Click rate-based product recommendation method, device, equipment and medium

Abstract

The present disclosure provides a click rate-based product recommendation method, device, equipment and medium, relating to the technical field of artificial intelligence. The method comprises: obtaining a plurality of historical transaction data of a user purchasing a product to form a training data set; obtaining m products to be recommended; extracting product features and user behavior features of each product in the m products according to the training data set to generate m feature vectors; calculating posterior probabilities corresponding to the m feature vectors respectively using a naive Bayes algorithm, and obtaining a first ranking list of the m products according to the posterior probabilities; inputting the m feature vectors into a deep learning network model to obtain click rate estimates of the m products, and obtaining a second ranking list of the m products according to the click rate estimates, wherein the deep learning network model is obtained by pre-training based on the training data set; correcting the first ranking list and the second ranking list to obtain a third ranking list, and sequentially recommending the m products according to the third ranking list.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, in particular to a product recommendation method and device based on click rate, electronic equipment, storage medium and program product. BACKGROUND

[0002] With the rise of product recommendation in financial services, the recommendation of fund, insurance and financial products (hereinafter referred to as fund, insurance and financial products) has become popular. If the products that customers care about and prefer can be placed at the top of the recommendation list, the transaction volume of such products will be increased. In the prior art, the common practice is to estimate the click-through rate (CTR) of fund, insurance and financial products according to the attributes of the products, and then form a recommendation list of products to be recommended according to the estimated CTR from high to low.

[0003] At present, the sorting method after CTR estimation of fund, insurance and financial products mainly includes two kinds: one is the traditional machine learning method represented by collaborative filtering, matrix decomposition algorithm, logistic regression (LR), gradient boosting decision tree (GBDT) + logistic regression and factorization machine (FM); the other is the deep learning method represented by DeepFM. The core of these two methods is to learn the features of the products, give the estimated CTR of the products, and then sort the products according to the estimated CTR from high to low.

[0004] However, in the process of implementing the present disclosure, the inventors found that at least the following problems exist in the related art: (1) In the traditional machine learning method, collaborative filtering and matrix decomposition algorithm can only give the similarity between different products to be recommended and the products purchased by the user in history, LR relies heavily on the selection of artificial features, GBDT+LR can automatically select features but cannot handle high-dimensional discrete features and the generalization ability of the model is weak, and FM can handle sparse features but does not consider the deep relationship between features. (2) In the deep learning method, the core is the modification of the neural network model, and the current better method is the DeepFM model. Although the DeepFM model considers both the shallow features of the features and the deep features of the features, the recommendation list of DeepFM is based on the order of the estimated CTR of each product from high to low, and does not make any correction to the estimated CTR and the sorting result, resulting in sometimes poor recommendation effect. SUMMARY

[0005] In view of the above problems, the present disclosure provides a product recommendation method and device based on click rate, electronic equipment, storage medium and program product.

[0006] According to a first aspect of the present disclosure, a click rate-based product recommendation method is provided, comprising: obtaining a plurality of historical transaction data of user purchasing products, constituting a training data set; obtaining m products to be recommended, m≥2 and m is an integer; extracting product features and user behavior features of each of the m products according to the training data set, and generating m feature vectors; calculating posterior probabilities corresponding to the m feature vectors respectively using a Naive Bayes algorithm, and obtaining a first ranking list of the m products according to the posterior probabilities; inputting the m feature vectors into a deep learning network model to obtain click rate estimates of the m products, and obtaining a second ranking list of the m products according to the click rate estimates, wherein the deep learning network model is obtained by pre-training based on the training data set; correcting the first ranking list and the second ranking list to obtain a third ranking list, and sequentially recommending the m products according to the third ranking list.

[0007] According to an embodiment of the present disclosure, the product features of each product include a plurality of sub-features, and the sub-features include product type, product praise degree, product sales, marketing novelty, and discount degree, each sub-feature including a plurality of attribute values; and the user behavior features of each product are discrete features, representing whether the user has purchased the product.

[0008] According to an embodiment of the present disclosure, the posterior probabilities corresponding to the m feature vectors respectively are calculated using the Naive Bayes algorithm, comprising: calculating a prior probability of each discrete feature in the training data set; calculating a conditional probability of different attribute values of each sub-feature under each discrete feature according to the prior probability; and calculating a posterior probability of each product in the m products belonging to each discrete feature according to the prior probability and the conditional probability.

[0009] According to an embodiment of the present disclosure, when the discrete features include a first discrete feature and a second discrete feature, the first discrete feature represents that the user has purchased the corresponding product, the second discrete feature represents that the user has not purchased the corresponding product, and the m products include a first product, after calculating the posterior probability of the first product belonging to each discrete feature, the method further comprises: determining whether the posterior probability of the first product belonging to the first discrete feature is greater than the posterior probability of the first product belonging to the second discrete feature, if yes, determining the posterior probability of belonging to the first discrete feature as a probability of the first product being recommended; otherwise, determining the probability of the first product being recommended according to a complementary event probability of the posterior probability of belonging to the second discrete feature; and determining the first ranking list of the m products according to the probability of the first product being recommended.

[0010] According to an embodiment of the present disclosure, when the posterior probability of the first product belonging to the first discrete feature is not greater than the posterior probability of the first product belonging to the second discrete feature, the probability of the first product being recommended is determined according to the following formula:

[0011]

[0012] wherein p k represents the probability that the first product is recommended; represents the posterior probability that the first product belongs to the second discrete feature.

[0013] According to an embodiment of the present disclosure, when the prior probability or the conditional probability is 0, a Laplace smoothing method is used for preprocessing.

[0014] According to an embodiment of the present disclosure, the deep learning network model is constructed according to a DeepFM network model, and an activation function σ2(x) of a hidden layer of the DeepFM network model is:

[0015]

[0016] wherein x represents an input of a hidden layer neuron; and σ2(x) represents an output of the hidden layer neuron.

[0017] According to an embodiment of the present disclosure, the deep learning network model is trained by: splitting a training data set into a training set, a validation set and a test set; training the constructed deep learning network model using the training set and the validation set, updating network parameters of the deep learning network model according to a preset loss function; and evaluating the effectiveness of the deep learning network model using the test set.

[0018] According to an embodiment of the present disclosure, the first ranking list and the second ranking list are modified to obtain a third ranking list, including: determining whether the first ranking list and the second ranking list are completely identical, if yes, determining the first ranking list or the second ranking list as the third ranking list; otherwise, multiplying each element in the first ranking list and the second ranking list respectively to obtain the third ranking list.

[0019] According to an embodiment of the present disclosure, after generating the m feature vectors, further comprising: extracting product identifiers of the m products according to the training data set, and establishing one-to-one mapping between the product identifiers of the m products and the m feature vectors.

[0020] According to an embodiment of the present disclosure, according to the third ranking list, the m products are recommended in turn, including: recommending the product identifier list mapped to the third ranking list in turn.

[0021] According to an embodiment of the present disclosure, the method further comprises: obtaining authorization of the user for a plurality of historical transaction data of the products purchased by the user; and obtaining the plurality of historical transaction data after obtaining the authorization of the user.

[0022] The second aspect of the present disclosure provides a product recommendation device based on click rate, comprising: a data acquisition module configured to acquire a plurality of historical transaction data of a user purchasing a product to form a training data set; a product acquisition module configured to acquire m products to be recommended, m is an integer greater than or equal to 2; a vector generation module configured to extract product features and user behavior features of each of the m products according to the training data set, and generate m feature vectors; a first sorting module configured to calculate posterior probabilities corresponding to the m feature vectors respectively using a naive Bayes algorithm, and obtain a first sorting list of the m products according to the posterior probabilities; a second sorting module configured to input the m feature vectors into a deep learning network model to obtain click rate estimates of the m products, and obtain a second sorting list of the m products according to the click rate estimates, wherein the deep learning network model is pre-trained based on the training data set; and a product recommendation module configured to correct the first sorting list and the second sorting list to obtain a third sorting list, and recommend the m products in turn according to the third sorting list.

[0023] The third aspect of the present disclosure provides an electronic device, comprising: one or more processors; a memory configured to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the product recommendation method based on click rate described above.

[0024] The fourth aspect of the present disclosure further provides a computer-readable storage medium having stored executable instructions, which, when executed by a processor, cause the processor to perform the product recommendation method based on click rate described above.

[0025] The fifth aspect of the present disclosure further provides a computer program product comprising a computer program, which, when executed by a processor, implements the product recommendation method based on click rate described above.

[0026] Compared with the prior art, the product recommendation method, device, electronic device, storage medium and program product based on click rate provided by the present disclosure have at least the following beneficial effects:

[0027] (1) The present disclosure combines traditional machine learning and deep learning to estimate the click rate of a plurality of products to be marketed, and then recommends the product information based on the click rate. This method does not require manual feature selection, can handle the relationship between shallow features and the relationship between deep features, and further corrects the estimated CTR given by the model, which can improve the accuracy of the recommendation;

[0028] (2) The present disclosure builds a federal learning platform with companies having user geographic location features, which helps to improve the accuracy of the bank abnormal user early warning model, can save assets, timely stop loss, prevent transactions, and even provide part of the information value of abnormal users to other external regulatory parties. Attached Figure Description

[0029] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0030] Figure 1 This illustration schematically depicts an application scenario suitable for implementing a click-through rate-based product recommendation method and apparatus according to embodiments of the present disclosure;

[0031] Figure 2 A flowchart illustrating a click-through rate-based product recommendation method according to an embodiment of the present disclosure is shown schematically.

[0032] Figure 3 A flowchart illustrating the posterior probability calculation process according to an embodiment of the present disclosure is shown schematically.

[0033] Figure 4 A flowchart illustrating the process of determining a first sorting list according to an embodiment of the present disclosure is shown schematically;

[0034] Figure 5 A flowchart illustrating the training process of a deep learning network model according to an embodiment of the present disclosure is shown schematically.

[0035] Figure 6 A flowchart illustrating a third sorting list modification process according to an embodiment of the present disclosure is shown schematically;

[0036] Figure 7 A flowchart illustrating the product identifier mapping establishment process according to an embodiment of the present disclosure is shown schematically.

[0037] Figure 8 A flowchart illustrating the product identifier list recommendation process according to an embodiment of the present disclosure is shown schematically.

[0038] Figure 9 A block diagram of a click-through rate-based product recommendation device according to an embodiment of the present disclosure is shown schematically.

[0039] Figure 10 A block diagram schematically illustrates an electronic device suitable for implementing a click-through rate-based product recommendation method according to an embodiment of the present disclosure. Detailed Implementation

[0040] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood, however, that the description is merely exemplary and is not intended to limit the scope of the present disclosure. In the following detailed description of the embodiments of the present disclosure, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it would be apparent to one skilled in the art that the present disclosure can be practiced without these specific details. In other instances, well-known structures and functions have been omitted to avoid unnecessarily complicating the present disclosure with details that will be readily apparent to those skilled in the art.

[0041] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. As used herein, the term "includes" and tautological expressions thereof, such as "including," "includes," "include," "contains," "containing," and so forth, shall not be taken to exclude

[0042] All terms used herein (including technical and scientific terms) have the meanings commonly understood by one of ordinary skill in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings that are consistent with the context of the specification, and should not be interpreted in an idealized or overly formal manner.

[0043] In the case where expressions similar to "at least one of A, B, and C, and the like" are used, in general, it should be interpreted that the meaning is the same as "one or more" of the A, B, and C (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, a system having B alone, a system having C alone, a system having both A and B, a system having both A and C, a system having both B and C, and / or a system having A, B, and C, etc.).

[0044] Some of the blocks and / or flowcharts in the drawings represent computer program instructions or programs. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, or other programmable data processing apparatus to create means for implementing the functions / operations specified in the block diagrams and / or flowcharts. The technology of the present disclosure can be implemented in the form of hardware and / or software (including firmware, microcode, etc.). In addition, the technology of the present disclosure can take the form of a computer program product on a computer readable storage medium having instructions stored thereon, which can be used by or in conjunction with an instruction execution system.

[0045] In the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information comply with relevant laws and regulations, necessary security measures are taken, and do not violate public order and good customs.

[0046] In the technical solutions of the present disclosure, the authorization or consent of the user is obtained before the personal information of the user is acquired or collected.

[0047] Embodiments of the present disclosure provide a product recommendation method and device based on click rate, equipment, storage medium and program product, relating to the technical field of artificial intelligence. The method comprises: acquiring a plurality of historical transaction data of a user purchasing a product to form a training data set; acquiring m products to be recommended, m>2 and m is an integer; extracting product features and user behavior features of each of the m products according to the training data set to generate m feature vectors; calculating the posterior probability corresponding to each of the m feature vectors using a naive Bayes algorithm, and obtaining a first ranking list of the m products according to the posterior probability; inputting the m feature vectors into a deep learning network model to obtain a click rate estimate value of the m products, and obtaining a second ranking list of the m products according to the click rate estimate value, wherein the deep learning network model is obtained by pre-training based on the training data set; correcting the first ranking list and the second ranking list to obtain a third ranking list, and sequentially recommending the m products according to the third ranking list.

[0048] Figure 1 An application scenario suitable for implementing the product recommendation method and device based on click rate according to the embodiments of the present disclosure is schematically shown. It should be noted that, Figure 1 The shown is only an example of an application scenario in which the embodiments of the present disclosure can be applied, to help those skilled in the art understand the technical content of the present disclosure, but does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0049] As Figure 1 shown, the application scenario 100 according to the embodiments can include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 can include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0050] The user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0051] The terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers and desktop computers, etc.

[0052] The server 105 can be a server that provides various services, such as a background management server that provides support for a website browsed by a user using the terminal device 101, 102, or 103 (as an example). The background management server can analyze and process received user requests and the like, and feed back the processing results (such as a webpage, information, or data obtained or generated according to a user request) to the terminal device.

[0053] It should be noted that the product recommendation method based on click rate provided by the embodiments of the present disclosure can generally be executed by the server 105. Accordingly, the product recommendation device based on click rate provided by the embodiments of the present disclosure can generally be arranged in the server 105. The product recommendation method based on click rate provided by the embodiments of the present disclosure can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the terminal device 101, 102, or 103 and / or the server 105. Accordingly, the product recommendation device based on click rate provided by the embodiments of the present disclosure can also be arranged in a server or a server cluster different from the server 105 and capable of communicating with the terminal device 101, 102, or 103 and / or the server 105.

[0054] It should be understood that the number of terminal devices, networks, and servers in the system 100 is merely illustrative. Any number of terminal devices, networks, and servers can be provided according to implementation needs. Figure 1

[0055] The product recommendation method based on click rate provided by the embodiments of the present disclosure will be described in detail below based on the application scenarios described above. Figure 1 Figures 2-8 The product recommendation method based on click rate provided by the embodiments of the present disclosure will be described in detail below based on the application scenarios described above.

[0056] Figure 2 An illustrative flowchart of the product recommendation method based on click rate according to the embodiments of the present disclosure is shown.

[0057] As shown in FIG. 2, the product recommendation method based on click rate of this embodiment can include operations S210-S260. Figure 2 In operation S210, a plurality of historical transaction data of a user purchasing a product is obtained to form a training data set.

[0058]

[0059] ​​​In the embodiments of the present disclosure, before the plurality of historical transaction data of the product purchased by the user is acquired, the method further includes: acquiring authorization of the user for the plurality of historical transaction data of the product purchased by the user; and acquiring the plurality of historical transaction data after the authorization of the user is obtained. Thus, before the plurality of historical transaction data of the product purchased by the user is acquired, the user is requested to acquire the plurality of historical transaction data of the product purchased by the user. In the case that the user agrees or authorizes the plurality of historical transaction data of the product purchased by the user to be acquired, the operation S210 is performed.

[0060] In operation S220, m products to be recommended are acquired, m is an integer greater than or equal to 2.

[0061] The number m of products is pre-set, and the embodiments mainly recommend the m products to the user in the best order to meet the interests and needs of the user, improve the marketing efficiency, and avoid interference of invalid recommended information on the user.

[0062] In operation S230, product features and user behavior features of each of the m products are extracted according to the training data set, and m feature vectors are generated.

[0063] In operation S240, the posterior probability corresponding to each of the m feature vectors is calculated using the Naive Bayes algorithm, and a first ranking list of the m products is obtained according to the posterior probability.

[0064] In operation S250, the m feature vectors are input into a deep learning network model to obtain click rate estimates of the m products, and a second ranking list of the m products is obtained according to the click rate estimates, wherein the deep learning network model is pre-trained based on the training data set.

[0065] In operation S260, the first ranking list and the second ranking list are corrected to obtain a third ranking list, and the m products are recommended in turn according to the third ranking list.

[0066] Through the embodiments of the present disclosure, the method combines traditional machine learning and deep learning to estimate the click rate of a plurality of products to be marketed, and then recommends the product information. Based on the ranking lists obtained by the two different algorithms of posterior probability and deep neural network, the final ranking list is obtained to recommend the products to the user in turn, and the product marketing efficiency and quality are improved. Moreover, the method does not need to perform artificial feature screening, can process the relationship between shallow features and the relationship between deep features, and further corrects the estimated click rate given by the model, which can improve the accuracy of the recommendation.

[0067] In the embodiments of the present disclosure, the product features of each product include a plurality of sub-features, and the sub-features include product type, product praise degree, product sales, marketing novelty and discount degree, each sub-feature includes a plurality of attribute values. The user behavior feature of each product is a discrete feature, indicating whether the user has purchased the product.

[0068] Specifically, the following gives a formal description of the technical problem to be solved in the embodiments. For the base insurance product, it is assumed that the user has purchased one or more products, and after the user's authorization, a plurality of historical transaction data are obtained. For each piece of historical transaction data, the product features and user behavior features of the product can be extracted. For the convenience of description, the product features and user behavior features of the product are formally represented as:

[0069]

[0070] wherein N is the total number of historical transaction data obtained; n is the number of sub-features included in a product feature of a piece of historical transaction data, n≥2 and is an integer; x i is the overall feature of the i th product; x i (j) is the j th sub-feature of the product i, j=1, 2, …, n, x i (j) ∈{a j1 , a j2 , …, a jp}, wherein a jl is the l th attribute value under the j th sub-feature, l=1, 2, …, p; p is the number of attribute values under the j th sub-feature, p is a positive integer; y i is the user behavior feature corresponding to the product i, which is a discrete feature, taking the value of -1 or +1; y i =+1 indicates that the user has clicked to purchase the product i; y i =-1 indicates that the user has not clicked to purchase the product i. Then, the recommendation problem of the base insurance product can be formally represented as:

[0071] Input:

[0072] (1) A plurality of historical transaction data of the user purchasing the product, constituting a training data set T:

[0073] T={(x1, y1), (x2, y2), …, (x N , y N )}

[0074] (2) The product features of m base insurance products to be recommended and the user behavior features of the current given user, generating m feature vectors:

[0075]

[0076] wherein x t is the tth feature vector, corresponding to the tth product to be recommended, t = 1, 2, …, m. At this time, the product to be recommended has not been sorted, so the serial number t is only obtained according to the generation order.

[0077] (3) Product identifiers id1, id2, …, id m of the m base insurance products to be recommended. The product identifiers of the m base insurance products to be recommended are in one-to-one mapping with the m feature vectors in (2).

[0078] Output:

[0079] A recommended list of length m, the content of the recommended list is the product identifier corresponding to the m products to be recommended according to the estimated click rate from high to low.

[0080] Based on the formal description of the above problem, the embodiments of the present disclosure first use the Naive Bayes algorithm to calculate the posterior probability corresponding to the m feature vectors respectively, and obtain the first sorting list of the m base insurance products to be recommended based on the size of the posterior probability; then, the deep neural network algorithm is used to train the training data set and obtain the trained deep learning network model, and the deep learning network model is used to estimate the click rate of the m base insurance products respectively, and the second sorting list of the m products is obtained based on the size of the estimated click rate; finally, the sorting lists obtained based on the two different algorithms are corrected to obtain the final sorting list, so as to sequentially recommend the m products.

[0081] Figure 3 The flowchart of the posterior probability calculation process according to the embodiments of the present disclosure is schematically shown.

[0082] As Figure 3 shown, in the embodiments of the present disclosure, the operation of using the Naive Bayes algorithm to calculate the posterior probability corresponding to the m feature vectors in the above operation S240 can specifically include operations S2401-S2403.

[0083] In operation S2401, the prior probability of each discrete feature in the training data set is calculated.

[0084] The prior probability is calculated according to the following formula:

[0085]

[0086]

[0087] wherein P(Y = +1) represents the prior probability that the user has purchased the product; P(Y = -1) represents the prior probability that the user has not purchased the product; I(y irepresents the number of times that the user has purchased the product i in the plurality of historical transaction data; I(y i represents the number of times that the user has not purchased the product i in the plurality of historical transaction data.

[0088] In operation S2402, the conditional probability of different attribute values of each sub-feature under each discrete feature is calculated according to the prior probability.

[0089] The conditional probability is specifically calculated according to the following formula:

[0090]

[0091]

[0092] wherein, P(X (j) = a jl represents the conditional probability of the lth attribute value under the jth sub-feature in the case that the user has purchased the product i; P(X (j) = a jl represents the conditional probability of the lth attribute value under the jth sub-feature in the case that the user has not purchased the product i; represents the lth attribute value under the jth sub-feature in the case that the user has not purchased the product i; represents the lth attribute value under the jth sub-feature in the case that the user has purchased the product i.

[0093] Further, based on the calculation result, when the prior probability or the conditional probability is 0, a Laplace smoothing method can be used for preprocessing.

[0094] In operation S2403, the posterior probability of each product in the m products belonging to each discrete feature is calculated according to the prior probability and the conditional probability.

[0095] Specifically, for the kth feature vector in the given m feature vectors:

[0096]

[0097] The posterior probability of the kth product to be recommended is calculated, and the formula is as follows:

[0098]

[0099]

[0100] wherein, P(Y = +1 | X = x k )、 represents the posterior probability of the kth product belonging to the first discrete feature; P(Y = -1 | X = xk represents the posterior probability that the kth product belongs to the second discrete feature.

[0101] Figure 4 A flowchart of a first ranking list determination process according to an embodiment of the present disclosure is schematically shown.

[0102] For ease of description, in embodiments of the present disclosure, when the discrete features include a first discrete feature and a second discrete feature, the first discrete feature represents that a user has purchased a corresponding product, the second discrete feature represents that a user has not purchased a corresponding product, and m products include a first product, as shown in Figure 4 After the posterior probability that the first product belongs to each discrete feature is calculated in operation S240, the above operation S240 can further include operation S2404 to operation S2405.

[0103] In operation S2404, it is determined whether the posterior probability that the first product belongs to the first discrete feature is greater than the posterior probability that the first product belongs to the second discrete feature. If yes, the posterior probability that belongs to the first discrete feature is determined as the probability that the first product is recommended; otherwise, the probability that the first product is recommended is determined according to the complementary event probability of the posterior probability that belongs to the second discrete feature.

[0104] In operation S2405, the first ranking list of the m products is determined according to the probability that the first product is recommended.

[0105] In embodiments of the present disclosure, when the posterior probability that the first product belongs to the first discrete feature is not greater than the posterior probability that the first product belongs to the second discrete feature, the probability that the first product is recommended is determined according to the following formula:

[0106]

[0107] wherein p k represents the probability that the first product is recommended; represents the posterior probability that the first product belongs to the second discrete feature.

[0108] Specifically, it is assumed that the posterior probability that the first product belongs to the first discrete feature is represented as the posterior probability that the first product belongs to the second discrete feature is represented as the probability that the first product is recommended is represented as p k If then let otherwise let

[0109] Then, for the m products to be recommended, the posterior probabilities calculated are sorted from large to small, and a probability list L p and a corresponding product identifier list L id, i.e. the first ranking list.

[0110] In the embodiments of the present disclosure, the deep learning network model is constructed according to a DeepFM network model, and an activation function σ2(x) of a hidden layer of the DeepFM network model is:

[0111]

[0112] wherein x represents an input of a hidden layer neuron; σ2(x) represents an output of the hidden layer neuron.

[0113] The activation function is different from an activation function of a conventional deep learning network model.

[0114] Figure 5 A flowchart of a deep learning network model training process according to an embodiment of the present disclosure is schematically shown.

[0115] As shown in Figure 5 , in the embodiments of the present disclosure, the deep learning network model in the operation S250 is trained through the following operations S2501-S2503.

[0116] In operation S2501, the training data set is split into a training set, a validation set and a test set.

[0117] In operation S2502, the training set and the validation set are used to train the constructed deep learning network model, and the network parameters of the deep learning network model are updated according to a preset loss function.

[0118] In operation S2503, the test set is used to evaluate the effectiveness of the deep learning network model.

[0119] Thus, the training data set T = {(x1, y1), (x2, y2),..., (xm, ym)} is split into a training set, a validation set and a test set for model training and evaluation. N N

[0120] Next, for m products to be recommended, the trained deep learning network model is used to obtain the click rate estimates of the m products.

[0121] Then, the click rate estimates are sorted from large to small, and a probability list L p 2 and a corresponding product identifier list L id 2 , i.e. the second ranking list.

[0122] Figure 6 A flowchart of a third ranking list correction process according to an embodiment of the present disclosure is schematically shown. ​​

[0123] As Figure 6 shown in the above operation S260, the first sorting list and the second sorting list are modified to obtain a third sorting list, which can specifically include operation S2601.

[0124] In operation S2601, it is determined whether the first sorting list and the second sorting list are completely consistent. If yes, the first sorting list or the second sorting list is determined as the third sorting list; otherwise, each element in the first sorting list and the second sorting list is multiplied to obtain the third sorting list.

[0125] Specifically, for the first sorting list L id and the second sorting list L id 2 , the comparison result is as follows:

[0126] (1) If they are completely consistent, then L id is the final sorting list;

[0127] (2) Otherwise, each element in the first sorting list L id and the second sorting list L id 2 is multiplied to obtain a probability list P The probabilities in P are sorted from large to small, and the product identifier list L recorded after sorting is the final sorting list.

[0128] Figure 7 The flowchart of the product identifier mapping establishment process according to an embodiment of the present disclosure is schematically shown.

[0129] As Figure 7 shown in the above operation S230, after the m feature vectors are generated, operation S2301 can be further included.

[0130] In operation S2301, according to the training data set, the product identifiers of the m products are extracted, and a one-to-one mapping between the product identifiers of the m products and the m feature vectors is established.

[0131] Figure 8 The flowchart of the product identifier list recommendation process according to an embodiment of the present disclosure is schematically shown.

[0132] As Figure 8 shown, further, in the above operation S260, according to the third sorting list, the m products are recommended in turn, which can further include operation S2602.

[0133] In operation S2602, the product identifier list mapped to the third sorting list is recommended in turn.

[0134] Therefore, the product information recommended to the user, specifically, the product identification list, is intuitive and concise.

[0135] Based on the product recommendation method based on click rate described above, the disclosure further provides a product recommendation device based on click rate, which will be described below in combination with Figure 9 The device will be described in detail.

[0136] Figure 9 The block diagram of the product recommendation device based on click rate according to the embodiment of the disclosure is schematically shown.

[0137] As Figure 9 shown, the product recommendation device based on click rate 900 of the embodiment includes a data acquisition module 910, a product acquisition module 920, a vector generation module 930, a first sorting module 940, a second sorting module 950, and a product recommendation module 960.

[0138] The data acquisition module 910 is configured to acquire a plurality of historical transaction data of the user purchasing products to form a training data set. In an embodiment, the data acquisition module 910 can be configured to perform the operation S210 described above, and details are not repeated here.

[0139] The product acquisition module 920 is configured to acquire m products to be recommended, where m is an integer greater than or equal to 2. In an embodiment, the product acquisition module 920 can be configured to perform the operation S220 described above, and details are not repeated here.

[0140] The vector generation module 930 is configured to extract product features and user behavior features of each of the m products according to the training data set to generate m feature vectors. In an embodiment, the vector generation module 930 can be configured to perform the operation S230 described above, and details are not repeated here.

[0141] The first sorting module 940 is configured to calculate the posterior probability corresponding to each of the m feature vectors using the Naive Bayes algorithm, and obtain a first sorting list of the m products according to the posterior probability. In an embodiment, the first sorting module 940 can be configured to perform the operation S240 described above, and details are not repeated here.

[0142] The second sorting module 950 is configured to input the m feature vectors into a deep learning network model to obtain click rate estimates of the m products, and obtain a second sorting list of the m products according to the click rate estimates, where the deep learning network model is pre-trained based on the training data set. In an embodiment, the second sorting module 950 can be configured to perform the operation S250 described above, and details are not repeated here.

[0143] The product recommendation module 960 is configured to correct the first sorting list and the second sorting list to obtain a third sorting list, and recommend the m products in sequence according to the third sorting list. In an embodiment, the product recommendation module 960 can be configured to perform the operation S260 described above, and details are not described herein again.

[0144] According to embodiments of the present disclosure, any one or more of the data acquisition module 910, the product acquisition module 920, the vector generation module 930, the first sorting module 940, the second sorting module 950 and the product recommendation module 960 can be combined in one module, or any one of them can be split into multiple modules. Alternatively, at least part of the function of one or more of these modules can be combined with at least part of the function of other modules, and implemented in one module. According to embodiments of the present disclosure, at least one of the data acquisition module 910, the product acquisition module 920, the vector generation module 930, the first sorting module 940, the second sorting module 950 and the product recommendation module 960 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on board, a system on package, an application specific integrated circuit (ASIC), or any other reasonable way of hardware or firmware that can be integrated or packaged, or any one of software, hardware and firmware or any appropriate combination of any of them. Alternatively, at least one of the data acquisition module 910, the product acquisition module 920, the vector generation module 930, the first sorting module 940, the second sorting module 950 and the product recommendation module 960 can be at least partially implemented as a computer program module which can perform corresponding functions when it is run.

[0145] Figure 10 A block diagram of an electronic device suitable for implementing the click rate based product recommendation method according to embodiments of the present disclosure is schematically shown.

[0146] As shown in Figure 10 The electronic device 1000 according to embodiments of the present disclosure includes a processor 1001 which can perform various appropriate actions and processes according to programs stored in a read only memory (ROM) 1002 or loaded from a storage portion 1008 into a random access memory (RAM) 1003. The processor 1001 can include, for example, a general purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special purpose microprocessor (such as an application specific integrated circuit (ASIC)), and the like. The processor 1001 can also include an on-board memory for cache use. The processor 1001 can include a single processing unit or multiple processing units for performing different actions of the method processes according to embodiments of the present disclosure.

[0147] In the RAM 1003, various programs and data required by the electronic device 1000 to operate are stored. The processor 1001, the ROM 1002, and the RAM 1003 are connected to each other via the bus 1004. The processor 1001 performs various operations of the method flow according to the embodiments of the present disclosure by executing the programs in the ROM 1002 and / or the RAM 1003. It needs to be noted that the programs can also be stored in one or more memories other than the ROM 1002 and the RAM 1003. The processor 1001 can also perform various operations of the method flow according to the embodiments of the present disclosure by executing the programs stored in the one or more memories.

[0148] According to an embodiment of the present disclosure, the electronic device 1000 can further include an input / output (I / O) interface 1005, which is also connected to the bus 1004. The electronic device 1000 can further include one or more of the following components connected to the I / O interface 1005: an input part 1006 including a keyboard, a mouse, etc.; an output part 1007 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 1008 including a hard disk, etc.; and a communication part 1009 including a network interface card such as a LAN card, a modem, etc. The communication part 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as necessary. A removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is mounted on the drive 1010 as necessary, so that a computer program read out therefrom is installed in the storage part 1008 as necessary.

[0149] The present disclosure also provides a computer readable storage medium, which can be included in the device / apparatus / system described in the above embodiments; or can exist separately without being assembled into the device / apparatus / system. The above computer readable storage medium carries one or more programs, when the one or more programs are executed, realizing the click rate based product recommendation method according to the embodiments of the present disclosure.

[0150] According to an embodiment of the present disclosure, the computer readable storage medium can be a nonvolatile computer readable storage medium, for example, can include, but is not limited to, 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), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any appropriate combination thereof. In the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, the computer readable storage medium can include one or more of the ROM 1002 and / or the RAM 1003 described above and / or one or more memory other than the ROM 1002 and the RAM 1003.

[0151] Embodiments of the present disclosure also include a computer program product, which includes a computer program containing program codes for executing the methods shown in the flowcharts. When the computer program product is run in a computer system, the program codes are used to make the computer system implement the click rate based product recommendation method provided by the embodiments of the present disclosure.

[0152] The above-described functions defined in the system / device of the embodiments of the present disclosure are performed when the computer program is executed by the processor 1001. According to an embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by computer program modules.

[0153] In one embodiment, the computer program can rely on a tangible storage medium such as an optical storage device, a magnetic storage device, etc. In another embodiment, the computer program can also be transmitted, distributed, and downloaded in the form of a signal on a network medium, and be downloaded and installed through the communication part 1009 and / or installed from the detachable medium 1011. The program codes contained in the computer program can be transmitted by any appropriate network medium, including but not limited to wireless, wired, etc., or any appropriate combination thereof.

[0154] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009 and / or installed from the detachable medium 1011. When the computer program is executed by the processor 1001, the above-described functions defined in the system of the embodiments of the present disclosure are performed. According to an embodiment of the present disclosure, the system, device, apparatus, module, unit, etc. described above can be implemented by computer program modules.

[0155] According to embodiments of the present disclosure, program code of the computer program for performing the methods provided by the embodiments of the present disclosure can be written in any combination of one or more programming languages, and can be implemented in a computer program product. Specifically, the computer program can be implemented in a high-level procedural and / or object-oriented programming language, and / or in assembly / machine language. The programming language includes, but is not limited to, Java, C++, python, “C” language, or similar programming languages. The program code can execute entirely on the user's computing device, partly on the user's device, and partly on a remote computing device, or entirely on the remote computing device or server. In the latter scenario, the remote computing device can be connected to the user's computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computing device, such as through the Internet using an Internet Service Provider (ISP).

[0156] The computer program product of the present disclosure can be a computer program product, which is a machine-readable medium (or computer readable medium) having stored therein a sequence of instructions readable by a machine (e.g., a computer). The instructions in the machine-readable medium focus on the function of the computer program (e.g., in the form of a program unit) and on the cooperation behavior between such program units. The program can be distributed over networks coupled to one or more processors or computers, and can be stored on computer readable media including computer program products, hard drives, RAMs, or any other device (internal or external) from which a processor or computer reads incoming data or instructions. In addition to being distributed over the Internet, or over any other publicly or privately accessed network, the aforementioned program product can also be stored in the memory of the respective machine and / or processor.

[0157] Those skilled in the art will understand that features of the various embodiments and / or claims of the present disclosure can be combined or / and integrated with one another, even though such combinations or integrations are not expressly disclosed in the present disclosure. In particular, features of the various embodiments and / or claims of the present disclosure can be combined and / or integrated with one another, without departing from the spirit and teachings of the present disclosure. All such combinations and / or integrations are within the scope of the present disclosure.

[0158] The above describes embodiments of the present disclosure. However, these embodiments are merely for illustrative purposes, and are not intended to limit the scope of the present disclosure. Although each embodiment is described above separately, this does not mean that the measures in each embodiment cannot be used advantageously in combination. The scope of the present disclosure is defined by the appended claims and their equivalents. Those skilled in the art can make various substitutions and modifications without departing from the scope of the present disclosure, and these substitutions and modifications should all fall within the scope of the present disclosure.

Claims

1. A product recommendation method based on click-through rate, comprising: Obtain multiple historical transaction records of user product purchases to form a training dataset; Get m products to be recommended, where m ≥ 2 and is an integer; Based on the training dataset, extract the product features and user behavior features of each of the m products to generate m feature vectors; The Naive Bayes algorithm is used to calculate the posterior probabilities corresponding to the m feature vectors respectively, and the first ranking list of the m products is obtained based on the posterior probabilities. The m feature vectors are input into a deep learning network model to obtain the click-through rate (CTR) estimates of the m products. Based on the CTR estimates, a second ranking list of the m products is obtained. The deep learning network model is pre-trained based on the training dataset. Determine whether the first sorted list and the second sorted list are completely identical. If they are, then determine the first sorted list or the second sorted list as the third sorted list; otherwise, multiply each element in the first sorted list and the second sorted list to obtain the third sorted list; and recommend the m products in sequence according to the third sorted list. The product features of each product include multiple sub-features, and each sub-feature includes multiple attribute values; the user behavior features of each product are discrete features, representing whether the user has purchased the product. The step of calculating the posterior probabilities corresponding to the m feature vectors using the Naive Bayes algorithm includes: Calculate the prior probability of each of the discrete features in the training dataset; Based on the prior probability, calculate the conditional probability of different attribute values ​​for each of the sub-features under each of the discrete features; Based on the prior probability and conditional probability, calculate the posterior probability of each of the m products belonging to each of the discrete features; When the discrete features include a first discrete feature and a second discrete feature, where the first discrete feature indicates that the user has purchased the corresponding product and the second discrete feature indicates that the user has not purchased the corresponding product, and when the m products include the first product, after calculating the posterior probability of the first product belonging to each of the discrete features, the method further includes: Determine whether the posterior probability of the first product belonging to the first discrete feature is greater than the posterior probability of the first product belonging to the second discrete feature. If so, the posterior probability of belonging to the first discrete feature is determined as the probability of the first product being recommended. Otherwise, the probability of the first product being recommended is determined based on the complementary event probability of the posterior probability of belonging to the second discrete feature. Based on the probability that the first product is recommended, determine the first sorted list of the m products; When the posterior probability of the first product belonging to the first discrete feature is not greater than the posterior probability of the first product belonging to the second discrete feature, the probability of the first product being recommended is determined according to the following formula: in, This indicates the probability that the first product will be recommended; This represents the posterior probability that the first product belongs to the second discrete feature.

2. The method according to claim 1, wherein the sub-features include product type, product satisfaction rating, product sales volume, marketing novelty, and discount level.

3. The method according to claim 1, wherein, When the prior probability or conditional probability is 0, the Laplace smoothing method is used for preprocessing.

4. The method according to claim 1, wherein, The deep learning network model is constructed based on the DeepFM network model, and the activation function of the hidden layer of the DeepFM network model is... for: Where x represents the input to the hidden layer neuron; This represents the output of neurons in the hidden layer.

5. The method according to claim 4, wherein, The deep learning network model is trained in the following way: The training dataset is split into a training set, a validation set, and a test set; The constructed deep learning network model is trained using the training set and validation set, and the network parameters of the deep learning network model are updated according to a preset loss function. The effectiveness of the deep learning network model is evaluated using the test set.

6. The method according to claim 1, wherein, After generating m feature vectors, the process also includes: Based on the training dataset, extract the product identifiers of the m products and establish a one-to-one mapping between the product identifiers of the m products and the m feature vectors.

7. The method according to claim 6, wherein, Based on the third sorting list, the m products are recommended sequentially, including: The product identifier list is recommended sequentially by mapping the third sorting list.

8. The method according to claim 1, wherein, The method further includes: Authorize the user to obtain multiple historical transaction data related to their purchased products; After obtaining authorization from the user, the multiple historical transaction data are retrieved.

9. A product recommendation device based on click-through rate, comprising: The data acquisition module is used to acquire multiple historical transaction data of users' product purchases to form a training dataset; The product acquisition module is used to acquire m products to be recommended, where m ≥ 2 and is an integer; The vector generation module is used to extract the product features and user behavior features of each of the m products from the training dataset and generate m feature vectors. The first sorting module is used to calculate the posterior probabilities corresponding to the m feature vectors using the Naive Bayes algorithm, and to obtain a first sorting list of the m products based on the posterior probabilities. The second ranking module is used to input the m feature vectors into a deep learning network model to obtain the click-through rate (CTR) estimates of the m products, and to obtain a second ranking list of the m products based on the CTR estimates. The deep learning network model is pre-trained based on the training dataset. The product recommendation module is used to determine whether the first sorting list and the second sorting list are completely identical. If they are, the first sorting list or the second sorting list is determined as the third sorting list. Otherwise, the elements in the first sorting list and the second sorting list are multiplied together to obtain the third sorting list. Based on the third sorting list, the m products are recommended in sequence. The product features of each product include multiple sub-features, and each sub-feature includes multiple attribute values; the user behavior features of each product are discrete features, representing whether the user has purchased the product. The step of calculating the posterior probabilities corresponding to the m feature vectors using the Naive Bayes algorithm includes: Calculate the prior probability of each of the discrete features in the training dataset; Based on the prior probability, calculate the conditional probability of different attribute values ​​for each of the sub-features under each of the discrete features; Based on the prior probability and conditional probability, calculate the posterior probability of each of the m products belonging to each of the discrete features; When the discrete features include a first discrete feature and a second discrete feature, where the first discrete feature indicates that the user has purchased the corresponding product and the second discrete feature indicates that the user has not purchased the corresponding product, and when the m products include the first product, after calculating the posterior probability of the first product belonging to each of the discrete features, the method further includes: Determine whether the posterior probability of the first product belonging to the first discrete feature is greater than the posterior probability of the first product belonging to the second discrete feature. If so, the posterior probability of belonging to the first discrete feature is determined as the probability of the first product being recommended. Otherwise, the probability of the first product being recommended is determined based on the complementary event probability of the posterior probability of belonging to the second discrete feature. Based on the probability that the first product is recommended, determine the first sorted list of the m products; When the posterior probability of the first product belonging to the first discrete feature is not greater than the posterior probability of the first product belonging to the second discrete feature, the probability of the first product being recommended is determined according to the following formula: in, This indicates the probability that the first product will be recommended; This represents the posterior probability that the first product belongs to the second discrete feature.

10. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 8.

12. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images