Hybrid Recommendation Method, Device, Electronic Device and Storage Medium for New and Old Products

Abstract

The present invention provides a method for mixed recommendation of new and old commodities, which method comprises: obtaining the commodity information of new commodities and the commodity information of each old commodity, and calculating the commodity similarity between each new commodity and each old commodity based on the commodity information of the new commodities and the commodity information of the old commodities, so as to obtain a similarity matrix between the new commodities and the old commodities, wherein the similarity matrix has the same arrangement order of old commodities as that of the first scoring matrix; adding the new commodities and the initial scoring values of the new commodities to the first scoring matrix to obtain a second scoring matrix; updating the scoring values of the new commodities and each old commodity in the second scoring matrix based on the similarity matrix and the first scoring matrix to obtain a third scoring matrix; and performing commodity recommendation based on the third scoring matrix. The present invention can perform mixed recommendation on new commodities and old commodities, thereby improving the shopping experience of users.

Description

Technical Field

[0001] The present invention relates to the field of product recommendation, and particularly to a method, device, electronic device and storage medium for hybrid recommendation of new and used products. Background Art

[0002] With the increasing personalized needs of people for items, various item recommendation methods have emerged. Most item recommendations are based on the user's behavior towards items, and through some mathematical algorithms, the items that the user likes are predicted and recommended to the user. Currently, existing recommendation methods mainly use the user's behavior towards products for recommendation. However, for new products, there is no user behavior towards new products. Therefore, in most cases, only brute-force recommendations can be made for new products. Regardless of whether the user wants to purchase the new product, the new product will be recommended, reducing the user's shopping experience. Summary of the Invention

[0003] An embodiment of the present invention provides a method for hybrid recommendation of new and used products, which can achieve hybrid recommendation of new and used products and improve the personalization degree and accuracy of recommendation. By obtaining a first rating matrix, obtaining the product information of new products and the product information of each used product, and calculating the product similarity between the new product and each used product based on the product information of the new product and the product information of the used product, a similarity matrix between the new product and the used product is obtained. The similarity matrix has the same arrangement order of used products as the first rating matrix. Add the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix, and update the rating values of the new product and each used product in the second rating matrix through the similarity matrix and the first rating matrix to obtain a third rating matrix, and use the third rating matrix for product recommendation, which can perform hybrid recommendation of new products and used products and improve the user's shopping experience.

[0004] In a first aspect, an embodiment of the present invention provides a method for hybrid recommendation of new and used products, the method comprising the following steps:

[0005] Obtain a first rating matrix, the first rating matrix including the rating values of different users for each used product;

[0006] Obtain the product information of the new product and the product information of each of the used products, and calculate the product similarity between the new product and each of the used products based on the product information of the new product and the product information of the used product, to obtain a similarity matrix between the new product and the used product, the similarity matrix having the same arrangement order of used products as the first rating matrix;

[0007] Add the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix;

[0008] Based on the similarity matrix and the first scoring matrix, update the scoring values of the new product and each of the old products in the second scoring matrix to obtain a third scoring matrix;

[0009] Perform product recommendation based on the third scoring matrix.

[0010] Optionally, adding the new product and the initial scoring value of the new product to the first scoring matrix to obtain a second scoring matrix includes:

[0011] Among each of the old products, determine whether there is a replacement product for the new product;

[0012] If there is a replacement product for the new product, then in the first scoring matrix, replace the replacement product with the new product and replace the scoring value of the replacement product with the initial score of the new product to obtain the second scoring matrix;

[0013] If there is no replacement product for the new product, then in the first scoring matrix, determine the adjacent products of the new product, and the adjacent products are the old products with the largest product similarity;

[0014] Insert the new product and the initial scoring value of the new product at the adjacent positions of the adjacent products to obtain the second scoring matrix.

[0015] Optionally, inserting the new product and the initial scoring value of the new product at the adjacent positions of the adjacent products to obtain the second scoring matrix includes:

[0016] Taking the adjacent product as the boundary, divide the similarity matrix into a first-side similarity matrix and a second-side similarity matrix, where the first-side similarity matrix is on one side of the adjacent product and the second-side similarity matrix is on the other side of the adjacent product;

[0017] Based on a preset first product similarity threshold, filter the product similarities less than the first product similarity threshold in the first-side similarity matrix and the second-side similarity matrix to obtain valid product similarities greater than or equal to the first product similarity threshold;

[0018] 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 determine the side where the first-side similarity matrix is located as the insertion side of the adjacent product;

[0019] If the sum of the effective product similarities in the first - side similarity matrix is less than the sum of the effective product similarities in the second - side similarity matrix, determine the other side where the second - side similarity matrix is located as the insertion side of the adjacent product;

[0020] Insert the new product and the initial score value of the new product at the adjacent position on the insertion side of the adjacent product to obtain the second score matrix.

[0021] Optionally, updating the score values of the new product and each of the old products in the second score matrix based on the similarity matrix and the first score matrix to obtain a third score matrix includes:

[0022] Perform matrix factorization on the first score matrix to obtain a first user matrix and a first product matrix. The first user matrix includes the first user eigenvalue corresponding to each user, and the first product matrix includes the first product eigenvalue corresponding to each of the old products;

[0023] Perform matrix factorization on the second score matrix to obtain a second user matrix and a second product matrix. The second user matrix includes the second user eigenvalue corresponding to each user, and the second product matrix includes the second product eigenvalue corresponding to the new product and each of the old products. The similarity between the second user matrix and the first user matrix is greater than a preset similarity;

[0024] Based on the first product matrix, the second product matrix, and the similarity matrix, calculate a third product matrix. The third product matrix includes the second product eigenvalue corresponding to the new product and each of the old products;

[0025] Perform matrix multiplication on the second user matrix and the third product matrix to obtain a third score matrix.

[0026] Optionally, the performing matrix factorization on the second score matrix to obtain a second user matrix and a second product matrix includes:

[0027] Perform matrix factorization on the second score matrix to obtain a second user matrix and a second product matrix;

[0028] Calculate the metric distance between the second user matrix and the first user matrix;

[0029] Taking the minimum of the metric distance as the optimization goal, iterate the process of matrix factorization. When the metric distance is less than the preset metric distance, stop the iteration to obtain the final second user matrix and the final second product matrix.

[0030] Optionally, calculating a third product matrix based on the first product matrix, the second product matrix, and the similarity matrix includes:

[0031] Performing dimensionality expansion on the similarity matrix in non-product dimensions through interpolation to obtain a similarity expansion matrix;

[0032] Performing matrix multiplication on the first product matrix and the similarity expansion matrix to obtain an intermediate matrix;

[0033] Performing matrix multiplication on the intermediate matrix and the second product matrix to obtain a third product matrix.

[0034] Optionally, before adding the new product and its initial score value to the first score matrix to obtain a second score matrix, the method further includes:

[0035] Based on a preset second product similarity threshold, selecting, from all the old products, the old products with a product similarity greater than the second product similarity threshold as reference products;

[0036] Obtaining the historical sales data of each reference product within a preset time period, where the start time of the preset time period is the launch time of the reference product;

[0037] Based on the score-sales data relationship table of each reference product, determining the historical score value corresponding to the historical sales data, and each old product corresponds to a score-sales data relationship table;

[0038] Calculating the average historical score value as the initial score value of the new product according to the historical score values of each old product.

[0039] In a second aspect, an embodiment of the present invention further provides a hybrid recommendation device for new and old products, where the hybrid recommendation device for new and old products includes:

[0040] A first acquisition module, configured to acquire a first score matrix, where the first score matrix includes score values of different users for each old product;

[0041] A second acquisition module, configured to acquire the product information of the new product and the product information of each old product, and calculate 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, to obtain the similarity threshold between the new product and the old product, and the similarity matrix and the first score matrix have the same arrangement order of old products;

[0042] 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;

[0043] An updating module, configured to update the rating values of the new product and each of the old products in the second rating matrix based on the similarity matrix and the first rating matrix to obtain a third rating matrix;

[0044] A product recommendation module, configured to perform product recommendation based on the third rating matrix.

[0045] In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, the steps in the hybrid recommendation method for new and old products provided by the embodiment of the present invention are implemented.

[0046] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the hybrid recommendation method for new and old products provided by the embodiment of the invention are implemented.

[0047] In the embodiment of the present invention, a first rating matrix is obtained, where the first rating matrix includes the rating values of different users for each old product; the product information of the new product and the 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 and the first rating matrix have the same arrangement order of old products; 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 recommendation is performed based on the third rating matrix. By obtaining the first rating matrix, obtaining the product information of the new product and the product information of each old product, calculating 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, adding the new product and the initial rating value of the new product to the first rating matrix to obtain a second rating matrix, updating the rating values of the new product and each old product in the second rating matrix through the similarity matrix and the first rating matrix to obtain a third rating matrix, and using the third rating matrix for product recommendation, the present invention can perform hybrid recommendation on new products and old products, improving the shopping experience of users. Description of the Drawings

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0049] Figure 1 is a flowchart of a method for hybrid recommendation of new and old commodities provided by an embodiment of the present invention;

[0050] Figure 2 is a schematic structural diagram of a device for hybrid recommendation of new and old commodities provided by an embodiment of the present invention;

[0051] Figure 3 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0053] As Figure 1 shown, Figure 1 is a flowchart of a method for hybrid recommendation of new and old commodities provided by an embodiment of the present invention. The method for hybrid recommendation of new and old commodities includes the following steps:

[0054] 101. Obtain a first rating matrix.

[0055] In the embodiments of the present invention, the above method for hybrid recommendation of new and old commodities can be applied to a commodity recommendation platform. The above commodity recommendation platform can be built based on a server or a distributed server. The above commodity recommendation platform includes a device interface (for data collection devices to upload the collected data), an algorithm container interface, a database, and a commodity recommendation program. The above device interface can be used to obtain the data of the data collection device (including the device identifier), the above algorithm container interface is used to obtain the data of the algorithm container (including the connected device identifier), the above database is used to store the association relationship between the device identifier and the task inference coefficient, and the above commodity recommendation program is used for task creation, commodity recommendation, and connecting the data collection device corresponding to the device identifier to the corresponding algorithm container. The above database can be a relational database, such as a mysql database. The above data collection device can be a data collection device such as a camera or a camera.

[0056] The above first rating matrix includes the rating values of different users for each old product. The above old products can be understood as existing products, representing products that already have rating data.

[0057] The above rating matrix can be represented in tabular form. Each row represents a user, each column represents a product, and each cell in the matrix contains the rating value of the corresponding user for the corresponding product. Each row in the rating matrix records the ratings of different users for each product.

[0058] For example, there are M users and N products, and the first rating matrix is as follows:

[0059]

[0060] Among them, the rating values of user 1 for products 1, 2, 3,..., N are 4.0, 4.3, 3.5,..., 4.3 respectively; the rating values of user 2 for products 1, 2, 3,..., N are 3.5, 4.7, 4.8,..., 4.2 respectively; the rating values of user 3 for products 1, 2, 3,..., N are 3.6, 4.3, 4.6,..., 3.9 respectively; the rating values of user M for products 1, 2, 3,..., N are 4.5, 4.5, 3.6,..., 4.5 respectively. The above rating values can be set within the range of 1-5. The rating value represents the degree of evaluation of the user for the product. A higher rating value indicates that the user is more satisfied with the product, while a lower rating value indicates that the user is less satisfied with the product. Specifically, by analyzing the rating matrix, the evaluation trend of users for different products can be understood.

[0061] 102. Obtain the product information of the new product and the product information of each old product, and calculate 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, to obtain the similarity matrix between the new product and the old product.

[0062] In the embodiment of the present invention, the above similarity matrix has the same arrangement order of old products as the first rating matrix. This means that the similarity matrix between the new product and the old products is sorted according to the arrangement order of the old products, that is, sorted according to the arrangement order of products 1, 2, 3,..., N, which can ensure that the similarity matrix and the first rating matrix can be corresponding, so as to ensure the consistency and comparability of the data, making the subsequent analysis or processing more accurate and reliable.

[0063] The above product information includes product name, brand, price, category, description, picture, etc.

[0064] Further, a similarity calculation method can be used to calculate the product similarity between the new product and each old product. The above similarity calculation method can be a text similarity calculation method, a cosine similarity calculation method, etc.

[0065] The above similarity matrix can show the product similarity between the new product and each old product, and the above product similarity is a value between 0 and 1. If two products show a high degree of similarity in content, category, or user behavior, then their corresponding values in the similarity matrix will be higher.

[0066] The above similarity matrix can be shown as follows:

[0067]

[0068] 103. Add the new product and the initial score value of the new product to the first score matrix to obtain the second score matrix.

[0069] In the embodiments of the present invention, the initial score value of the above new product can be understood as the score set for the new product when adding the new product to the score matrix. The initial score can be any reasonable value. For example, if a product recommendation system is being established and a new product wants to be added to the system. An initial score can be set for the new product, and this score can be set based on factors such as the name, brand, price, category, etc. of the product, or can be set based on the historical sales data of the old products similar to the new product.

[0070] The above second score matrix includes the score values of different users for each old product, the new product, and the initial score values of different users for the new product.

[0071] For example, assume there is a first score matrix containing several products and their scores. Now, a new product needs to be added and an initial score is set for the new product. In order to take the new product and its score into consideration, the original first score matrix needs to be expanded to form a second score matrix. The number of new products can be one or more.

[0072] For example, if there was initially a first score matrix containing 10 products and their scores, and later 5 new products were discovered and initial scores were set for them respectively. Then, these 5 new products and their scores can be added to the original first score matrix to obtain a second score matrix containing 15 products and their scores.

[0073] 104. Based on the similarity matrix and the first score matrix, update the score values of the new product and each old product in the second score matrix to obtain the third score matrix.

[0074] In the embodiments of the present invention, considering the addition of new products, it will affect the rating values of users for different old products. This kind of influence varies. Specifically, the older products that are more similar to the new product are more affected. Therefore, according to the similarity matrix and the first rating matrix, the rating values of the new product and each old product in the second rating matrix can be updated, so that the rating values of the new product and the old products are re-predicted.

[0075] The first rating matrix can be matrix decomposed to obtain a first user matrix and a first product matrix. The second rating matrix can be matrix decomposed to obtain a second user matrix and a second product matrix. And according to the first product matrix, the second product matrix and the similarity matrix, a third product matrix is calculated through a similarity calculation method. Then, matrix multiplication is performed on the second user matrix and the third product matrix to obtain a third rating matrix.

[0076] The above matrix decomposition can be understood as a processing 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 as to more easily discover the potential characteristics of users and products, which helps to improve the accuracy and efficiency of the recommendation system.

[0077] Through matrix decomposition and matrix multiplication, the above third rating matrix integrates the product feature information of the second product matrix and the third product matrix, enabling the new product and the old products to be integrated in the dimension of product features, and predicting the rating values of each user for the new product. The third rating matrix can reflect the degree of preference or satisfaction of users for the new product and the old products.

[0078] In a possible embodiment, the above first rating matrix is a matrix of M×N, the above similarity matrix is a matrix of T×N, and the above second rating matrix is a matrix of M×(N + T). A linear matrix of T×(N + T) can be used to perform a linear transformation on the transposed matrix N×T of the similarity 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 the matrix multiplication of the transposed matrix N×T of the similarity matrix and the linear matrix T×(N + T). The linearly transformed similarity matrix N×(N + T) is a matrix affected by product similarity. The linearly transformed similarity matrix N×(N + T) is normalized to obtain a normalized similarity matrix N×(N + T). The first rating matrix M×N is multiplied by the transposed matrix N×(N + T) of the normalized similarity matrix to obtain a rating update matrix M×(N + T). Subtracting the corresponding rating update value of the rating update matrix M×(N + T) from the rating value in the third rating matrix M×(N + T), the third rating matrix M×(N + T) can be obtained.

[0079] 105. Perform product recommendations based on the third scoring matrix.

[0080] In an embodiment of the present invention, product recommendations can be made according to the third scoring matrix.

[0081] Specifically, through the third scoring matrix, predict the degree of preference of users for new products, so as to make product recommendations, so that new products can be recommended together with old products, giving users more choices and improving the shopping experience of users.

[0082] In an embodiment of the present invention, the present invention can analyze the historical purchase records, evaluated products and scoring situations of users, combine the attributes of the products themselves and the scoring data of other users, and use a recommendation algorithm to calculate which products are more in line with the interests and needs of the current user, so as to perform personalized recommendations. For example, if a user has frequently purchased and given positive reviews to a certain type of product before, then the system can infer that the user may also be interested in similar or related products, and then screen out products that match the user's characteristics from numerous products for recommendation, which can improve the accuracy of product recommendations and user satisfaction.

[0083] In an embodiment of the present invention, obtain the first scoring matrix, where the first scoring matrix includes the scoring values of different users for each old product; obtain the product information of the new product and the product information of each old product, and calculate 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, to obtain the similarity matrix between the new product and the old product, and the similarity matrix has the same arrangement order of old products as the first scoring matrix; add the new product and the initial scoring value of the new product to the first scoring matrix to obtain the second scoring matrix; update the scoring values of the new product 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; perform product recommendations based on the third scoring matrix. The present invention obtains the first scoring matrix, obtains the product information of the new product and the product information of each old product, 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, obtains the 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 the second scoring matrix, 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 the third scoring matrix, and uses the third scoring matrix to perform product recommendations, which can perform mixed recommendations on new products and old products and improve the shopping experience of users.

[0084] It can be understood that in the specific implementation of this application, data related to commodity 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 needs to be obtained, and the collection, use, and processing of relevant data, as well as the training, deployment, and invocation of algorithm models, all need to comply with relevant laws, regulations, and standards in relevant countries and regions.

[0085] Optionally, in the step of adding a new commodity and its initial score value to the first score matrix to obtain the second score matrix, it is possible to determine whether there is a replacement commodity for the new commodity among the old commodities; if there is a replacement commodity for the new commodity, then replace the replacement commodity with the new commodity in the first score matrix, and replace the score value of the replacement commodity with the initial score of the new commodity to obtain the second score matrix; if there is no replacement commodity for the new commodity, then in the first score matrix, determine the adjacent commodities of the new commodity; insert the new commodity and its initial score value at the adjacent positions of the adjacent commodities to obtain the second score matrix.

[0086] In the embodiments of the present invention, the above-mentioned adjacent commodities are the old commodities with the greatest similarity to the new commodity, and the new commodity can be added to the adjacent positions of the adjacent commodities.

[0087] The above-mentioned old commodities can be understood as commodities that already exist and have been evaluated by users.

[0088] The above-mentioned replacement commodities can be understood as commodities that are similar or identical to the new commodity in terms of function, use, or attributes. The new commodity is introduced to replace this old commodity, and the old commodity as the replacement commodity is no longer sold. Therefore, the old commodity can be replaced with the new commodity. In a possible embodiment, after replacing the old commodity as the replacement commodity in the first score matrix with the new commodity, the score value of the old commodity can be used as the initial score value of the new commodity, that is, the new commodity inherits the score value of the old commodity to obtain the second score matrix. At this time, no new commodity column will be added to the second score matrix.

[0089] Specifically, it is possible to determine whether there is an alternative commodity in the old commodities that can replace the new commodity by analyzing the characteristics, categories, brands, etc. of the commodities. If there is a replacement commodity for the new commodity, then replace the score of the replacement commodity with the initial score of the new commodity in the first score matrix. It can be understood that, for example, there is a score matrix that contains various commodities and their corresponding scores. If in this matrix, one commodity can be replaced by another new commodity, then the score of the replaced commodity should be replaced with the score of the new commodity. In this way, a new score matrix is obtained, which contains the information and scores of the new commodity.

[0090] If there is no replacement product for the new product, it is necessary to find the old product with the highest similarity to the new product in the first rating matrix as the adjacent product of the new product. Specifically, methods such as cosine similarity and Pearson correlation coefficient can be used to calculate the product similarity between the new product and the old product, and the old product with the highest product similarity is used as the adjacent product of the new product. It is also possible to directly find the old product with the highest product similarity in the similarity matrix as the adjacent product of the new product.

[0091] Furthermore, a new rating matrix can be constructed by inserting the new product and its initial rating value at adjacent product positions.

[0092] It should be noted that first, among all the old products, it is determined whether there is a replacement product for the new product. If there is an old product that is the replacement product for the new product, then in the first rating matrix, the replacement product is replaced with the new product, and the rating value of the replacement product is replaced with the initial rating of the new product to obtain the second rating 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 rating value at the adjacent position of the adjacent product to obtain the second rating matrix.

[0093] Optionally, in the step of inserting the new product and its initial rating value at the adjacent position of the adjacent product to obtain the second rating matrix, with the adjacent product as the boundary, the similarity matrix is divided into the first-side similarity matrix and the second-side similarity matrix; based on a preset first product similarity threshold, the product similarity values less than the first product similarity threshold in the first-side similarity matrix and the second-side similarity matrix are filtered to obtain the effective product similarity greater than or equal to the first product similarity threshold; if the sum of the effective product similarities in the first-side similarity matrix is greater than the sum of the effective product similarities in the second-side similarity matrix, it is determined that the side where the first-side similarity matrix is located is the insertion side of the adjacent product; if the sum of the effective product similarities in the first-side similarity matrix is less than the sum of the effective product similarities in the second-side similarity matrix, it is determined that the other side where the second-side similarity matrix is located is the insertion side of the adjacent product; at the adjacent position on the insertion side of the adjacent product, the new product and its initial rating value are inserted to obtain the second rating matrix.

[0094] In the embodiments of the present invention, the above similarity threshold is the similarity threshold between the new product and the old product.

[0095] The above first-side similarity matrix is located on one side of the adjacent product, and the above second-side similarity matrix is located on the other side of the adjacent product.

[0096] The above preset first product similarity threshold is the first product similarity threshold preset by the system, which is used to filter and screen product similarity data.

[0097] 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, and these two parts respectively represent the product similarity situations on both sides of the adjacent products; then, a 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 similarities; then, the sums of these two parts of valid product similarities are compared. If the sum of the valid similarities in the first-side similarity matrix is greater than that in the second side, it is considered that the first side is the insertion side of the adjacent product; otherwise, the second side is the insertion side; finally, a new product and its initial score value are inserted at the adjacent position on the insertion side, so as to generate a new score matrix, that is, the second score matrix.

[0098] Optionally, in the step of updating the score values of the new product and each old product in the second score matrix based on the similarity matrix and the first score matrix to obtain the third score matrix, the first score matrix can be factorized to obtain the first user matrix and the first product matrix; the second score matrix is factorized to obtain the second user matrix and the second product matrix; based on the first product matrix, the second product matrix and the similarity matrix, the third product matrix is calculated; the second user matrix and the third product matrix are multiplied to obtain the third score matrix.

[0099] In the embodiment of the present invention, the above-mentioned first user matrix includes the first user eigenvalue corresponding to each user, and the above-mentioned first product matrix includes the first product eigenvalue corresponding to each old product. For example, factorizing the first score matrix M×N can obtain the first user matrix M×U and the first product matrix U×N.

[0100] The above-mentioned first user matrix M×U can be shown as follows:

[0101]

[0102] Among them, the first user eigenvalue corresponding to user 1 is u 1,1 、u 1,2 、u 1,3 、……、u 1,U , and the first user eigenvalue corresponding to user 2 is u 2,1 、u 2,2 、u 2,3 、……、u 2,U , and so on.

[0103] The above-mentioned first product matrix U×N can be shown as follows:

[0104]

[0105] Among them, the first commodity eigenvalue corresponding to Commodity 1 is v 1,1 , v 2,1 , v 3,1 , ……, v u,1 , and the first commodity eigenvalue corresponding to Commodity 2 is v 1,1 , v 2,2 , v 3,2 , ……, v u,2 , and so on.

[0106] Similarly, by performing matrix factorization on the second rating matrix M×(N + T), the second user matrix M×U and the second commodity matrix U×(N + T) can be obtained.

[0107] In a possible embodiment, considering that the new commodity affects the old commodity pair, while the new commodity has a minimal impact on users, therefore, the change in user eigenvalues can be not considered, and the first user matrix M×U and the second commodity matrix U×(N + T) can be directly multiplied to obtain a new rating matrix M×(N + T). Interpolation is performed on the column corresponding to the new commodity in the similarity matrix T×N, and the interpolation value of the column is 1 (i.e., the commodity similarity is 1) to obtain the first new similarity matrix T×(N + T). Then, interpolation is performed on the row of the first new similarity matrix T×(N + T), and the interpolation value of the row is 0 (i.e., the commodity similarity is 0) to obtain the second new similarity matrix (N + T)×(N + T). The new rating matrix M×(N + T) and the second new similarity matrix (N + T)×(N + T) are multiplied to obtain the third rating matrix M×(N + T).

[0108] The above-mentioned second user matrix includes the second user eigenvalues corresponding to each user, the above-mentioned second commodity matrix includes the new commodity and the second commodity eigenvalues corresponding to each old commodity, and the similarity between the second user matrix and the first user matrix is greater than a preset similarity. It should be noted that since the addition of the new commodity has a small impact on users, therefore, by the similarity between the second user matrix and the first user matrix being greater than the preset similarity, when performing matrix factorization on the second rating matrix, the change in user eigenvalues can be limited to be small, which can more prominently show the change in the commodity eigenvalue dimension between the new commodity and the old commodity.

[0109] The above-mentioned third commodity matrix includes the new commodity and the third commodity eigenvalues corresponding to each old commodity.

[0110] The above-mentioned matrix factorization can be understood as a processing process of splitting a complex matrix into the product of multiple simple matrices. The purpose of matrix factorization is to reduce the original high-dimensional rating matrix to a low-rank matrix, so as to more easily discover the potential characteristics of users and commodities, which helps to improve the accuracy and efficiency of the recommendation system.

[0111] The above matrix multiplication calculation is an operation method for multiplying two matrices. It can be understood that for a matrix A of M×N and a matrix B of N×P, their product is a matrix C of M×P. Each element c(i, j) in matrix C is the sum of the products of the corresponding elements in the i-th row of A and the j-th column of B. Matrix multiplication is an operation for multiplying two matrices.

[0112] Each element in the third scoring matrix is obtained by performing a dot product operation on the corresponding user feature vector in the second user matrix and the corresponding product feature vector in the third product matrix.

[0113] Optionally, in the step of performing matrix factorization on the second scoring matrix to obtain the second user matrix and the second product matrix, the second scoring matrix can be factorized to obtain the second user matrix and the second product matrix; calculate the metric distance between the second user matrix and the first user matrix; with the minimum metric distance as the optimization goal, iterate the matrix factorization process, and stop the iteration when the metric distance is less than the preset metric distance to obtain the final second user matrix and the final second product matrix.

[0114] In the embodiment of the present invention, the above matrix factorization is a processing process of splitting a complex matrix into the product of multiple simple matrices.

[0115] The above second user matrix includes second user eigenvalues corresponding to each user.

[0116] The above second product matrix includes new products and second product eigenvalues corresponding to each old product.

[0117] The above metric distance can be understood as a quantization index of the similarity or difference degree between two vectors. The metric distance is used to measure the difference degree 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 and the second user matrix are.

[0118] Furthermore, by iteratively optimizing the matrix factorization process, the metric distance can be gradually reduced until it reaches the preset threshold. At this time, it is considered that the difference between the two matrices is small enough, and the iteration can be stopped to obtain the final second user matrix and the final second product matrix.

[0119] Optionally, in the step of calculating the third commodity matrix based on the first commodity matrix, the second commodity matrix, and the similarity matrix, the similarity matrix can be dimensionally extended on the non-commodity dimension by interpolation to obtain a similarity extended matrix; the first commodity matrix is multiplied by the similarity extended matrix through matrix multiplication to obtain an intermediate matrix; the intermediate matrix and the second commodity matrix are multiplied through matrix multiplication to obtain the third commodity matrix.

[0120] In the embodiments of the present invention, the above-mentioned interpolation method can be understood as interpolating the columns and rows of the similarity matrix, thereby extending the dimension of the similarity matrix. For example, interpolation is performed on the columns corresponding to new commodities in the similarity matrix T×N, and the interpolation value for the columns is 1 (i.e., the commodity similarity is 1), obtaining the first new similarity matrix T×(N+T), and then interpolation is performed on the rows of the first new similarity matrix T×(N+T), and the interpolation value for the rows is 0 (i.e., the commodity similarity is 0), obtaining the second new similarity matrix (N+T)×(N+T).

[0121] In the similarity matrix T×N, the commodity dimension is the column, and the non-commodity dimension is the row.

[0122] The above-mentioned dimensional extension can be understood as an extension process of extending the similarity matrix from a single dimension to multiple dimensions.

[0123] The above-mentioned matrix multiplication calculation is a process of multiplying the corresponding elements of two matrices and then summing. Specifically, each element of the result matrix is obtained by summing the products of the elements corresponding to the rows of the first matrix and the columns of the second matrix.

[0124] Specifically, after obtaining the first commodity matrix U×N and the second commodity matrix U×(N+T), the similarity matrix T×N can be dimensionally extended on the non-commodity dimension (row) (interpolating the rows of the similarity matrix T×N, and the interpolation value for the rows is 0), obtaining the similarity extended matrix U×N. The first commodity matrix U×N is multiplied by the transposed matrix N×U of the similarity extended matrix through matrix multiplication to obtain the intermediate matrix U×U. The intermediate matrix U×U and the second commodity matrix U×(N+T) are multiplied through matrix multiplication to obtain the third commodity matrix U×(N+T).

[0125] For example, there is an original similarity matrix. The original similarity matrix may only consider the direct similarity between users and commodities. By interpolating the non-commodity dimension (such as rows) of the similarity matrix, a similarity extended matrix can be created, which considers the direct similarity between commodities.

[0126] The above-mentioned intermediate matrix integrates the commodity information of the first commodity matrix and the similarity information in the similarity extended matrix.

[0127] The above-mentioned third product matrix integrates the similarity information of the second product matrix and the intermediate matrix.

[0128] Optionally, before the step of adding a new product and its initial score value to the first score matrix to obtain the second score matrix, old products with a product similarity greater than the second product similarity threshold can be selected from all old products as reference products based on a preset second product similarity threshold; historical sales data of each reference product within a preset time period can be obtained; based on the score value - sales data relationship table of each reference product, the historical score value corresponding to the historical sales data can be determined; and according to the historical score values of each old product, the average historical score value can be calculated as the initial score value of the new product.

[0129] In the embodiments of the present invention, the above-mentioned preset second product similarity threshold is a similarity threshold preset by the system, which is used to measure the similarity degree between products, and can help the system determine which old products have a high similarity with the new product, and use the old products with a product similarity greater than the second product similarity threshold as reference products.

[0130] The start 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 after launch, etc.

[0131] The above-mentioned historical sales data can be understood as the sales records of each reference product in the past period of time, including information such as sales volume and sales amount.

[0132] Each old product corresponds to a score value - sales data relationship table. The above-mentioned score value - sales data relationship table records the relationship between each product and its score value, and is a relationship table used to convert sales data into score values.

[0133] It should be noted that in the present invention, old products are screened by using a preset second product similarity threshold, and old products with a product similarity exceeding the preset second product similarity threshold are selected as reference products. Such screening helps to ensure that the reference products and the new product have a high similarity in attributes, categories or other aspects, thereby improving the accuracy of subsequent score calculation. For each selected reference product, its sales data within the preset time period is collected. For each reference product, its score value - sales data relationship table is used to determine the historical score value corresponding to its historical sales data. The historical score values of all selected reference products are averaged to obtain an average score value, and this average score value will be used as the initial score value of the new product.

[0134] As Figure 2 shown, the embodiments of the present invention provide a hybrid recommendation device for new and old products, and the hybrid recommendation device for new and old products includes:

[0135] The first acquisition module 201 is configured to acquire a first scoring matrix, where the first scoring matrix includes scoring values of different users for each old product.

[0136] The second acquisition module 202 is configured to acquire product information of a new product and product information of each of the old products, and calculate a 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, where the similarity matrix has the same arrangement order of the old products as the first scoring matrix.

[0137] The addition module 203 is configured to add the new product and an initial scoring value of the new product to the first scoring matrix to obtain a second scoring matrix.

[0138] The update module 204 is configured to update scoring 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.

[0139] The product recommendation module 205 is configured to perform product recommendation based on the third scoring matrix.

[0140] Optionally, the addition module 203 is further configured to determine whether there is a replacement product of the new product among the old products; if there is a replacement product of the new product, replace the replacement product with the new product in the first scoring matrix, and replace the scoring value of the replacement product with the initial scoring of the new product to obtain the second scoring matrix; if there is no replacement product of the new product, determine an adjacent product of the new product in the first scoring matrix, where the adjacent product is the old product with the largest product similarity; insert the new product and the initial scoring value of the new product at an adjacent position of the adjacent product to obtain the second scoring matrix.

[0141] Optionally, the adding module 203 is further configured to divide the similarity matrix into a first-side similarity matrix and a second-side similarity matrix with the adjacent product as the 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, filter the product similarities less than the first product similarity threshold in the first-side similarity matrix and the second-side similarity matrix 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, determine the side where the first-side similarity matrix is located as 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, determine the other side where the second-side similarity matrix is located as the insertion side of the adjacent product. Insert the new product and the initial score value of the new product at the adjacent position on the insertion side of the adjacent product to obtain the second score matrix.

[0142] Optionally, the updating module 204 is further configured to perform matrix decomposition on the first score matrix to obtain a first user matrix and a first product matrix. The first user matrix includes first user eigenvalues corresponding to each user, and the first product matrix includes first product eigenvalues corresponding to each old product. Perform matrix decomposition on the second score matrix to obtain a second user matrix and a second product matrix. The second user matrix includes second user eigenvalues corresponding to each user, and the second product matrix includes the new product and second product eigenvalues corresponding to each old product. 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, calculate a third product matrix. The third product matrix includes the new product and third product eigenvalues corresponding to each old product. Perform matrix multiplication calculation on the second user matrix and the third product matrix to obtain a third score matrix.

[0143] Optionally, the updating module 204 is further configured to perform matrix decomposition on the second score 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; use the minimum metric distance as the optimization goal, iterate the matrix decomposition process, and stop the iteration when the metric distance is less than a preset metric distance to obtain the final second user matrix and the final second product matrix.

[0144] Optionally, the update module 204 is further configured to perform dimensionality expansion on the similarity matrix in the non-commodity dimension by interpolation to obtain a similarity expansion matrix; perform matrix multiplication on the first commodity matrix and the similarity expansion matrix to obtain an intermediate matrix; perform matrix multiplication on the intermediate matrix and the second commodity matrix to obtain a third commodity matrix.

[0145] Optionally, the apparatus further includes:

[0146] A selection module, configured to select, from all the old commodities, the old commodities whose commodity similarity is greater than the second commodity similarity threshold as reference commodities based on a preset second commodity similarity threshold;

[0147] A third acquisition module, configured to acquire historical sales data of each of the reference commodities within a preset time period, where the start time of the preset time period is the launch time of the reference commodity;

[0148] A determination module, configured to determine a historical score value corresponding to the historical sales data based on a score value - sales data relationship table of each reference commodity, and each old commodity corresponds to a score value - sales data relationship table;

[0149] A calculation module, configured to calculate an average historical score value as an initial score value of the new commodity according to the historical score values of each of the old commodities.

[0150] As Figure 3 shown, an embodiment of the present invention further provides an electronic device, including a processor, and the above processor can execute any one of the above new and old commodity hybrid recommendation methods.

[0151] Specifically, it includes a processor 301 and a memory 302, and a computer program for executing the new and old commodity hybrid recommendation method stored on the memory 302 and capable of running on the processor 301, where:

[0152] The processor 301 runs the calculator program of the new and old commodity hybrid recommendation method stored in the memory 302 and executes the following steps:

[0153] Obtain a first score matrix, where the first score matrix includes score values of different users for each old commodity;

[0154] Obtain commodity information of the new commodity and commodity information of each of the old commodities, and calculate a commodity similarity between the new commodity and each of the old commodities based on the commodity information of the new commodity and the commodity information of the old commodities to obtain a similarity matrix between the new commodity and the old commodities, and the similarity matrix has the same arrangement order of old commodities as the first score matrix;

[0155] Add the new product and its initial rating value to the first rating matrix to obtain a second rating matrix;

[0156] Based on the similarity matrix and the first rating matrix, update the rating values of the new product and each of the old products in the second rating matrix to obtain a third rating matrix;

[0157] Perform product recommendation based on the third rating matrix.

[0158] Optionally, the step of adding the new product and its initial rating value to the first rating matrix to obtain a second rating matrix, which is executed by the processor 301, includes:

[0159] Among all the old products, determine whether there is a replacement product for the new product;

[0160] If there is a replacement product for the new product, then in the first rating matrix, replace the replacement product with the new product and replace the rating value of the replacement product with the initial rating of the new product to obtain the second rating matrix;

[0161] If there is no replacement product for the new product, then in the first rating matrix, determine the adjacent product of the new product, and the adjacent product is the old product with the highest product similarity;

[0162] Insert the new product and its initial rating value at the adjacent position of the adjacent product to obtain the second rating matrix.

[0163] Optionally, the step of inserting the new product and its initial rating value at the adjacent position of the adjacent product to obtain the second rating matrix, which is executed by the processor 301, includes:

[0164] Taking the adjacent product as the boundary, divide the similarity matrix into a first-side similarity matrix and a second-side similarity matrix, where the first-side similarity matrix is on one side of the adjacent product and the second-side similarity matrix is on the other side of the adjacent product;

[0165] Based on a preset first product similarity threshold, filter the product similarities less than the first product similarity threshold in the first-side similarity matrix and the second-side similarity matrix to obtain valid product similarities greater than or equal to the first product similarity threshold;

[0166] 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 determine the side where the first-side similarity matrix is located as the insertion side of the adjacent product;

[0167] If the total similarity of valid products in the first-side similarity matrix is less than the total similarity of valid products in the second-side similarity matrix, determine the other side where the second-side similarity matrix is located as the insertion side of the adjacent product;

[0168] Insert the new product and the initial score value of the new product at the adjacent position on the insertion side of the adjacent product to obtain the second score matrix.

[0169] Optionally, the processor 301 executes updating the score values of the new product and each of the old products in the second score matrix based on the similarity matrix and the first score matrix to obtain a third score matrix, including:

[0170] Perform matrix factorization on the first score matrix to obtain a first user matrix and a first product matrix, where the first user matrix includes first user eigenvalues corresponding to each of the users, and the first product matrix includes first product eigenvalues corresponding to each of the old products;

[0171] Perform matrix factorization on the second score matrix to obtain a second user matrix and a second product matrix, where the second user matrix includes second user eigenvalues corresponding to each of the users, the second product matrix includes the second product eigenvalues corresponding to the new product 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;

[0172] Based on the first product matrix, the second product matrix, and the similarity matrix, calculate to obtain a third product matrix, where the third product matrix includes the second product eigenvalues corresponding to the new product and each of the old products;

[0173] Perform matrix multiplication calculation on the second user matrix and the third product matrix to obtain a third score matrix.

[0174] Optionally, the processor 301 executes performing matrix factorization on the second score matrix to obtain a second user matrix and a second product matrix, including:

[0175] Perform matrix factorization on the second score matrix to obtain a second user matrix and a second product matrix;

[0176] Calculate the metric distance between the second user matrix and the first user matrix;

[0177] Taking the minimum of the measured distances as the optimization objective, iterate the process of matrix factorization. When the measured distance is less than the preset measured distance, stop the iteration to obtain the final second user matrix and the final second commodity matrix.

[0178] Optionally, the processor 301 calculates the third commodity matrix based on the first commodity matrix, the second commodity matrix, and the similarity matrix, including:

[0179] Expand the dimension of the similarity matrix on the non-commodity dimension by interpolation to obtain a similarity expansion matrix;

[0180] Perform matrix multiplication on the first commodity matrix and the similarity expansion matrix to obtain an intermediate matrix;

[0181] Perform matrix multiplication on the intermediate matrix and the second commodity matrix to obtain the third commodity matrix.

[0182] Optionally, before adding the new commodity and the initial score value of the new commodity to the first score matrix to obtain the second score matrix, the method executed by the processor 301 further includes:

[0183] Based on a preset second commodity similarity threshold, select, from all the old commodities, the old commodities whose commodity similarity is greater than the second commodity similarity threshold as reference commodities;

[0184] Obtain the historical sales data of each reference commodity within a preset time period, where the start time of the preset time period is the listing time of the reference commodity;

[0185] Based on the score - sales data relationship table of each reference commodity, determine the historical score value corresponding to the historical sales data, and each old commodity corresponds to a score - sales data relationship table;

[0186] Calculate the average historical score value as the initial score value of the new commodity according to the historical score values of each old commodity.

[0187] An embodiment of the present invention further provides a computer - readable storage medium. A computer program is stored on the computer - readable storage medium. When the computer program is executed by a processor, it implements each process of the method for hybrid recommendation of new and old commodities provided by the embodiment of the present invention and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0188] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above various methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM, for short), etc.

[0189] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by 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; specifically comprising: performing matrix decomposition on the first scoring matrix to obtain a first user matrix and a first product matrix, the first user matrix including first user feature values ​​corresponding to each of the users, and the first product matrix including first product feature values ​​corresponding to each of the old products; performing matrix decomposition on the second scoring matrix to obtain a second user matrix and a second product matrix, the second user matrix including second user feature values ​​corresponding to each of the users, the second product matrix including 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, the third product matrix including third product feature values ​​corresponding to the new products and each of the old products; performing matrix multiplication calculation on the second user matrix and the third product matrix 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 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.

5. The mixed recommendation method of new and used goods according to claim 1, 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.

6. The mixed recommendation method for new and used goods according to any one of claims 1 to 5, 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.

7. A mixed recommendation device for 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 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, 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 is used to update the scoring 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; specifically comprising: performing matrix decomposition on the first scoring 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 scoring 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; calculating a third product matrix based on the first product matrix, the second product matrix and the similarity matrix, wherein the third product matrix includes third product feature values ​​corresponding to the new products and each of the old products; performing matrix multiplication calculation on the second user matrix and the third product matrix to obtain a third scoring matrix; A product recommendation module is used to make product recommendations based on the third rating matrix.

8. 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 6 are implemented.

9. 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 6.

Images