A method and apparatus for recommending a product

Abstract

The application discloses a kind of commodity recommendation method and device, it is related to computer technical field.The specific embodiment of the method includes: according to the historical browsing information of user, determine the special moment of user;Determine the historical commemorative goods corresponding to the special moment of the user;According to the historical commemorative goods, determine the commodity set that there is association relationship with the historical commemorative goods;According to the commodity set, recommend commodity for the user at the special moment.This embodiment solves the technical defects that existing technology cannot remind user to purchase corresponding commodity in the special moment of user, and then reaches the technical effect of improving the purchasing power of user, improving the shopping experience of user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a method and device for recommending goods. BACKGROUND

[0002] The recommendation system of e-commerce platform is attracting more and more attention, and recommending the products that the user wants to the user can improve the purchasing power of the user and bring better shopping experience to the user. The existing recommendation system can be roughly divided into collaborative filtering-based recommendation, content-based recommendation and time-based recommendation.

[0003] In the process of implementing the present application, the inventors have found that at least the following problems exist in the prior art:

[0004] 1. When there is no previous purchase history of the user, but the goods need to be recommended to the user, the technical problem of cold start exists.

[0005] 2. When recommending goods to the user, the special moment of the user is ignored, such as wedding anniversary and special holidays of the user. SUMMARY

[0006] Therefore, the embodiments of the present application provide a method and device for recommending goods, which can solve the technical defects that the prior art cannot remind the user to purchase the corresponding goods at the special moment of the user by the technical means of determining the special moment of the user according to the historical browsing information, and further achieve the technical effects of improving the purchasing power of the user and improving the shopping experience of the user. Moreover, the present application further determines the goods to be recommended according to the historical commemorative goods of the user, and further achieves the technical effects of more accurately and efficiently recommending goods, and further improves the user experience.

[0007] To achieve the above-mentioned purpose, according to one aspect of the embodiments of the present application, a method for recommending goods is provided, comprising:

[0008] determining the special moment of the user according to the historical browsing information of the user;

[0009] determining the historical commemorative goods corresponding to the special moment of the user;

[0010] determining a set of goods having an association relationship with the historical commemorative goods according to the historical commemorative goods;

[0011] recommending goods to the user at the special moment according to the set of goods;

[0012] wherein the historical browsing information comprises historical order information and / or historical search goods information of the user.

[0013] Optionally, the special moment of the user is determined according to the historical browsing information of the user, including:

[0014] The keyword of the historical order information is determined according to the historical browsing information of the user.

[0015] The keyword is scored according to the words in the preset corpus, and the score of the keyword and the words corresponding to the keyword in the corpus are determined.

[0016] The special moment of the user is determined according to the score of the keyword and the words corresponding to the keyword.

[0017] Optionally, the keyword of the historical order information is determined according to the historical browsing information of the user, including:

[0018] When the historical browsing information of the user is the historical order information of the user, the keyword of the historical order information is determined by using the term frequency-inverse document frequency index statistical method.

[0019] And / or,

[0020] When the historical browsing information of the user is the historical search commodity information of the user, the historical search commodity information in a preset time interval is summarized into an article according to a preset time interval, and the keyword in each article is determined by using the term frequency-inverse document frequency index statistical method.

[0021] Optionally, the special moment of the user is determined according to the score of the keyword and the words corresponding to the keyword, including:

[0022] The preset score threshold is compared with the score of the keyword, and the keyword with a score not less than the score threshold is selected as a first keyword.

[0023] The historical order information and / or historical search commodity information corresponding to the first keyword of the user is determined.

[0024] The order delivery time in the historical order information corresponding to the first keyword and / or the search time in the historical search commodity information is set as the special moment of the user.

[0025] The type of the special moment of the user is determined according to the words corresponding to the keyword.

[0026] Optionally, the words in the corpus include at least one of the following: love, marriage, anniversary, anniversary, birthday.

[0027] The type of the special moment of the user includes at least one of the following: love anniversary, wedding anniversary, anniversary, birthday.

[0028] Optionally, the determining, according to the historical commemorative product, a product set having a correlation relationship with the historical commemorative product comprises:

[0029] obtaining a neural network model for calculating a product correlation degree and a product set to be recommended;

[0030] determining, according to the neural network model, a correlation degree between a product in the product set to be recommended and the historical commemorative product;

[0031] screening, according to the correlation degree, the product in the product set to be recommended to obtain the product set having the correlation relationship with the historical commemorative product.

[0032] Optionally, before the obtaining the neural network model for calculating the product correlation degree, the method comprises:

[0033] obtaining a correlation degree matrix between products in a training set for training a neural network;

[0034] establishing, according to the correlation degree matrix, a near neighbor matrix between the products in the training set;

[0035] training, according to the near neighbor matrix, the neural network model for calculating the product correlation degree.

[0036] Optionally, the obtaining the correlation degree matrix between the products in the training set for training the neural network comprises:

[0037] establishing an element in the correlation degree matrix between the products in the training set for training the neural network by using the following formula:

[0038]

[0039] wherein, |U i | represents a number of users corresponding to a first product i in the training set for training the neural network; |U j | represents a number of users corresponding to a second product j having a correlation relationship with the first product i; |U i ∩U j | represents a number of users corresponding to both the first product i and the second product j.

[0040] Optionally, the establishing, according to the correlation degree matrix, the near neighbor matrix between the products in the training set comprises:

[0041] sorting elements in the correlation degree matrix according to a value corresponding to each element from large to small to obtain a first sequence;

[0042] updating a value of a first preset number of elements in the first sequence to 1 and updating a value of the remaining elements to 0;

[0043] According to the updated element, a near neighbor matrix between commodities in the training set is established.

[0044] Optionally, according to the near neighbor matrix, the neural network model for calculating the commodity correlation degree is trained, comprising:

[0045] According to the near neighbor matrix and a loss function, an output parameter of the neural network model is determined;

[0046] The output parameter is trained by using a gradient, and the neural network model for calculating the commodity correlation degree is generated;

[0047] The expression of the loss function is:

[0048] L = S ij (||e i -e j || 2 )+(1-S ij )[α-||e i -e j || 2 ] +

[0049] [α-||e i -e j || 2 ] + =max(α-||e i -e j || 2 ,0)

[0050] Wherein, S ii is the i-th row and j-th column of the near neighbor matrix S; e i is an output parameter determined by the neural network model for commodity i; e j is an output parameter determined by the neural network model for commodity j.

[0051] Optionally, according to the set of commodities, the user is recommended a commodity at the special time, comprising:

[0052] The historical commemorative commodity corresponding to the user at the special time is obtained;

[0053] The historical commemorative commodity is input into the neural network model as a parameter to determine the output parameter determined for the historical commemorative commodity;

[0054] The commodities in the set of commodities are input into the neural network model as a parameter to determine the output parameter determined for the commodities in the set of commodities;

[0055] recommend the goods in the special time to the user when the Euclidean distance between the output parameter of the historical commemorative goods and the output parameter of the goods in the goods set is the smallest.

[0056] According to still another aspect of the embodiments of the present application, a device for recommending goods is provided, comprising:

[0057] a special time determining module, configured to determine a special time of a user according to historical browsing information of the user;

[0058] a historical commemorative goods determining module, configured to determine historical commemorative goods corresponding to the special time of the user;

[0059] a goods set determining module, configured to determine a goods set having an association relationship with the historical commemorative goods according to the historical commemorative goods;

[0060] a user recommended goods generating module, configured to recommend goods to the user at the special time according to the goods set;

[0061] The historical browsing information comprises historical order information and / or historical search goods information of the user.

[0062] According to another aspect of the embodiments of the present application, an electronic device for recommending goods is provided, comprising:

[0063] one or more processors;

[0064] a storage device, configured to store one or more programs,

[0065] when the one or more programs are executed by the one or more processors, the one or more processors implement the method for recommending goods provided by the present application.

[0066] According to still another aspect of the embodiments of the present application, a computer readable medium having a computer program stored thereon is provided, the program is executed by a processor to implement the method for recommending goods provided by the present application.

[0067] One of the embodiments of the above-mentioned application has the following advantages or beneficial effects:

[0068] The technical means of determining the special time of the user according to the historical browsing information provided by the present application can solve the technical defects that the prior art cannot remind the user to purchase corresponding goods at the special time of the user, and further achieve the technical effects of improving the purchasing power of the user and enhancing the shopping experience of the user.

[0069] The scheme also determines the to-be-recommended goods according to the historical memorial goods of the user, thereby achieving the technical effect of more accurate and efficient recommendation of goods, and further improving the user experience.

[0070] Further effects of the above-described non-conventional optional mode will be described in the following in conjunction with the specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0071] The accompanying drawings are used to better understand the present application and do not constitute undue limitations on the present application. Among them:

[0072] Figure 1 is a schematic diagram of the main flow of a method for recommending goods according to an embodiment of the present application, as shown in

[0073] Figure 2 is a schematic diagram of the overall flow of a method for recommending goods according to an embodiment of the present application;

[0074] Figure 3 is a schematic diagram of the main modules of a device for recommending goods according to an embodiment of the present application;

[0075] Figure 4 is an exemplary system architecture diagram to which embodiments of the present application can be applied;

[0076] Figure 5 is a structural schematic diagram of a computer system of a terminal device or a server suitable for implementing embodiments of the present application. DETAILED DESCRIPTION

[0077] The exemplary embodiments of the present application are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present application to help understanding, and should be considered only as exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Also, for the sake of clarity and conciseness, the description below omits the description of well-known functions and structures.

[0078] Figure 1 is a schematic diagram of the main flow of a method for recommending goods according to an embodiment of the present application, as shown in Figure 1 , including:

[0079] Step S101, determining a special moment of a user according to historical browsing information of the user;

[0080] Step S102, determining a historical memorial good corresponding to the special moment of the user;

[0081] Step S103, determining a set of goods having an association relationship with the historical memorial good according to the historical memorial good;

[0082] recommend a commodity for the user at the special time according to the commodity set;

[0083] The historical browsing information comprises historical order information and / or historical search commodity information of the user.

[0084] The present application can estimate the time of arrival of the commodity purchased by the user as the special time of the user by obtaining the historical browsing information of the user. For example, the user purchases a birthday gift for a friend on an e-commerce platform on October 10, 2000, and the delivery date is October 12, 2000, so it can be inferred that the birthday of the friend is around October 12. The method provided by the present application can enable the user to receive the system-recommended commodity about the birthday gift around October 12 every year, wherein October 12 is a special time of the user. Since there are cases where the user does not use the e-commerce platform to purchase a gift for a friend, the historical search commodity information of the user can also be used as a way to determine the special time.

[0085] The technical means for determining the special time of the user according to the historical browsing information provided by the present application can solve the technical defects of the prior art that cannot remind the user to purchase the corresponding commodity at the special time of the user, thereby achieving the technical effects of improving the purchasing power of the user and improving the shopping experience of the user.

[0086] For different special times, the commodities purchased by the user can be different. For example, when the special time is a wedding anniversary, the commodity purchased by the user can be flowers, customized souvenirs, etc. However, when the special time is a friend's birthday, the commodity purchased by the user can be a birthday cake, a candle, etc. By determining the historical souvenir purchased at the special time, the type of the special time can be more determined, and it is convenient to recommend commodities to the user subsequently.

[0087] The technical solution provided by the present application can recommend more suitable commodities to the user according to the type of the special time and the historical souvenir commodity information. Specifically, the user purchases a birthday cake for a friend on an e-commerce platform on October 10, 2000, and it can be inferred that the special time is October 12, and the type of the special time is the birthday of the friend, so the platform will recommend other styles of birthday cakes, birthday dolls, etc. to the user around October 12 of the next year.

[0088] The type of the special time of the user comprises but is not limited to one of the following: a love anniversary, a wedding anniversary, an anniversary, a birthday.

[0089] The commodity in the present solution can also be an article.

[0090] Optionally, in step S101, a special moment of the user is determined according to the historical browsing information of the user, including:

[0091] According to the historical browsing information of the user, a keyword of the historical order information is determined;

[0092] According to the keyword in the preset corpus, the keyword is scored to determine the score of the keyword and the keyword corresponding to the corpus;

[0093] According to the score of the keyword and the keyword corresponding to the keyword, the special moment of the user is determined.

[0094] Among them, the score score of the keyword k The higher, the more accurate the keyword is, and the more likely it is to be a description of a commemorative product. Specifically, a threshold can be set, and once the score of the keyword reaches the preset threshold, it can be considered that the moment corresponding to the historical browsing information is a special moment of the user.

[0095] Among them, the keyword in the corpus includes but is not limited to one of the following: love, marriage, anniversary, anniversary, birthday.

[0096] Optionally, according to the historical browsing information of the user, the keyword of the historical order information is determined, including:

[0097] When the historical browsing information of the user is the historical order information of the user, the term frequency-inverse document frequency index statistical method (TF-IDF) is used to determine the keyword of the historical order information;

[0098] And / or,

[0099] When the historical browsing information of the user is the historical search product information of the user, according to a preset time interval, the historical search product information in a preset time interval is summarized into an article; the term frequency-inverse document frequency index statistical method is used to determine the keyword in each article.

[0100] Among them, the TF-IDF is a statistical method that can be used to evaluate the importance of a word to an article in a corpus.

[0101] The keyword recognition algorithm used in the present application constructs the keywords of interest in advance by taking the description of the product and the user search history as the articles of the corpus, thereby completing the search for the commemorative object and the determination of the special moment.

[0102] TF-IDF can be used to evaluate a word d i For an article p in the corpus D kDegree of importance. Its main idea is to start from two aspects. The first is to look at the frequency of the word appearing in the document (TF: Term Frequency)

[0103]

[0104] where n ik represents the number of times the word d i appears in the article p k The greater the frequency of a word, the more important it is to this article. However, there are many words that do not actually determine the type of the article, such as: "le", "de", etc. These words often appear in almost every article. Therefore, the inverse document frequency (IDF: Inverse Document Frequency) is also needed

[0105]

[0106] Among them, |D| in the above formula represents the total number of articles in the corpus, and |{k|d i ∈p k}| represents the number of articles in the corpus that contain the word d i , and +1 is for smoothing to prevent the denominator from being 0. The final tfidf ik = tf ik * idf ik , which well eliminates the influence of those words that appear many times in each article. Therefore, if tfidf ik is larger, it means that the word d i is more likely to be the keyword of the article p k . As long as we construct a word library of the words we care about in advance, and then regard the description of each product as an article, we can determine whether this product is the commemorative product we care about. Therefore, we can first determine our word library as: {"love", "anniversary", "marriage", "anniversary", "birthday"}. Then determine the keywords for each article p k Determine the keywords

[0107] By determining the keywords in two different ways, namely, the user's historical order information and the historical search product information, the obtained keywords can be more accurate and more conducive to determining the user's special moment<000032​​​​​​​​​​​

[0111] Specifically, the special moment of the user is determined according to the score of the keyword and the word corresponding to the keyword, including:

[0112] The score threshold is compared with the score of the keyword, and the keyword with a score not less than the score threshold is selected as the first keyword;

[0113] The historical order information and / or historical search commodity information of the user corresponding to the first keyword are determined;

[0114] The order delivery time in the historical order information corresponding to the first keyword and / or the search time in the historical search commodity information are set as the special moment of the user;

[0115] The type of the special moment of the user is determined according to the word corresponding to the keyword.

[0116] Optionally, according to the historical commemorative goods, a set of goods having an association relationship with the historical commemorative goods is determined, including:

[0117] A neural network model for calculating the association degree of goods and a set of goods to be recommended are obtained;

[0118] According to the neural network model, the association degree between the goods in the set of goods to be recommended and the historical commemorative goods is determined;

[0119] According to the association degree, the set of goods having an association relationship with the historical commemorative goods is obtained by screening the goods in the set of goods to be recommended.

[0120] In actual operation, only it is necessary to detect whether there is a user reaching a special moment needs to be predicted, and then the historical purchase or browsing record of the user is extracted, and the technical means of calculating the association degree between goods through the trained recommendation model is used to overcome the technical defects of the prior art, such as complex calculation of association degree, low efficiency and inaccuracy, and then the most likely purchased goods combination of the user is predicted, and the technical effect of quickly completing goods recommendation is achieved.

[0121] In the present application, the technical means of deep metric learning is used, and the purpose of learning a metric function satisfying the semantic relationship can be achieved by constructing a deep model.

[0122] The following is a specific real-time way to illustrate how the neural network is trained in this scheme:

[0123] Optionally, before obtaining the neural network model for calculating the association degree of goods, it includes:

[0124] An association degree matrix between goods in a training set for training the neural network is obtained;

[0125] Based on the correlation matrix, a nearest neighbor matrix is ​​established among the products in the training set;

[0126] The neural network model used to calculate the correlation between products is trained based on the nearest neighbor matrix.

[0127] This application utilizes a technique that transforms the correlation matrix of goods in the training set into a proximity matrix composed of 0s and 1s, thereby reducing the computational load used to train the model and improving the efficiency of the generative neural network model.

[0128] Specifically, the correlation matrix between items in the training set used to train the neural network is obtained, including:

[0129] The following formula is used to construct the elements of the correlation matrix between products in the training set for training the neural network:

[0130]

[0131] Among them, |U i | represents the number of users corresponding to the first item i in the training set used to train the neural network; |U j | Represents the number of users corresponding to the second product j that is associated with the first product i; |U i ∩U j | represents the number of users corresponding to both the first product i and the second product j.

[0132] In practical applications, it has been found that users who prefer a certain type of product are likely to continue purchasing that type of product or similar products. Therefore, the correlation between two types of products can be calculated by observing whether users purchase the same or similar products. This facilitates the training of neural networks.

[0133] Optionally, based on the correlation matrix, a nearest neighbor matrix is ​​established among the items in the training set, including:

[0134] The elements in the correlation matrix are sorted in descending order of their corresponding numerical values ​​to obtain the first sequence;

[0135] Update the values ​​of the first preset number of elements in the first sequence to 1, and update the values ​​of the remaining elements to 0;

[0136] Based on the updated elements, a nearest neighbor matrix is ​​constructed among the items in the training set.

[0137] By normalizing the values ​​in the multi-association degree matrix, the original correlation degrees ranging from 0 to 1 are transformed into 0 or 1. Specifically, elements with high correlation degrees are mapped to 1, and elements with low correlation degrees are mapped to 0, thus transforming the original correlation degree matrix into a nearest neighbor matrix. Specifically, Sij is the i-th row and j-th column of the proximity matrix S. The value of S ii represents a flag indicating whether commodity i and commodity j are proximate or not. If S ij = 1, they are proximate, and if S ij = 0, they are not proximate.

[0138] Optionally, the neural network model for calculating the commodity correlation degree is trained according to the proximity matrix, comprising:

[0139] According to the proximity matrix and the loss function, the output parameter of the neural network model is determined; by adjusting the change of the input proximity matrix, the size of the loss function is reduced as much as possible, so that the output parameter obtained by training is more accurate.

[0140] The output parameter is trained by using the gradient, and the neural network model for calculating the correlation degree between two commodities is generated; specifically, the features of two commodities or users are input to obtain the correlation degree between the two commodities or the two users.

[0141] The expression of the loss function is:

[0142] L = S ij (||e i -e j || 2 )+(1-S ij )[α-||e i -e j || 2 ] +

[0143] [α-||e i -e j || 2 ] + = max(α-||e i -e j || 2 , 0)

[0144] Wherein, S ii is the i-th row and j-th column of the proximity matrix S; e i is the output parameter determined by the neural network model for commodity i; e j is the output parameter determined by the neural network model for commodity j.

[0145] For each commodity i, only the k j with the highest correlation degree with it are taken as proximate, that is, in the i-th row of W, W jThe k largest k j of i are the neighbors of i. Next, by taking the basic attributes of the commodities as inputs, a deep neural network is used as a model, and the distance between commodities with an association relationship is required to be as close as possible, and the distance between dissimilar commodities is required to be greater than alpha at least, a deep metric learning loss L is constructed; wherein, alpha is the distance between commodity i and commodity j for determining the association degree.

[0146] Through the above neural network model, whether a commodity is similar to a commodity previously purchased by a user can be determined, and if so, the commodity to be recommended can be recommended to the user.

[0147] In actual application, the algorithm of collaborative filtering can also be used to calculate the association degree between commodities.

[0148] Optionally, according to the set of commodities, the user is recommended a commodity at the special time, comprising:

[0149] The historical commemorative commodity corresponding to the user at the special time is obtained;

[0150] The historical commemorative commodity is input into the neural network model as a parameter to determine the output parameter determined for the historical commemorative commodity;

[0151] The commodities in the set of commodities are input into the neural network model as parameters to determine the output parameters determined for the commodities in the set of commodities;

[0152] When the Euclidean distance between the output parameters determined for the historical commemorative commodity and the output parameters determined for the commodities in the set of commodities is the smallest, the commodities in the set of commodities are recommended to the user at the special time.

[0153] Through the neural network model, the most likely purchased commodity of a user to be predicted can be calculated, and it is assumed that the set of commemorative objects previously purchased by the user u is S u , and the commodity to be recommended needs to satisfy the following conditions:

[0154]

[0155] In the above formula, I represents the set of all commemorative commodities to be recommended. According to the commodities in the set of commodities (for example, the set includes {commemorative object, red wine, candle}), other closest matching commodities are found.

[0156] The flow of the present scheme is described below with a specific embodiment. Figure 2 is a schematic diagram of the overall flow of a commodity recommendation method according to an embodiment of the present application, as Figure 2 shown, comprising:

[0157] S201, acquire historical browsing information of a user;

[0158] S202, identify keywords from the historical browsing information according to a TF-IDF technique;

[0159] S203, determine a special moment according to the identification result of S202;

[0160] S204, determine a historical record commodity according to the identification result of S202;

[0161] S205, generate a neural network model corresponding to S206 for the historical record commodity of S204 through deep learning;

[0162] S207, output a recommended commodity according to the neural network model and the special moment determined by S203.

[0163] Figure 3 is a schematic diagram of main modules of a device for a commodity recommendation method according to an embodiment of the present application;

[0164] According to still another aspect of the present application, a device 300 for commodity recommendation is provided, as shown in Figure 3 , comprising:

[0165] A special moment determination module 301 is configured to determine a special moment of a user according to historical browsing information of the user.

[0166] A historical record commodity determination module 302 is configured to determine a historical record commodity corresponding to the special moment of the user.

[0167] A commodity set determination module 303 is configured to determine a commodity set having an association relationship with the historical record commodity according to the historical record commodity.

[0168] A user recommended commodity generation module 304 is configured to recommend commodities for the user at the special moment according to the commodity set.

[0169] The historical browsing information includes historical order information and / or historical search commodity information of the user.

[0170] Optionally, determining a special moment of a user according to historical browsing information of the user comprises:

[0171] Determining keywords of the historical order information according to the historical browsing information of the user;

[0172] Scoring the keywords according to words in a preset corpus to determine scores of the keywords and the words in the corpus corresponding to the keywords.

[0173] determine a special moment of the user according to the score of the keyword and a word corresponding to the keyword.

[0174] Optionally, the keyword of the historical order information is determined according to the historical browsing information of the user, including:

[0175] When the historical browsing information of the user is the historical order information of the user, the keyword of the historical order information is determined by using a term frequency-inverse document frequency index statistical method.

[0176] and / or,

[0177] When the historical browsing information of the user is the historical search commodity information of the user, the historical search commodity information in a preset time interval is summarized into an article according to a preset time interval, and the keyword in each article is determined by using a term frequency-inverse document frequency index statistical method.

[0178] Optionally, the special moment of the user is determined according to the score of the keyword and a word corresponding to the keyword, including:

[0179] A preset score threshold is compared with the score of the keyword, and a keyword with a score not less than the score threshold is selected as a first keyword.

[0180] The historical order information and / or the historical search commodity information corresponding to the first keyword are determined.

[0181] The order delivery time in the historical order information corresponding to the first keyword and / or the search time in the historical search commodity information are set as the special moment of the user.

[0182] The type of the special moment of the user is determined according to the word corresponding to the keyword.

[0183] Optionally, the words in the corpus include at least one of the following: love, marriage, anniversary, birthday, and the like.

[0184] The type of the special moment of the user includes at least one of the following: love anniversary, marriage anniversary, anniversary, and birthday.

[0185] Optionally, the set of commodities associated with the historical commemorative commodity is determined according to the historical commemorative commodity, including:

[0186] A neural network model for calculating commodity association degree and a set of commodities to be recommended are obtained.

[0187] The association degree between the commodities in the set of commodities to be recommended and the historical commemorative commodity is determined according to the neural network model.

[0188] According to the correlation degree, the goods in the set of goods to be recommended are screened to obtain a set of goods having a correlation relationship with the historical commemorative goods.

[0189] Optionally, before obtaining the neural network model for calculating the correlation degree of goods, the following steps are further included:

[0190] Obtaining a correlation degree matrix between goods in a training set for training a neural network;

[0191] According to the correlation degree matrix, a near neighbor matrix between the goods in the training set is established;

[0192] According to the near neighbor matrix, the neural network model for calculating the correlation degree of goods is trained.

[0193] Optionally, obtaining the correlation degree matrix between goods in a training set for training a neural network includes:

[0194] An element in the correlation degree matrix between goods in a training set for training a neural network is established by using the following formula:

[0195]

[0196] Wherein, |U i | represents the number of users corresponding to the first good i in the training set for training a neural network; |U j | represents the number of users corresponding to the second good j having a correlation relationship with the first good i; |U i ∩U j | represents the number of users corresponding to the first good i and the second good j.

[0197] Optionally, according to the correlation degree matrix, the near neighbor matrix between the goods in the training set is established, including:

[0198] The elements in the correlation degree matrix are sorted in descending order according to the numerical values of the elements to obtain a first sequence;

[0199] The numerical values of the first preset number of elements in the first sequence are updated to 1, and the numerical values of the remaining elements are updated to 0;

[0200] According to the updated elements, the near neighbor matrix between the goods in the training set is established.

[0201] Optionally, according to the near neighbor matrix, the neural network model for calculating the correlation degree of goods is trained, including:

[0202] According to the near neighbor matrix and the loss function, the output parameters of the neural network model are determined;

[0203] training the output parameter by using the gradient, to generate the neural network model used for calculating the commodity correlation degree;

[0204] The expression of the loss function is:

[0205] L=S ij (||e i -e j || 2 )+(1-S ij )[α-||e i -e j || 2 +

[0206] Wherein, [α-||e i -e j || 2 ] + =max(α-||e i -e j || 2 ,0)

[0207] Wherein, S ij is the i-th row and j-th column of the near neighbor matrix S; e i is the output parameter determined by the neural network model for commodity i; e j is the output parameter determined by the neural network model for commodity j.

[0208] Optionally, according to the commodity set, the user is recommended a commodity at the special time, comprising:

[0209] Obtaining the historical commemorative commodity corresponding to the user at the special time;

[0210] The historical commemorative commodity is input into the neural network model as a parameter to determine the output parameter determined for the historical commemorative commodity;

[0211] The commodities in the commodity set are input into the neural network model as a parameter to determine the output parameter determined for the commodities in the commodity set;

[0212] When the Euclidean distance between the output parameter determined for the historical commemorative commodity and the output parameter determined for the commodities in the commodity set is the smallest, the commodities in the commodity set are recommended to the user at the special time.

[0213] Figure 4 An exemplary system architecture 400 of a commodity recommendation method or a commodity recommendation device to which embodiments of the present application can be applied is shown.

[0214] ​like Figure 4 As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405. Network 404 serves as the medium for providing communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0215] Users can use terminal devices 401, 402, and 403 to interact with server 405 via network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, and 403, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0216] Terminal devices 401, 402, and 403 can be various electronic devices with displays that support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0217] Server 405 can be a server that provides various services, such as a backend management server that supports shopping websites browsed by users using terminal devices 401, 402, and 403 (for example only). The backend management server can analyze and process data such as received product information query requests, and feed back the processing results (such as target push information, product information - for example only) to the terminal device.

[0218] It should be noted that the product recommendation method provided in this embodiment of the invention is generally executed by server 405, and correspondingly, the product recommendation device is generally set in server 405.

[0219] It should be understood that Figure 4 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0220] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system 500 suitable for implementing a terminal device of the present invention. Figure 5 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0221] like Figure 5As shown, the computer system 500 includes a central processing module (CPU) 501 which can perform various appropriate actions and processes in accordance with programs stored in a read only memory (ROM) 502 or loaded from a storage section 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the system 500 are also stored. The CPU 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0222] Connected to the I / O interface 505 are an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as necessary. A removable media 511 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 510 as necessary, so that a computer program read therefrom is installed into the storage section 508 as necessary.

[0223] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program in accordance with the embodiments disclosed herein. For example, the embodiments disclosed herein include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 509 and / or installed from the removable media 511. When the computer program is executed by the central processing module (CPU) 501, the above-described functions defined in the system of the present invention are performed.

[0224] It should be noted that the computer-readable medium shown in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0225] The flowcharts and block diagrams in the drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment or a part of code containing one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order than that shown in the drawings. For example, two blocks that are shown in succession can actually be executed substantially in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams or flowcharts, and the combination of blocks in the block diagrams or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0226] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. The described modules can also be arranged in a processor, for example, a processor can be described as including a sending module, an obtaining module, a determining module and a first processing module. In some cases, the names of these modules do not constitute a limitation on the modules themselves, for example, the sending module can also be described as "a module that sends a picture obtaining request to a connected server".

[0227] As another aspect, the present application also provides a computer readable medium, which can be included in the device described in the above embodiments, or can exist independently without being assembled into the device. The computer readable medium carries one or more programs, which, when executed by the device, cause the device to include:

[0228] determine a special moment of the user according to historical browsing information of the user;

[0229] determine a historical commemorative product corresponding to the special moment of the user;

[0230] determine a product set having an association relationship with the historical commemorative product according to the historical commemorative product;

[0231] recommend a product to the user at the special moment according to the product set;

[0232] The historical browsing information includes historical order information and / or historical search product information of the user.

[0233] The technical solution according to the embodiments of the present application can achieve the following technical effects:

[0234] The technical means for determining a special moment of a user according to historical browsing information provided by the present solution can solve the technical defect that the prior art cannot remind a user to purchase a corresponding product at a special moment of the user, thereby achieving the technical effects of improving the purchasing power of the user and improving the shopping experience of the user.

[0235] The present solution further determines a product to be recommended according to a historical commemorative product of the user, thereby achieving the technical effects of more accurately and efficiently recommending a product and further improving the user experience.

[0236] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made depending on design requirements and other factors. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for recommending products, characterized in that, include: Determine the user's special moments based on their browsing history; Identify the historical commemorative items corresponding to the user at that special moment; Based on the historical commemorative items, determine a set of items that are associated with the historical commemorative items; Based on the product set, recommend products to the user at the specific time; The historical browsing information includes: the user's historical order information and / or historical search product information; The step of determining a user's special moment based on the user's browsing history includes: determining keywords of the historical order information based on the user's browsing history; scoring the keywords based on words in a preset corpus to determine the score of the keywords and the corresponding words in the corpus; determining the user's special moment based on the score of the keywords and the words corresponding to the keywords; the special moment can be of various types.

2. The method according to claim 1, characterized in that, Based on the user's browsing history, keywords for the historical order information are determined, including: When the user's browsing history information is the user's order history information, the keywords of the order history information are determined using the word frequency-inverse text frequency index statistical method; And / or, When the user's browsing history information is the user's historical search product information, the historical search product information within a preset time interval is summarized into an article; the keywords in each article are determined using the word frequency-inverse text frequency index statistical method.

3. The method according to claim 1, characterized in that, Based on the scores of the keywords and the words corresponding to the keywords, the user's special moments are determined, including: The preset score threshold is compared with the score of the keyword, and keywords with scores not less than the score threshold are selected as the first keyword; Determine the user's historical order information and / or historical search product information corresponding to the first keyword; Set the order delivery time in the historical order information corresponding to the first keyword, and / or the search time in the historical search product information to the user's special moment; The type of the user's special moment is determined based on the words corresponding to the keywords.

4. The method according to claim 3, characterized in that, The words in the corpus include at least one of the following: love, marriage, anniversary, birthday; The types of special moments for the user include at least one of the following: dating anniversary, wedding anniversary, anniversary, and birthday.

5. The method according to claim 1, characterized in that, Based on the historical commemorative merchandise, a set of merchandise related to the historical commemorative merchandise is determined, including: Obtain the neural network model used to calculate product relevance and the set of products to be recommended; Based on the neural network model, the correlation between the products in the set of products to be recommended and the historical commemorative products is determined; Based on the correlation, products in the set of products to be recommended are filtered to obtain a set of products that have a correlation with the historical commemorative products.

6. The method according to claim 5, characterized in that, Before obtaining the neural network model used to calculate product correlation, the following steps are included: Obtain the correlation matrix between items in the training set used to train the neural network; Based on the correlation matrix, a nearest neighbor matrix is ​​established among the products in the training set; The neural network model used to calculate the correlation between products is trained based on the nearest neighbor matrix.

7. The method according to claim 6, characterized in that, Obtain the association matrix between items in the training set used to train the neural network, including: The following formula is used to construct the elements of the correlation matrix between products in the training set for training the neural network: Among them, |U i | represents the number of users corresponding to the first item i in the training set used to train the neural network; |U j | Represents the number of users corresponding to the second product j that is associated with the first product i; |U i ∩U j | represents the number of users corresponding to both the first product i and the second product j.

8. The method according to claim 6, characterized in that, Based on the correlation matrix, a nearest neighbor matrix is ​​established among the items in the training set, including: The elements in the correlation matrix are sorted in descending order of their corresponding numerical values ​​to obtain the first sequence; Update the values ​​of the first preset number of elements in the first sequence to 1, and update the values ​​of the remaining elements to 0; Based on the updated elements, a nearest neighbor matrix is ​​constructed among the items in the training set.

9. The method according to claim 6, characterized in that, Based on the nearest neighbor matrix, train the neural network model for calculating the correlation between products, including: Based on the nearest neighbor matrix and the loss function, determine the output parameters of the neural network model; The output parameters are trained using gradients to generate the neural network model used to calculate the correlation between products; The expression for the loss function is: L=S ij (||and i -And j || 2 )+(1-S ij )[α-||e i -And j || 2 ] + [α-||e i -and j || 2 ] + =max(α-||e i -and j || 2 0) Among them, S ij e is the i-th row and j-th column of the nearest neighbor matrix S; i e represents the output parameters determined by the neural network model for product i. j α represents the output parameters determined by the neural network model for product j; α is the distance between product i and product j.

10. The method according to claim 9, characterized in that, Based on the product set, recommending products to the user at the specific time includes: Retrieve the historical commemorative items corresponding to the user at the special moment; The historical commemorative merchandise is input as an input parameter into the neural network model to determine the output parameters specific to the historical commemorative merchandise. The products in the product set are input as input parameters to the neural network model to determine the output parameters for the products in the product set; When the Euclidean distance between the output parameters determined for historical commemorative items and the output parameters determined for items in the item set is minimized, the items in the item set are recommended to the user at that special moment.

11. A device for recommending products, characterized in that, include: The special moment determination module is used to determine the user's special moments based on the user's browsing history. The historical commemorative merchandise identification module is used to identify the historical commemorative merchandise corresponding to the user at the special moment. The product set determination module is used to determine a product set that is associated with the historical commemorative products based on the historical commemorative products. The user-recommended product generation module recommends products to the user at the specific time based on the product set. The historical browsing information includes: the user's historical order information and / or historical search product information; The step of determining a user's special moment based on the user's browsing history includes: determining keywords of the historical order information based on the user's browsing history; scoring the keywords based on words in a preset corpus to determine the score of the keywords and the corresponding words in the corpus; determining the user's special moment based on the score of the keywords and the words corresponding to the keywords; the special moment can be of various types.

12. An electronic device for product recommendation, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-10.

13. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-10.

Images