adjacent products are old products with the greatest product similarity, and the new products can be added to the adjacent positions of the adjacent products.

[0049] The above-mentioned old products can be understood as products that already exist and have been evaluated by users.

[0050] The replacement product can be understood as a product that is similar or identical to the new product in function, use or attribute. The new product is launched to replace the old product. The old product as the replacement product is no longer sold. Therefore, the new product can replace the old product. In a possible embodiment, after the old product as the replacement product in the first scoring matrix is ​​replaced by the new product, the score value of the old product can be used as the initial score value of the new product, that is, the new product inherits the score value of the old product to obtain the second scoring matrix. At this time, the second scoring matrix will not add a new product column.

[0051] Specifically, it is possible to determine whether there are alternative products that can replace the new products in the old products by analyzing the characteristics, categories, brands, etc. of the products. If there are alternative products for the new products, then the scores of the alternative products are replaced with the initial scores of the new products in the first scoring matrix. It can be understood that, assuming there is a scoring matrix that contains multiple products and their corresponding scores, if there is a product in this matrix that can be replaced by another new product, then the score of the replaced product should be replaced with the score of the new product. In this way, a new scoring matrix is ​​obtained, which contains the information and scores of the new products.

[0052] If there is no replacement product for the new product, it is necessary to find the old product with the greatest similarity to the new product in the first scoring matrix as the adjacent product of the new product. Specifically, the product similarity between the new product and the old product can be calculated by methods such as cosine similarity and Pearson correlation coefficient, and the old product with the greatest product similarity can be used as the adjacent product of the new product. It is also possible to directly find the old product with the greatest product similarity in the similarity matrix as the adjacent product of the new product.

[0053] Furthermore, a new rating matrix can be constructed by inserting new products and initial rating values ​​of the new products at adjacent product positions.

[0054] It should be noted that, first, among the old products, it is determined whether there is a replacement product for the new product. If there is an old product that is a replacement product for the new product, the replacement product is replaced by the new product in the first scoring matrix, and the score of the replacement product is replaced by the initial score of the new product to obtain the second scoring matrix; if there is no replacement product, it is necessary to find the old product with the highest similarity to the new product as the adjacent product, and then insert the new product and its initial score value in the adjacent position of the adjacent product to obtain the second scoring matrix.

[0055] Optionally, in the step of inserting new products and initial ratings of the new products at adjacent positions of adjacent products to obtain a second rating matrix, the similarity matrix is ​​divided into a first-side similarity matrix and a second-side similarity matrix with the adjacent products as the boundary; based on a preset first product similarity threshold, the product similarity values ​​in the first-side similarity matrix and the second-side similarity matrix that are less than the first product similarity threshold are filtered to obtain effective product similarities greater than or equal to the first product similarity threshold; if the sum of effective product similarities in the first-side similarity matrix is ​​greater than the sum of effective product similarities in the second-side similarity matrix, the side where the first-side similarity matrix is ​​located is determined to be the insertion side of the adjacent products; if the sum of effective product similarities in the first-side similarity matrix is ​​less than the sum of effective product similarities in the second-side similarity matrix, the other side where the second-side similarity matrix is ​​located is determined to be the insertion side of the adjacent products; and the new products and the initial ratings of the new products are inserted at adjacent positions on the insertion side of the adjacent products to obtain a second rating matrix.

[0056] In the embodiment of the present invention, the similarity threshold is a similarity threshold between the new product and the old product.

[0057] The first-side similarity matrix is ​​located on one side of the adjacent products, and the second-side similarity matrix is ​​located on the other side of the adjacent products.

[0058] The above-mentioned preset first commodity similarity threshold is a first commodity similarity threshold preset by the system and is used to filter and screen commodity similarity data.

[0059] Specifically, the similarity matrix is ​​divided into two parts according to the positions of adjacent products, namely the first-side similarity matrix and the second-side similarity matrix, which respectively represent the similarity of the products on both sides of the adjacent products; then, the preset first product similarity threshold is used to filter the product similarity values ​​in the first-side similarity matrix and the second-side similarity matrix, and only the similarity values ​​greater than or equal to the preset first product similarity threshold will be retained as valid product similarity; then, the sum of the two parts of valid product similarity is compared. If the effective sum in the first-side similarity matrix is ​​greater than that of the second side, the first side is considered to be the insertion side of the adjacent product; otherwise, the second side is the insertion side; finally, the new product and its initial score value are inserted at the adjacent position of the insertion side, thereby generating a new score matrix, namely the second score matrix.

[0060] Optionally, in the step of updating the score values ​​of new products and each old product in the second scoring matrix based on the similarity matrix and the first scoring matrix to obtain the third scoring matrix, the first scoring matrix can be matrix decomposed to obtain a first user matrix and a first product matrix; the second scoring matrix can be matrix decomposed to obtain a second user matrix and a second product matrix; the third product matrix can be calculated based on the first product matrix, the second product matrix and the similarity matrix; and matrix multiplication can be performed on the second user matrix and the third product matrix to obtain the third scoring matrix.

[0061] In the embodiment of the present invention, the first user matrix includes the first user feature value corresponding to each user, and the first product matrix includes the first product feature value corresponding to each old product. For example, matrix decomposition of the first rating matrix M×N can yield the first user matrix M×U and the first product matrix U×N.

[0062] The first user matrix M×U can be expressed as follows:

[0063] Among them, the first user characteristic value corresponding to user 1 is u 1,1 、u 1,2 、u 1,3 、……、u 1,U , the first user feature value corresponding to user 2 is u 2,1 、u 2,2 、u 2,3 、……、u 2,U , and so on.

[0064] The first commodity matrix U×N above can be shown as follows:

[0065] Among them, the first product feature value corresponding to product 1 is v1,1 、v 2,1 、v 3,1 ,……,v u,1 , the first product feature value corresponding to product 2 is v 1,1 、v 2,2 、v 3,2 ,……,v u,2 , and so on.

[0066] Similarly, by performing matrix decomposition on the second rating matrix M×(N+T), we can obtain the second user matrix M×U and the second product matrix U×(N+T).

[0067] In a possible embodiment, considering that the new product affects the old product, while the new product has little impact on the user, the change of the user feature value can be ignored, and the first user matrix M×U and the second product matrix U×(N+T) are directly matrix multiplied to obtain a new rating matrix M×(N+T), and the columns corresponding to the new products in the similarity matrix T×N are interpolated, and the column interpolation value is 1 (that is, the product similarity is 1), and the first new similarity matrix T×(N+T) is obtained. Then, the rows of the first new similarity matrix T×(N+T) are interpolated, and the row interpolation value is 0 (that is, the product similarity is 0), and the second new similarity matrix (N+T)×(N+T) is obtained. The new rating matrix M×(N+T) is matrix multiplied with the second new similarity matrix (N+T)×(N+T) to obtain a third rating matrix M×(N+T).

[0068] The second user matrix includes the second user feature value corresponding to each user, the second product matrix includes the second product feature value corresponding to the new product and each old product, and the similarity between the second user matrix and the first user matrix is ​​greater than the preset similarity. It should be noted that since the addition of new products has little impact on users, the similarity between the second user matrix and the first user matrix is ​​greater than the preset similarity, when the second rating matrix is ​​decomposed, the user feature value change can be limited to be small, which can better highlight the change in the product feature value dimension between the new product and the old product.

[0069] The third commodity matrix includes the third commodity feature values ​​corresponding to the new commodity and each old commodity.

[0070] The above matrix decomposition can be understood as the process of splitting a complex matrix into the product of multiple simple matrices. The purpose of matrix decomposition is to reduce the original high-dimensional rating matrix to a low-rank matrix, so that it is easier to discover the potential characteristics of users and products, which helps to improve the accuracy and efficiency of the recommendation system.

[0071] The above matrix multiplication calculation is a method of multiplying two matrices. It can be understood that for an M×N matrix A and an N×P matrix B, their product is an M×P matrix C. Each element c (i, j) in the matrix C is the sum of the products of the corresponding elements of the i-th row of A and the j-th column of B. Matrix multiplication is an operation that multiplies two matrices.

[0072] Each element in the third rating matrix is ​​obtained by performing a dot product operation on a corresponding user feature vector in the second user matrix and a corresponding product feature vector in the third product matrix.

[0073] Optionally, in the step of performing matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix, the second rating matrix can be matrix decomposed to obtain a second user matrix and a second product matrix; the metric distance between the second user matrix and the first user matrix is ​​calculated; the matrix decomposition process is iterated with the minimum metric distance as the optimization goal, and when the metric distance is less than the preset metric distance, the iteration is stopped to obtain a final second user matrix and a final second product matrix.

[0074] In the embodiment of the present invention, the matrix decomposition is a process of splitting a complex matrix into the product of multiple simple matrices.

[0075] The second user matrix includes the second user characteristic value corresponding to each user.

[0076] The second commodity feature values ​​corresponding to the new commodities and each old commodity in the second commodity matrix.

[0077] The above metric distance can be understood as a quantitative indicator of the similarity or difference between two vectors. The metric distance is used to measure the degree of difference between two user matrices. The Euclidean distance can be used to calculate the metric distance between the second user matrix and the first user matrix. The smaller the metric distance, the more similar the first user matrix is ​​to the second user matrix.

[0078] Furthermore, by iteratively optimizing the matrix decomposition process, the metric distance can be gradually reduced until a preset threshold is reached. At this time, it is considered that the difference between the two matrices is small enough and the iteration can be stopped, thereby obtaining the final second user matrix and the final second product matrix.

[0079] Optionally, in the step of calculating the third product matrix based on the first product matrix, the second product matrix and the similarity matrix, the similarity matrix can be dimensionally expanded on a non-product dimension by interpolation to obtain a similarity expansion matrix; matrix multiplication is performed on the first product matrix and the similarity expansion matrix to obtain an intermediate matrix; and matrix multiplication is performed on the intermediate matrix and the second product matrix to obtain a third product matrix.

[0080] In the embodiment of the present invention, the above interpolation method can be understood as interpolating the columns and rows of the similarity matrix, thereby expanding the dimension of the similarity matrix. For example, the columns corresponding to the new products in the similarity matrix T×N are interpolated, and the value of the column interpolation is 1 (that is, the product similarity is 1), and the first new similarity matrix T×(N+T) is obtained. Then, the rows of the first new similarity matrix T×(N+T) are interpolated, and the value of the row interpolation is 0 (that is, the product similarity is 0), and the second new similarity matrix (N+T)×(N+T) is obtained.

[0081] In the similarity matrix T×N, the product dimension is the column and the non-product dimension is the row.

[0082] The above-mentioned dimensional expansion can be understood as an expansion process of expanding the similarity matrix from a single dimension to multiple dimensions. The above matrix multiplication calculation is the process of multiplying the corresponding elements of two matrices and then summing them. Specifically, each element of the result matrix is ​​obtained by summing the products of the rows of the first matrix and the corresponding elements of the columns of the second matrix.

[0083] Specifically, after obtaining the first product matrix U×N and the second product matrix U×(N+T), the similarity matrix T×N can be dimensionally expanded on the non-product dimension (row) (interpolate the rows of the similarity matrix T×N, and the row interpolation value is 0) to obtain the similarity expansion matrix U×N, perform matrix multiplication on the first product matrix U×N and the transposed matrix N×U of the similarity expansion matrix to obtain the intermediate matrix U×U, perform matrix multiplication on the intermediate matrix U×U and the second product matrix U×(N+T) to obtain the third product matrix U×(N+T).

[0084] For example, there is an original similarity matrix, which may only consider the direct similarity between users and items. By interpolating non-item dimensions (such as rows) in the similarity matrix, an extended similarity matrix can be created, which considers the direct similarity between items.

[0085] The above intermediate matrix combines the product information of the first product matrix and the similarity information in the similarity extension matrix.

[0086] The third product matrix mentioned above integrates the similarity information of the second product matrix and the intermediate matrix.

[0087] Optionally, before adding the new product and the initial score value of the new product to the first score matrix to obtain the second score matrix, it is also possible to select old products whose product similarity is greater than the second product similarity threshold from all old products as reference products based on a preset second product similarity threshold; obtain the historical sales data of each reference product within a preset time period; determine the historical score value corresponding to the historical sales data based on the score value-sales data relationship table of each reference product; and calculate the average historical score value based on the historical score values ​​of each old product as the initial score value of the new product.

[0088] In an embodiment of the present invention, the above-mentioned preset second product similarity threshold is a similarity threshold pre-set by the system, which is used to measure the similarity between products. It can help the system determine which old products have a higher similarity with new products, and use old products whose product similarity is greater than the second product similarity threshold as reference products.

[0089] The starting time of the above-mentioned preset time period is the launch time of the reference product, and the end time can be set by the user. For example, the above-mentioned preset time period can be within one month or two months of the launch.

[0090] The above historical sales data can be understood as the sales record of each reference product in the past period of time, including sales volume, sales revenue and other information.

[0091] Each old product corresponds to a rating value-sales data relationship table. The rating value-sales data relationship table records the relationship between each product and its rating value, and is used to convert sales data into a rating value relationship table.

[0092] It should be noted that the present invention uses a preset second commodity similarity threshold to screen old commodities, and old commodities whose commodity similarity exceeds the preset second commodity similarity threshold are selected as reference commodities. Such screening helps to ensure that the reference commodities have a high similarity with the new commodities in attributes, categories or other aspects, thereby improving the accuracy of subsequent rating calculations. For each selected reference commodity, its sales data within a preset time period is collected. For each reference commodity, its rating value-sales data relationship table is used to determine the historical rating value corresponding to its historical sales data. The historical rating values ​​of all selected reference commodities are averaged to obtain an average rating value, which will be used as the initial rating value of the new commodity.

[0093] like Figure 2 As shown, an embodiment of the present invention provides a mixed recommendation device for new and old goods, and the mixed recommendation device for new and old goods includes: A first acquisition module 201 is used to acquire a first rating matrix, where the first rating matrix includes ratings of different users for each used product; The second acquisition module 202 is used to acquire the product information of the new product and the product information of each of the old products, and calculate the product similarity between the new product and each of the old products based on the product information of the new product and the product information of the old products, so as to obtain a similarity matrix between the new product and the old products, wherein the similarity matrix has the same arrangement order of the old products as the first scoring matrix; An adding module 203 is used to add the new product and the initial score value of the new product to the first scoring matrix to obtain a second scoring matrix; An updating module 204 is used to update the score values ​​of the new products and each of the old products in the second scoring matrix based on the similarity matrix and the first scoring matrix to obtain a third scoring matrix; The product recommendation module 205 is used to make product recommendations based on the third rating matrix.

[0094] Optionally, the adding module 203 is also used to determine whether there is a replacement product for the new product in each of the old products; if there is a replacement product for the new product, the replacement product is replaced by the new product in the first scoring matrix, and the score value of the replacement product is replaced by the initial score of the new product to obtain the second scoring matrix; if there is no replacement product for the new product, the adjacent products of the new product are determined in the first scoring matrix, and the adjacent products are the old products with the greatest similarity to the products; the new product and the initial score value of the new product are inserted into adjacent positions of the adjacent products to obtain the second scoring matrix.

[0095] Optionally, the adding module 203 is further used to divide the similarity matrix into a first side similarity matrix and a second side similarity matrix with the adjacent product as a boundary, the first side similarity matrix is ​​located on one side of the adjacent product, and the second side similarity matrix is ​​located on the other side of the adjacent product; based on a preset first product similarity threshold, the product similarities in the first side similarity matrix and the second side similarity matrix that are less than the first product similarity threshold are filtered to obtain valid product similarities greater than or equal to the first product similarity threshold; if the sum of the valid product similarities in the first side similarity matrix is ​​greater than the sum of the valid product similarities in the second side similarity matrix, then the side where the first side similarity matrix is ​​located is determined to be the insertion side of the adjacent product; if the sum of the valid product similarities in the first side similarity matrix is ​​less than the sum of the valid product similarities in the second side similarity matrix, then the other side where the second side similarity matrix is ​​located is determined to be the insertion side of the adjacent product; and the new product and the initial score value of the new product are inserted into the adjacent position of the insertion side of the adjacent product to obtain the second score matrix.

[0096] Optionally, the update module 204 is further used to perform matrix decomposition on the first rating matrix to obtain a first user matrix and a first product matrix, wherein the first user matrix includes first user feature values ​​corresponding to each of the users, and the first product matrix includes first product feature values ​​corresponding to each of the old products; perform matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix, wherein the second user matrix includes second user feature values ​​corresponding to each of the users, and the second product matrix includes second product feature values ​​corresponding to the new products and each of the old products, and the similarity between the second user matrix and the first user matrix is ​​greater than a preset similarity; based on the first product matrix, the second product matrix and the similarity matrix, a third product matrix is ​​calculated, wherein the third product matrix includes third product feature values ​​corresponding to the new products and each of the old products; and matrix multiplication is performed on the second user matrix and the third product matrix to obtain a third rating matrix.

[0097] Optionally, the update module 204 is also used to perform matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix; calculate the metric distance between the second user matrix and the first user matrix; iterate the matrix decomposition process with the minimum metric distance as the optimization goal, and stop the iteration when the metric distance is less than a preset metric distance, so as to obtain the final second user matrix and the final second product matrix.

[0098] Optionally, the updating module 204 is further used to perform dimension expansion of the similarity matrix on a non-product dimension by interpolation to obtain a similarity expansion matrix; perform matrix multiplication calculation on the first product matrix and the similarity expansion matrix to obtain an intermediate matrix; perform matrix multiplication calculation on the intermediate matrix and the second product matrix to obtain a third product matrix.

[0099] Optionally, the device further comprises: A selection module, configured to select, from among all the old commodities, the old commodities whose commodity similarity is greater than the second commodity similarity threshold value as reference commodities based on a preset second commodity similarity threshold value; A third acquisition module is used to acquire the historical sales data of each reference commodity within a preset time period, where the starting time of the preset time period is the listing time of the reference commodity; A determination module, configured to determine the historical score value corresponding to the historical sales data based on the score value-sales data relationship table of each reference commodity, wherein each old commodity corresponds to one score value-sales data relationship table; The calculation module is used to calculate an average historical score value as the initial score value of the new product according to the historical score values ​​of each of the products.

[0100] like Figure 3 As shown, an embodiment of the present invention further provides an electronic device, including a processor, and the processor can execute any one of the above-mentioned mixed recommendation methods for new and used goods.

[0101] Specifically, the method includes a processor 301 and a memory 302, and a computer program for executing a mixed recommendation method for new and used products, which is stored in the memory 302 and can be run on the processor 301, wherein: The processor 301 runs the computer program of the mixed recommendation method of new and used goods stored in the memory 302, and performs the following steps: Obtaining a first rating matrix, where the first rating matrix includes ratings of different users for each used product; Acquire product information of the new product and product information of each of the old products, and calculate product similarity between the new product and each of the old products based on the product information of the new product and the product information of the old products, to obtain a similarity matrix between the new product and the old products, wherein the similarity matrix has the same arrangement order of the old products as the first scoring matrix; Adding the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix; Based on the similarity matrix and the first scoring matrix, the scoring values ​​of the new products and each of the old products in the second scoring matrix are updated to obtain a third scoring matrix; Product recommendations are made based on the third rating matrix.

[0102] Optionally, the adding of the new product and the initial score value of the new product to the first score matrix to obtain a second score matrix performed by the processor 301 includes: Determine whether there is a replacement product for the new product among each of the old products; If there is a replacement product for the new product, the replacement product is replaced by the new product in the first rating matrix, and the rating value of the replacement product is replaced by the initial rating of the new product to obtain the second rating matrix; If there is no replacement product for the new product, then in the first scoring matrix, an adjacent product of the new product is determined, where the adjacent product is the old product with the greatest similarity to the new product; The new product and the initial rating value of the new product are inserted into adjacent positions of the adjacent products to obtain the second rating matrix.

[0103] Optionally, the processor 301 inserts the new product and the initial score value of the new product at the adjacent position of the adjacent product to obtain the second score matrix, including: Taking the adjacent product as a boundary, dividing the similarity matrix into a first-side similarity matrix and a second-side similarity matrix, wherein the first-side similarity matrix is ​​located on one side of the adjacent product, and the second-side similarity matrix is ​​located on the other side of the adjacent product; Based on a preset first product similarity threshold, filtering product similarities in the first-side similarity matrix and the second-side similarity matrix that are less than the first product similarity threshold to obtain effective product similarities that are greater than or equal to the first product similarity threshold; If the sum of similarities of valid products in the similarity matrix on the first side is greater than the sum of similarities of valid products in the similarity matrix on the second side, determining the side where the similarity matrix on the first side is located as the insertion side of the adjacent product; If the sum of similarities of valid commodities in the first-side similarity matrix is ​​less than the sum of similarities of valid commodities in the second-side similarity matrix, determining the other side where the second-side similarity matrix is ​​located as the insertion side of the adjacent commodity; The new product and the initial score value of the new product are inserted into the adjacent position on the insertion side of the adjacent product to obtain the second score matrix.

[0104] Optionally, the processor 301 executes the updating of the score values ​​of the new product and each of the old products in the second scoring matrix based on the similarity matrix and the first scoring matrix to obtain a third scoring matrix, including: Performing matrix decomposition on the first rating matrix to obtain a first user matrix and a first product matrix, wherein the first user matrix includes first user feature values ​​corresponding to each of the users, and the first product matrix includes first product feature values ​​corresponding to each of the old products; Performing matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix, wherein the second user matrix includes second user feature values ​​corresponding to each of the users, and the second product matrix includes second product feature values ​​corresponding to the new products and each of the old products, and the similarity between the second user matrix and the first user matrix is ​​greater than a preset similarity; Based on the first commodity matrix, the second commodity matrix and the similarity matrix, a third commodity matrix is ​​calculated, wherein the third commodity matrix includes the third commodity feature values ​​corresponding to the new commodity and each of the old commodities; Perform matrix multiplication on the second user matrix and the third product matrix to obtain a third rating matrix.

[0105] Optionally, the processor 301 performs matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix, including: By performing matrix decomposition on the second rating matrix, a second user matrix and a second product matrix are obtained; Calculating a metric distance between the second user matrix and the first user matrix; Taking the minimum metric distance as the optimization goal, the matrix decomposition process is iterated, and when the metric distance is less than a preset metric distance, the iteration is stopped to obtain the final second user matrix and the final second product matrix.

[0106] Optionally, the processor 301 calculates a third product matrix based on the first product matrix, the second product matrix, and the similarity matrix, including: By interpolation, the similarity matrix is ​​dimensionally expanded in a non-product dimension to obtain a similarity expansion matrix; Performing matrix multiplication calculation on the first product matrix and the similarity extension matrix to obtain an intermediate matrix; Perform matrix multiplication calculation on the intermediate matrix and the second product matrix to obtain a third product matrix.

[0107] Optionally, before adding the new product and the initial score value of the new product to the first score matrix to obtain the second score matrix, the method executed by the processor 301 further includes: Based on a preset second product similarity threshold, selecting the old products whose product similarity is greater than the second product similarity threshold from among all the old products as reference products; Acquire historical sales data of each of the reference products within a preset time period, where the starting time of the preset time period is the time when the reference product was launched on the market; Based on the score value-sales data relationship table of each reference commodity, determine the historical score value corresponding to the historical sales data, each old commodity corresponds to one score value-sales data relationship table; According to the historical rating values ​​of the respective commodities, an average historical rating value is calculated as the initial rating value of the new commodity.

[0108] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the mixed recommendation method for new and old goods provided by the embodiment of the present invention are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0109] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0110] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A mixed recommendation method for new and old goods, characterized in that: The method comprises the following steps: Obtaining a first rating matrix, where the first rating matrix includes ratings of different users for each used product; Acquire product information of the new product and product information of each of the old products, and calculate product similarity between the new product and each of the old products based on the product information of the new product and the product information of the old products, to obtain a similarity matrix between the new product and the old products, wherein the similarity matrix has the same arrangement order of the old products as the first scoring matrix; Adding the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix; Based on the similarity matrix and the first scoring matrix, the scoring values ​​of the new products and each of the old products in the second scoring matrix are updated to obtain a third scoring matrix; Product recommendations are made based on the third rating matrix.

2. The mixed recommendation method of new and old goods according to claim 1, characterized in that: The step of adding the new product and the initial score value of the new product to the first score matrix to obtain a second score matrix includes: Determine whether there is a replacement product for the new product among each of the old products; If there is a replacement product for the new product, the replacement product is replaced by the new product in the first rating matrix, and the rating value of the replacement product is replaced by the initial rating of the new product to obtain the second rating matrix; If there is no replacement product for the new product, then in the first scoring matrix, an adjacent product of the new product is determined, where the adjacent product is the old product with the greatest similarity to the new product; The new product and the initial rating value of the new product are inserted into adjacent positions of the adjacent products to obtain the second rating matrix.

3. The mixed recommendation method of new and old goods according to claim 2, characterized in that: Inserting the new product and the initial score value of the new product into the adjacent position of the adjacent product to obtain the second score matrix includes: Taking the adjacent product as a boundary, dividing the similarity matrix into a first-side similarity matrix and a second-side similarity matrix, wherein the first-side similarity matrix is ​​located on one side of the adjacent product, and the second-side similarity matrix is ​​located on the other side of the adjacent product; Based on a preset first product similarity threshold, filtering product similarity values ​​in the first-side similarity matrix and the second-side similarity matrix that are less than the first product similarity threshold, to obtain effective product similarities that are greater than or equal to the first product similarity threshold; If the sum of similarities of valid products in the first-side similarity matrix is ​​greater than the sum of similarities of valid products in the second-side similarity matrix, determining the side where the first-side similarity matrix is ​​located as the insertion side of the adjacent product; If the sum of similarities of valid commodities in the first-side similarity matrix is ​​less than the sum of similarities of valid commodities in the second-side similarity matrix, determining the other side where the second-side similarity matrix is ​​located as the insertion side of the adjacent commodity; The new product and the initial score value of the new product are inserted into the adjacent position on the insertion side of the adjacent product to obtain the second score matrix.

4. The mixed recommendation method of new and used goods according to claim 1, characterized in that: The step of updating the score values ​​of the new product and each of the old products in the second scoring matrix based on the similarity matrix and the first scoring matrix to obtain a third scoring matrix includes: Performing matrix decomposition on the first rating matrix to obtain a first user matrix and a first product matrix, wherein the first user matrix includes first user feature values ​​corresponding to each of the users, and the first product matrix includes first product feature values ​​corresponding to each of the old products; Performing matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix, wherein the second user matrix includes second user feature values ​​corresponding to each of the users, and the second product matrix includes second product feature values ​​corresponding to the new products and each of the old products, and the similarity between the second user matrix and the first user matrix is ​​greater than a preset similarity; Based on the first commodity matrix, the second commodity matrix and the similarity matrix, a third commodity matrix is ​​calculated, wherein the third commodity matrix includes the third commodity feature values ​​corresponding to the new commodity and each of the old commodities; Perform matrix multiplication on the second user matrix and the third product matrix to obtain a third rating matrix.

5. The mixed recommendation method of new and old goods according to claim 4, characterized in that: The performing matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix includes: Performing matrix decomposition on the second rating matrix to obtain a second user matrix and a second product matrix; Calculating a metric distance between the second user matrix and the first user matrix; Taking the minimum metric distance as the optimization goal, the matrix decomposition process is iterated, and when the metric distance is less than a preset metric distance, the iteration is stopped to obtain the final second user matrix and the final second product matrix.

6. The mixed recommendation method of new and old goods according to claim 4, characterized in that: The calculating a third product matrix based on the first product matrix, the second product matrix and the similarity matrix includes: By interpolation, the similarity matrix is ​​dimensionally expanded in a non-product dimension to obtain a similarity expansion matrix; Performing matrix multiplication calculation on the first product matrix and the similarity extension matrix to obtain an intermediate matrix; Perform matrix multiplication calculation on the intermediate matrix and the second product matrix to obtain a third product matrix.

7. The mixed recommendation method for new and used goods according to any one of claims 1 to 6, characterized in that: Before adding the new product and the initial rating value of the new product to the first rating matrix to obtain the second rating matrix, the method further includes: Based on a preset second product similarity threshold, selecting the old products whose product similarity is greater than the second product similarity threshold from among all the old products as reference products; Acquire historical sales data of each of the reference products within a preset time period, where the starting time of the preset time period is the launch time of the reference product; Based on the score value-sales data relationship table of each reference commodity, determine the historical score value corresponding to the historical sales data, each old commodity corresponds to one score value-sales data relationship table; According to the historical rating values ​​of each of the old commodities, an average historical rating value is calculated as the initial rating value of the new commodity.

8. A device for recommending mixed new and old goods, characterized in that: The mixed recommendation device for new and old goods includes: A first acquisition module is used to acquire a first rating matrix, wherein the first rating matrix includes ratings of different users for each used product; A second acquisition module is used to acquire product information of the new product and product information of each of the old products, and based on the product information of the new product and the product information of the old products, calculate the product similarity between the new product and each of the old products to obtain a similarity threshold between the new product and the old products, wherein the similarity matrix has the same arrangement order of the old products as the first scoring matrix; An adding module, configured to add the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix; An updating module, configured to update the score values ​​of the new products and each of the old products in the second scoring matrix based on the similarity matrix and the first scoring matrix, to obtain a third scoring matrix; A product recommendation module is used to make product recommendations based on the third rating matrix.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps in the mixed recommendation method for new and used goods as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the mixed recommendation method for new and used goods as described in any one of claims 1 to 7.

Images