Merchant recommendation method, device, electronic device and readable storage medium

Abstract

The present disclosure discloses a merchant recommendation method, device, electronic device and readable storage medium. The method includes: obtaining click sequence information of a target user; obtaining the recommendation score of each candidate merchant through a preset recommendation model according to the click sequence information; obtaining the target recommended merchant of the target user based on the recommendation score of the candidate merchant and pushing it to the target user; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model. This solves the technical problem of poor accuracy of existing recommendation methods. The beneficial effect of improving the accuracy of recommended merchants is achieved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of machine learning, and in particular to a merchant recommendation method, device, electronic device, and readable storage medium. Background Art

[0002] With the rapid development of the Internet and machine learning technology, more and more e-commerce platforms recommend personalized merchants that meet users' needs through recommendation systems. The recommendation system generally ranks the candidate merchants for the current user to determine the final recommended merchant.

[0003] At present, the application of machine learning model recommendation scenarios has achieved certain results, such as FNN (Feedforward Neural Network), RNN (Recurrent Neural Network) and DNN (Deep Neural Networks) networks have achieved certain results. At present, there are also methods in the industry to add sequence information to Xgb (eXtreme Gradient Boosting) and DNN for recommendation, but it is necessary to train the RNN separately and then input the features learned by the RNN into the DNN. When the machine learning models are trained separately, the training coordinate systems of the two models may be different, which can easily lead to the invalid training of the learned user sequence information in the machine learning model, thereby affecting the accuracy of the recommendation results. It can be seen that the existing recommendation schemes still have the problem of poor accuracy of recommendation results. Summary of the invention

[0004] The present disclosure provides a merchant recommendation method, device, electronic device and readable storage medium to partially or completely solve the above-mentioned problems related to the merchant recommendation process in the prior art.

[0005] According to a first aspect of the present disclosure, a merchant recommendation method is provided, comprising:

[0006] Obtain click sequence information of target users;

[0007] According to the click sequence information, a recommendation score of each candidate merchant is obtained through a preset recommendation model;

[0008] Based on the recommendation scores of the candidate merchants, the target recommended merchants of the target user are obtained and pushed to the target user;

[0009] Among them, the recommendation model is a combined model obtained by jointly training a two-layer recurrent neural network model and a deep neural network model, the two-layer recurrent neural network model includes a first-layer recurrent neural network model and a second-layer recurrent neural network model, and the input data of the second-layer recurrent neural network model includes the output data of the first-layer recurrent neural network model.

[0010] According to a second aspect of the present disclosure, there is provided a merchant recommendation device, comprising:

[0011] A click sequence information acquisition module is used to obtain the click sequence information of the target user;

[0012] A recommendation score acquisition module is used to acquire the recommendation score of each candidate merchant according to the click sequence information through a preset recommendation model;

[0013] A target recommended merchant acquisition module is used to acquire the target recommended merchant of the target user based on the recommendation score of the candidate merchant and push it to the target user;

[0014] Among them, the recommendation model is a combined model obtained by jointly training a two-layer recurrent neural network model and a deep neural network model, the two-layer recurrent neural network model includes a first-layer recurrent neural network model and a second-layer recurrent neural network model, and the input data of the second-layer recurrent neural network model includes the output data of the first-layer recurrent neural network model.

[0015] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0016] A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the aforementioned merchant recommendation method when executing the program.

[0017] According to a fourth aspect of the present disclosure, a readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the aforementioned merchant recommendation method.

[0018] According to the merchant recommendation method disclosed in the present invention, the click sequence information of the target user can be obtained; according to the click sequence information, the recommendation score of each candidate merchant can be obtained through a preset recommendation model; based on the recommendation score of the candidate merchant, the target recommended merchant of the target user is obtained and pushed to the target user; wherein, the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model. This solves the technical problem of poor accuracy of the existing recommendation method. The beneficial effect of improving the accuracy of recommended merchants is achieved.

[0019] The above description is only an overview of the technical solution of the present disclosure. In order to more clearly understand the technical means of the present disclosure, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present disclosure more obvious and easy to understand, the specific implementation methods of the present disclosure are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present disclosure. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0021] Figure 1 A flowchart of the steps of a merchant recommendation method according to an embodiment of the present disclosure is shown;

[0022] Figure 2 A flowchart showing the steps of another merchant recommendation method according to an embodiment of the present disclosure is shown;

[0023] Figure 3 A schematic diagram showing click sequence information according to an embodiment of the present disclosure is shown;

[0024] Figure 4 A schematic diagram showing the structure of a merchant recommendation device according to an embodiment of the present disclosure; and

[0025] Figure 5 A schematic structural diagram of another merchant recommendation device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0026] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0027] Embodiment 1

[0028] A recommended method provided by an embodiment of the present disclosure is described in detail.

[0029] Reference Figure 1 , showing a step flow chart of a recommendation method in an embodiment of the present disclosure.

[0030] Step 110: Obtain click sequence information of the target user.

[0031] In the disclosed embodiment, in order to recommend target recommended merchants that meet the current status of the target user, it is necessary to obtain relevant reference information for determining the target recommended merchants. In addition, in practical applications, the merchants that the user browsed by clicking or other methods in the historical time can reflect his current needs to a certain extent. Therefore, in the disclosed embodiment, the click sequence information of the target user can be obtained as reference information for determining the target recommended merchants.

[0032] The click sequence information may include feature information of each merchant browsed by the target user by clicking or the like within a preset time period, user information of the corresponding target user, interaction information between the user and each merchant, etc. The click sequence information may include features describing the user, merchant, context, and the interaction relationship between the user and the merchant. The specific content contained in the click sequence information and the data arrangement method in the click sequence information may be preset according to the needs, and the embodiments of the present disclosure are not limited to this.

[0033] Step 120, according to the click sequence information, obtain the recommendation score of each candidate merchant through a preset recommendation model; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model.

[0034] After obtaining the click sequence information of the target user, the recommendation score of each candidate merchant can be obtained through the preset recommendation model. The specific content of the candidate merchant can be preset according to the needs, and this embodiment of the disclosure is not limited to this.

[0035] Furthermore, when obtaining the recommendation scores of each candidate merchant, the parameters input into the recommendation model may include click sequence information and attribute information of each candidate merchant. The specific content of the attribute information may also be preset according to the requirements, which is not limited in the embodiment of the present disclosure.

[0036] For example, in the takeaway merchant recommendation scenario, the attribute information may include 7 discrete features and 119 continuous features. The corresponding target user's click sequence information may also include the 7 discrete features and 119 continuous features of the takeaway merchants clicked by the target user within a preset time period, as well as the dish sequence of the corresponding takeaway merchant, and so on.

[0037] Furthermore, the recommendation score of each candidate merchant can be obtained according to the click sequence information of the target user and the attribute information of each candidate merchant through a preset recommendation model.

[0038] The recommended model is a combination model obtained by jointly training a two-layer recurrent neural network (RNN) model and a deep neural network (DNN) model. A deep neural network literally means a deep neural network.

[0039] A recurrent neural network is a type of recurrent neural network that takes sequence data as input, performs recursion in the direction of sequence evolution, and all nodes (recurrent units) are connected in a chain. Recurrent neural networks have memory, parameter sharing, and Turing completeness, so they can learn the nonlinear features of sequences with high efficiency. RNN models can include many types, such as LSTM (Long Short Term Memory) network models, GRU network models, bidirectional RNN models, and so on.

[0040] The double-layer recurrent neural network model can be understood as a neural network model constructed by two recurrent neural network models. For example, assuming that the double-layer recurrent neural network model includes a first-layer recurrent neural network model and a second-layer recurrent neural network model, and the input data of the second-layer recurrent neural network model includes the output data of the first-layer recurrent neural network model. At this time, the input end of the first-layer recurrent neural network model can be set as the input end of the corresponding double-layer recurrent neural network model, the output end of the first-layer recurrent neural network model is connected to the input end of the second-layer recurrent neural network model, and the output end of the second-layer recurrent neural network model is used as the output end of the corresponding double-layer recurrent neural network model.

[0041] Deep neural network literally means deep neural network. In many fields, DNN can surpass human accuracy. The outstanding performance of DNN comes from its ability to extract high-level features from raw sensory data using statistical learning methods and obtain effective representation of input space in a large amount of data. DNN models can include deep convolutional neural network models, FNN (feedforward neural network) models, and so on.

[0042] In the embodiment of the present disclosure, since the click sequence information of the target user can be in the form of a sequence, the recommendation model can be set as a combination of a double-layer RNN model and a DNN model. Moreover, the recommendation model can include a double-layer RNN model composed of two RNN models, and at least one DNN model, which can be pre-set according to the needs, and the specific type of each RNN model, the specific number of DNN models, and the specific type of each DNN model can be pre-set according to the needs, which is not limited by the embodiment of the present disclosure.

[0043] Moreover, when training the recommendation model, a combined model obtained by jointly training the double-layer recurrent neural network model and the deep neural network model can be used as the recommendation model.

[0044] Step 130: based on the recommendation scores of the candidate merchants, obtain the target recommended merchants of the target user and push them to the target user.

[0045] After obtaining the recommendation scores of each candidate merchant, the target recommended merchant of the target user can be obtained according to the recommendation scores of the candidate merchants and pushed to the target user. In the embodiment of the present disclosure, the higher the recommendation score, the higher the matching degree between the corresponding candidate merchant and the target user. In this case, a recommendation score critical value can be set according to the demand, and then the candidate merchants with recommendation scores greater than or equal to the recommendation score critical value are obtained as the target recommended merchants of the target user, and then the target recommended merchants can be pushed to the target user; or, if the higher the recommendation score is set, the lower the matching degree between the corresponding candidate merchant and the target user is, then the candidate merchants with recommendation scores less than or equal to the recommendation score critical value can be obtained as the target recommended merchants of the target user, and then the target recommended merchants can be pushed to the target user.

[0046] Among them, any available method can be used to send the target recommended merchant to the target user, and the specific method of sending the target recommended merchant to the target user can also be preset according to needs, which is not limited in the embodiments of the present disclosure.

[0047] For example, for a food delivery platform, if the target recommended merchant is a food delivery merchant, the electronic business card of the corresponding food delivery merchant can be sent to the target user. If the target user clicks on the electronic business card, he or she can be redirected to the corresponding food delivery merchant.

[0048] In the disclosed embodiment, by obtaining the click sequence information of the target user; according to the click sequence information, obtaining the recommendation score of each candidate merchant through a preset recommendation model; based on the recommendation score of the candidate merchant, obtaining the target recommended merchant of the target user and pushing it to the target user; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model. The beneficial effect of improving the accuracy of recommended merchants is achieved.

[0049] Embodiment 2

[0050] A recommended method provided by an embodiment of the present disclosure is described in detail.

[0051] Reference Figure 2 , showing a step flow chart of a recommendation method in an embodiment of the present disclosure.

[0052] Step 210, obtaining characteristic information of the target user clicking on the merchant within a preset time period, and constructing click sequence information of the target user based on the characteristic information.

[0053] As mentioned above, in the embodiment of the present disclosure, the recommendation score of each candidate merchant is obtained based on the click sequence characteristics of the target user, so the acquisition of click sequence information is crucial. In the embodiment of the present disclosure, in order to improve the accuracy and completeness of the click sequence information, the characteristic information of the target user clicking on the merchant within a preset time period can be obtained, and the click sequence information of the target user is constructed based on the characteristic information.

[0054] Among them, the specific content contained in the feature information can be pre-set according to the needs, and this embodiment of the present disclosure does not limit this. For example, for the takeaway merchant recommendation scenario, the clicked merchant at this time can be a takeaway merchant, and the feature information can be set to include but not limited to the merchant characteristics of the takeaway merchant, dish characteristics, user-merchant cross-features, user characteristics of the target user, user characteristics of other users who have consumption records in the corresponding takeaway merchant, and so on. The merchant characteristics can include but are not limited to the merchant name, the category to which the merchant belongs, the merchant location, the merchant's sales volume, the merchant's evaluation, the merchant's per capita consumption level, and so on; the dish characteristics can include but are not limited to the dish name, dish price, dish quantity, dish sales, dish evaluation, and so on; the user characteristics can include but are not limited to the user's gender, occupation, age, consumption tendency, real-time location, and so on; the user-merchant cross-features can include the user's operation record for the corresponding merchant, the corresponding merchant's feedback on the user's operation, and so on.

[0055] Moreover, the arrangement order of each feature information in the click sequence information can also be pre-set according to the needs, which is not limited by the embodiments of the present disclosure. In order to improve the accuracy of the model, for the same recommendation model, the arrangement order of each feature information in the corresponding click sequence information can be set to be consistent. Of course, if necessary, it can also be set to be inconsistent, which is not limited by the embodiments of the present disclosure.

[0056] For example, assuming that for a takeaway scenario, 7 discrete features and 119 continuous features can be extracted for each clicked merchant, and the dish sequence of each clicked merchant can also be extracted, then if the arrangement order of the feature information for each clicked merchant in the click sequence information is set to dish sequence + discrete features + continuous features, then the following can be obtained: Figure 3 Click sequence information shown.

[0057] Optionally, in an embodiment of the present disclosure, the click sequence information includes a first feature sequence and a second feature sequence; the first feature sequence includes dish features, and the second feature sequence includes at least one of merchant features, user features, and user-merchant cross features.

[0058] The preset time period can be preset according to the needs, and the embodiments of the present disclosure are not limited to this. For example, according to different business needs, the characteristic information of the merchants clicked in different time periods can be selected, such as the characteristic information of the merchants clicked on the same day, or the characteristic information of all merchants clicked by the target user within one month before the current time, etc.

[0059] In actual applications, in ordinary e-commerce environments, there is only a simple demand for recommending products. In the takeaway scenario, the description of the merchant can be considered directly from the merchant's perspective, or the merchant can be described from the perspective of the dish. Therefore, in the disclosed embodiment, in order to facilitate the description of the corresponding merchant based on the dish features, the dish features can be separated from the click sequence information for separate learning. Then at this time, the click sequence information can be set to include a first feature sequence and a second feature sequence, and the first feature sequence includes dish features, and the second feature sequence includes at least one of merchant features, user features, and user-merchant cross features.

[0060] Step 220, based on the preset training sample data, jointly train the recommendation model obtained by combining the double-layer recurrent neural network model and the deep neural network model; wherein, in each training process, the input data of the double-layer recurrent neural network model includes the sample sequence information in the training sample data, and the input data of the deep neural network model includes the output data of the double-layer recurrent neural network model, the feature sequence of the sample merchant corresponding to the sample sequence information, and the label value corresponding to the sample merchant.

[0061] In order to use the recommendation model to obtain the recommendation score of each candidate merchant, it is necessary to first train the corresponding recommendation model. Specifically, the recommendation model obtained by combining the two-layer recurrent neural network model and the deep neural network model can be jointly trained according to preset training sample data.

[0062] Among them, each training sample data may include a sample sequence information of a sample user's historical clicks on merchants, a feature sequence of the sample merchants corresponding to the corresponding sample users, and a label value of the corresponding sample user for the corresponding sample merchant. The label value has different meanings in different business scenarios. For example, in the scenario of predicting CTR (Click-Through-Rate), the label value indicates whether it is clicked or not. In the scenario of predicting CVR (Conversion Rate), the label value indicates whether it is purchased. Generally speaking, if a click or purchase operation is performed, the corresponding label value can be set to 1, otherwise the corresponding label value can be set to 0. The sample sequence information can refer to the above-mentioned click sequence information, which will not be elaborated here.

[0063] Moreover, in order to improve the accuracy of the recommendation model, the training sample data needs to include positive and negative sample data. In each positive training sample data, the corresponding sample merchant is the merchant finally selected by the corresponding sample user, so the label value of the corresponding sample merchant can be 1; and in each negative training sample data, the corresponding sample merchant can be any merchant that the corresponding sample user has not finally selected, so the label value of the corresponding sample merchant can be 0.

[0064] Then, in the process of jointly training the recommendation model using each training sample data, the input data of the two-layer recurrent neural network model includes the sample sequence information in the training sample data, and the input data of the deep neural network model includes the output data of the two-layer recurrent neural network model, the feature sequence of the sample merchant corresponding to the sample sequence information, and the label value of the sample merchant.

[0065] Step 230: input the click sequence information into the double-layer recurrent neural network model, and obtain the output vector of the double-layer recurrent neural network model.

[0066] Optionally, in the embodiment of the present disclosure, step 230 may further include:

[0067] Sub-step 231, inputting the first feature sequence into the first-layer recurrent neural network model, and obtaining a first output vector of the first-layer recurrent neural network model;

[0068] Sub-step 232, input the first output vector and the second feature sequence into the second-layer recurrent neural network model, and obtain the second output vector of the second-layer recurrent neural network model as the output vector of the double-layer recurrent neural network model.

[0069] As mentioned above, in an ordinary e-commerce environment, there is only a simple demand for recommending products, while in the takeaway scenario, the description of the merchant can be directly considered from the merchant's perspective, or the merchant can be described from the perspective of the dishes. Therefore, in the present embodiment, an RNN model is designed to describe the merchant's dishes. Therefore, the two-layer recurrent neural network model in the present embodiment includes a first-level recurrent neural network model and a second-level recurrent neural network model.

[0070] At this time, the first feature sequence including the characteristics of the dishes can be input into the first-level recurrent neural network model, and the first output vector of the first-level recurrent neural network model can be obtained. Then, the first output vector and the second feature sequence can be input into the second-level recurrent neural network model, and the second output vector of the second-level recurrent neural network model can be obtained as the output vector of the double-layer recurrent neural network model. Moreover, when the first feature sequence is input into the first-level RNN model, the characteristics of the dishes can be input into the first-level RNN model in the order of the click-through rate of the corresponding dishes from low to high, so that the dishes with high click-through rates can better represent the characteristics of the corresponding merchants. Of course, the characteristics of the dishes can also be input in other orders, which can be pre-set according to the needs, and the embodiments of the present disclosure are not limited to this.

[0071] Optionally, in the disclosed embodiment, the first-layer recurrent neural network model and the second-layer recurrent neural network model may include but are not limited to dynamic long short-term memory network models (bucketing LSTM / dynamicLSTM). Moreover, if the first-layer recurrent neural network model and the second-layer recurrent neural network model are both LSTM models, then the aforementioned two-layer recurrent neural network model may also be referred to as a two-layer dynamic LSTM model.

[0072] In practical applications, the RNN model is responsible for processing the sample sequence information in each training sample data, and the length of the feature sequence contained in the sample sequence information can vary. For a general RNN model, its input port is fixed. If the length of the feature sequence corresponding to the sample sequence information does not match the length of the input port of the RNN model, especially when the length of the feature sequence is less than the length of the input port of the RNN model, zeros will be padded for the corresponding feature sequence, which not only increases resource usage, but also introduces noise that easily reduces the accuracy of the recommendation model.

[0073] Therefore, in the disclosed embodiment, the first-level recurrent neural network model and the second-level recurrent neural network model can be set to include a dynamic long short-term memory network model. The dynamic LSTM can be adaptively adjusted according to the length of the input sequence. For example, if the input sequence length is 5, that is, the sample sequence information contains 5 feature sequences of click merchants, then the input port of the dynamic LSTM model becomes 5 sequences to receive each feature sequence accordingly; and if 10 sequence data are input, the dynamic LSTM will adaptively become 10.

[0074] Alternatively, according to requirements, only at least one of the first-level recurrent neural network model and the second-level recurrent neural network model may be set as a dynamic LSTM model, which is not limited in this embodiment of the present invention.

[0075] Optionally, in the embodiment of the present disclosure, the sub-step 232 may further include:

[0076] Sub-step 2321, based on a preset encoding matrix, and the coordinates of each discrete feature in the first output vector and the second feature sequence relative to the encoding matrix, obtaining a first embedding vector of the discrete feature relative to the encoding matrix;

[0077] Sub-step 2322, obtaining a second embedding vector of the first output vector and the second feature sequence according to the first embedding vector of each offline feature and the continuous features in the first output vector and the second feature sequence;

[0078] Sub-step 2323, input the second embedding vector into the second-layer recurrent neural network model, and obtain the second output vector of the second-layer recurrent neural network model as the output vector of the two-layer recurrent neural network model.

[0079] As mentioned above, in order to be able to process features other than the dish feature sequence, the second-level RNN model also needs to support the input of user features, user-merchant cross features, and merchant features. Therefore, the input data of the second-level RNN model is converted to meet actual business needs.

[0080] Specifically, based on a preset encoding matrix and the coordinates of each discrete feature in the first output vector and the second feature sequence relative to the encoding matrix, a first embedding vector of the discrete feature relative to the encoding matrix can be obtained, and then according to the first embedding vector of each offline feature and the continuous features in the first output vector and the second feature sequence, a second embedding vector of the first output vector and the second feature sequence can be obtained, and finally the second embedding vector is input into a second-layer recurrent neural network model, and the second output vector of the second-layer recurrent neural network model is obtained as the output vector of the double-layer recurrent neural network model.

[0081] The encoding matrix can be preset as required, which is not limited in the embodiment of the present disclosure, and the coordinates of each discrete feature relative to the encoding matrix can also be preset as required, which is not limited in the embodiment of the present disclosure.

[0082] For example, if a discrete feature corresponds to the i-th row and j-th column in the encoding matrix, then the i-th row vector in the encoding matrix can be obtained, and then the row vectors extracted for each discrete feature are concatenated or averaged, so as to obtain the first embedding vector of the discrete feature relative to the encoding matrix. Of course, in the embodiment of the present disclosure, if a discrete feature corresponds to the i-th row and j-th column in the encoding matrix, it can also be set to obtain the j-th column vector in the encoding matrix to form the first embedding vector, which can be pre-set according to the needs, and the embodiment of the present disclosure is not limited to this.

[0083] Furthermore, according to the first embedding vector of each offline feature, and the continuous features in the first output vector and the second feature sequence, the second embedding vector of the first output vector and the second feature sequence is obtained. Specifically, each continuous feature can be connected to the first embedding vector corresponding to the discrete feature to obtain the second embedding vector of the first output vector and the second feature sequence. Of course, according to the first embedding vector of each offline feature, and the continuous features in the first output vector and the second feature sequence, the specific method of obtaining the second embedding vector can also be preset according to the needs, and this is not limited in the embodiment of the present disclosure.

[0084] Alternatively, in the disclosed embodiment, only the coordinates of each discrete feature in the second feature sequence relative to the encoding matrix may be used to obtain the first embedding vector of the discrete feature relative to the encoding matrix; and then the first output vector and the second embedding vector of the second feature sequence may be obtained based on the first embedding vector of each offline feature, the continuous features in the second feature sequence, and the first output vector. In this case, only the first embedding vector of the discrete feature relative to the encoding matrix may be obtained for the coordinates of each discrete feature in the second feature sequence relative to the encoding matrix, without operating the first output vector.

[0085] Step 240: Obtain a recommendation score for each of the candidate merchants through the deep neural network model according to the output vector and the feature information of the candidate merchants.

[0086] Step 250: based on the recommendation scores of the candidate merchants, obtain the target recommended merchants of the target user and push them to the target user.

[0087] In the disclosed embodiment, by obtaining the click sequence information of the target user; according to the click sequence information, obtaining the recommendation score of each candidate merchant through a preset recommendation model; based on the recommendation score of the candidate merchant, obtaining the target recommended merchant of the target user and pushing it to the target user; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model. The beneficial effect of improving the accuracy of recommended merchants is achieved.

[0088] Moreover, in the embodiment of the present disclosure, it is also possible to obtain feature information of the target user clicking on the merchant within a preset time period, and to construct the click sequence information of the target user based on the feature information. Also, the click sequence information is input into the double-layer recurrent neural network model, and the output vector of the double-layer recurrent neural network model is obtained; according to the output vector and the feature information of the candidate merchant, the recommendation score of each candidate merchant is obtained through the deep neural network model. The click sequence information includes a first feature sequence and a second feature sequence; the first feature sequence includes dish features, and the second feature sequence includes at least one of merchant features, user features, and user-merchant cross features. Moreover, the first feature sequence is input into the first-level recurrent neural network model, and the first output vector of the first-level recurrent neural network model is obtained; the first output vector and the second feature sequence are input into the second-level recurrent neural network model, and the second output vector of the second-level recurrent neural network model is obtained as the output vector of the double-layer recurrent neural network model. Based on the preset encoding matrix, and the coordinates of each discrete feature in the first output vector and the second feature sequence relative to the encoding matrix, the first embedding vector of the discrete feature relative to the encoding matrix is ​​obtained; according to the first embedding vector of each offline feature, and the continuous features in the first output vector and the second feature sequence, the second embedding vector of the first output vector and the second feature sequence is obtained; the second embedding vector is input into the second-level recurrent neural network model, and the second output vector of the second-level recurrent neural network model is obtained as the output vector of the double-layer recurrent neural network model. This can further improve the accuracy of the recommendation results.

[0089] In addition, in an embodiment of the present disclosure, a recommendation model obtained by combining the two-layer recurrent neural network model and the deep neural network model is jointly trained based on preset training sample data; wherein, in each training process, the input data of the two-layer recurrent neural network model includes the sample sequence information in the training sample data, and the input data of the deep neural network model includes the output data of the two-layer recurrent neural network model, the feature sequence of the sample merchant corresponding to the sample sequence information, and the label value of the sample merchant. Both the first recurrent neural network model and the second recurrent neural network model include dynamic long short-term memory network models. The accuracy of the recommendation model can also be further improved.

[0090] For the method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the embodiments of the present disclosure are not limited by the order of the actions described, because according to the embodiments of the present disclosure, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present disclosure.

[0091] Embodiment 3

[0092] A merchant recommendation device provided by an embodiment of the present disclosure is introduced in detail.

[0093] Reference Figure 4 , showing a schematic structural diagram of a merchant recommendation device in an embodiment of the present disclosure.

[0094] The click sequence information acquisition module 310 is used to acquire the click sequence information of the target user.

[0095] The recommendation score acquisition module 320 is used to obtain the recommendation score of each candidate merchant according to the click sequence information through a preset recommendation model; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, and the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model.

[0096] The target recommended merchant acquisition module 330 is used to acquire the target recommended merchant of the target user based on the recommendation score of the candidate merchant and push it to the target user.

[0097] In the disclosed embodiment, by obtaining the click sequence information of the target user; according to the click sequence information, obtaining the recommendation score of each candidate merchant through a preset recommendation model; based on the recommendation score of the candidate merchant, obtaining the target recommended merchant of the target user and pushing it to the target user; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model. The beneficial effect of improving the accuracy of recommended merchants is achieved.

[0098] Embodiment 4

[0099] A merchant recommendation device provided by an embodiment of the present disclosure is introduced in detail.

[0100] Reference Figure 5 , showing a schematic structural diagram of a merchant recommendation device in an embodiment of the present disclosure.

[0101] The click sequence information acquisition module 410 is used to acquire the click sequence information of the target user.

[0102] Optionally, in the embodiment of the present disclosure, the click sequence information acquisition module 410 may further include:

[0103] The click sequence information acquisition submodule 411 is used to acquire characteristic information of the target user clicking on the merchant within a preset time period, and construct the click sequence information of the target user based on the characteristic information.

[0104] The recommendation model training module 420 is used to jointly train the recommendation model obtained by combining the double-layer recurrent neural network model and the deep neural network model according to preset training sample data; wherein, in each training process, the input data of the double-layer recurrent neural network model includes the sample sequence information in the training sample data, and the input data of the deep neural network model includes the output data of the double-layer recurrent neural network model, the feature sequence of the sample merchant corresponding to the sample sequence information, and the label value of the sample merchant.

[0105] The recommendation score acquisition module 430 is used to obtain the recommendation score of each candidate merchant according to the click sequence information through a preset recommendation model; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, and the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model.

[0106] Optionally, in an embodiment of the present disclosure, the click sequence information includes a first feature sequence and a second feature sequence; the first feature sequence includes dish features, and the second feature sequence includes at least one of merchant features, user features, and user-merchant cross features.

[0107] Optionally, in the embodiment of the present disclosure, the recommendation score acquisition module 430 may further include:

[0108] A first input submodule 431 is used to input the click sequence information into the double-layer recurrent neural network model and obtain an output vector of the double-layer recurrent neural network model;

[0109] Optionally, in the embodiment of the present disclosure, the first input submodule 431 may further include:

[0110] A first feature sequence input unit, used for inputting the first feature sequence into the first-layer recurrent neural network model, and obtaining a first output vector of the first-layer recurrent neural network model;

[0111] The second feature sequence input unit is used to input the first output vector and the second feature sequence into the second-layer recurrent neural network model, and obtain the second output vector of the second-layer recurrent neural network model as the output vector of the double-layer recurrent neural network model.

[0112] Optionally, in an embodiment of the present disclosure, the first-layer recurrent neural network model and the second-layer recurrent neural network model both include dynamic long short-term memory network models.

[0113] Optionally, in the embodiment of the present disclosure, the second feature sequence input unit may further include:

[0114] A first embedding vector acquisition subunit, configured to acquire a first embedding vector of the discrete feature relative to the encoding matrix based on a preset encoding matrix and the coordinates of each discrete feature in the first output vector and the second feature sequence relative to the encoding matrix;

[0115] A second embedding vector acquisition subunit, configured to acquire a second embedding vector of the first output vector and the second feature sequence according to the first embedding vector of each offline feature and continuous features in the first output vector and the second feature sequence;

[0116] The output vector acquisition subunit is used to input the second embedding vector into the second-layer recurrent neural network model and obtain the second output vector of the second-layer recurrent neural network model as the output vector of the two-layer recurrent neural network model.

[0117] The recommendation score acquisition submodule 432 is used to acquire the recommendation score of each candidate merchant through the deep neural network model according to the output vector and the feature information of the candidate merchants.

[0118] The target recommended merchant acquisition module 440 is used to acquire the target recommended merchant of the target user based on the recommendation score of the candidate merchant and push it to the target user.

[0119] In the disclosed embodiment, by obtaining the click sequence information of the target user; according to the click sequence information, obtaining the recommendation score of each candidate merchant through a preset recommendation model; based on the recommendation score of the candidate merchant, obtaining the target recommended merchant of the target user and pushing it to the target user; wherein the recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, the double-layer recurrent neural network model includes a first-level recurrent neural network model and a second-level recurrent neural network model, and the input data of the second-level recurrent neural network model includes the output data of the first-level recurrent neural network model. The beneficial effect of improving the accuracy of recommended merchants is achieved.

[0120] Moreover, in the embodiment of the present disclosure, it is also possible to obtain feature information of the target user clicking on the merchant within a preset time period, and to construct the click sequence information of the target user based on the feature information. Also, the click sequence information is input into the double-layer recurrent neural network model, and the output vector of the double-layer recurrent neural network model is obtained; according to the output vector and the feature information of the candidate merchant, the recommendation score of each candidate merchant is obtained through the deep neural network model. The click sequence information includes a first feature sequence and a second feature sequence; the first feature sequence includes dish features, and the second feature sequence includes at least one of merchant features, user features, and user-merchant cross features. Moreover, the first feature sequence is input into the first-level recurrent neural network model, and the first output vector of the first-level recurrent neural network model is obtained; the first output vector and the second feature sequence are input into the second-level recurrent neural network model, and the second output vector of the second-level recurrent neural network model is obtained as the output vector of the double-layer recurrent neural network model. Based on the preset encoding matrix, and the coordinates of each discrete feature in the first output vector and the second feature sequence relative to the encoding matrix, the first embedding vector of the discrete feature relative to the encoding matrix is ​​obtained; according to the first embedding vector of each offline feature, and the continuous features in the first output vector and the second feature sequence, the second embedding vector of the first output vector and the second feature sequence is obtained; the second embedding vector is input into the second-level recurrent neural network model, and the second output vector of the second-level recurrent neural network model is obtained as the output vector of the double-layer recurrent neural network model. This can further improve the accuracy of the recommendation results.

[0121] In addition, in the disclosed embodiment, the recommendation model obtained by combining the two-layer recurrent neural network model and the deep neural network model is jointly trained based on the preset training sample data; wherein, in each training process, the input data of the two-layer recurrent neural network model includes the sample sequence information in the training sample data, and the input data of the deep neural network model includes the output data of the two-layer recurrent neural network model, the feature sequence of the sample merchant corresponding to the sample sequence information, and the label value of the sample merchant. The two-layer recurrent neural network model includes a dynamic long short-term memory network model. The accuracy of the recommendation model can also be further improved.

[0122] The present disclosure also discloses an electronic device, including:

[0123] A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the aforementioned merchant recommendation methods is implemented.

[0124] A readable storage medium is also disclosed in an embodiment of the present disclosure. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute any of the aforementioned merchant recommendation methods.

[0125] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0126] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems may also be used together with the teachings based thereon. According to the above description, it is apparent that the structure required for constructing such systems is. In addition, the present disclosure is not directed to any particular programming language either. It should be understood that various programming languages ​​may be utilized to implement the disclosure described herein, and the description of the specific languages ​​above is intended to disclose the best mode of implementation of the present disclosure.

[0127] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present disclosure can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0128] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present disclosure, the various features of the present disclosure are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the following intention: the claimed disclosure requires more features than the features explicitly recited in each claim. More specifically, as reflected in the claims below, the inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present disclosure.

[0129] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0130] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of this disclosure and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0131] The various component embodiments of the present disclosure may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the merchant recommendation device according to the embodiment of the present disclosure. The present disclosure may also be implemented as a device or apparatus program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present disclosure may be stored on a computer-readable medium, or may be in the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0132] It should be noted that the above embodiments illustrate the present disclosure rather than limit the present disclosure, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of multiple such elements. The present disclosure may be implemented by means of hardware including several different elements and by means of appropriately programmed computers. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names.

Claims

1. A merchant recommendation method, characterized in that: include: Obtain click sequence information of target users; According to the click sequence information, a recommendation score of each candidate merchant is obtained through a preset recommendation model, including: Inputting the click sequence information into a double-layer recurrent neural network model, and obtaining an output vector of the double-layer recurrent neural network model; According to the output vector and the characteristic information of the candidate merchants, obtaining a recommendation score for each of the candidate merchants through a deep neural network model; Based on the recommendation scores of the candidate merchants, the target recommended merchants of the target user are obtained and pushed to the target user; The recommendation model is a combined model obtained by jointly training a double-layer recurrent neural network model and a deep neural network model, wherein the double-layer recurrent neural network model includes a first-layer recurrent neural network model and a second-layer recurrent neural network model, and the input data of the second-layer recurrent neural network model includes the output data of the first-layer recurrent neural network model; The click sequence information includes a first feature sequence and a second feature sequence; the first feature sequence includes a dish feature, and the second feature sequence includes at least one of a merchant feature, a user feature, and a user-merchant cross feature; The step of inputting the click sequence information into the double-layer recurrent neural network model and obtaining the output vector of the double-layer recurrent neural network model comprises: Inputting the first feature sequence into the first-layer recurrent neural network model, and obtaining a first output vector of the first-layer recurrent neural network model; Inputting the first output vector and the second feature sequence into the second-layer recurrent neural network model, and obtaining the second output vector of the second-layer recurrent neural network model as the output vector of the double-layer recurrent neural network model; The step of inputting the first output vector and the second feature sequence into the second-layer recurrent neural network model, and obtaining the second output vector of the second-layer recurrent neural network model as the output vector of the double-layer recurrent neural network model includes: Based on a preset encoding matrix, and the coordinates of each discrete feature in the first output vector and the second feature sequence relative to the encoding matrix, obtaining a first embedding vector of the discrete feature relative to the encoding matrix; According to the first embedding vector of each offline feature and the continuous features in the first output vector and the second feature sequence, obtain the second embedding vector of the first output vector and the second feature sequence; input the second embedding vector into the second-layer recurrent neural network model, and obtain the second output vector of the second-layer recurrent neural network model as the output vector of the double-layer recurrent neural network model.

2. The method according to claim 1, characterized in that The first-layer recurrent neural network model and the second-layer recurrent neural network model both include dynamic long short-term memory network models.

3. The method according to claim 1, characterized in that The step of obtaining the click sequence information of the target user includes: Acquire characteristic information of the target user clicking on a merchant within a preset time period, and construct click sequence information of the target user based on the characteristic information.

4. The method according to claim 1, characterized in that: Before the step of obtaining the recommendation score of each candidate merchant through a preset recommendation model according to the click sequence information, the method further includes: According to the preset training sample data, jointly train the recommendation model obtained by combining the double-layer recurrent neural network model and the deep neural network model; In which, during each training process, the input data of the two-layer recurrent neural network model includes the sample sequence information in the training sample data, and the input data of the deep neural network model includes the output data of the two-layer recurrent neural network model, the feature sequence of the sample merchant corresponding to the sample sequence information, and the label value corresponding to the sample merchant.

5. A merchant recommendation device, characterized in that: include: A click sequence information acquisition module is used to obtain the click sequence information of the target user; The click sequence information includes a first feature sequence and a second feature sequence; The first feature sequence includes dish features, and the second feature sequence includes at least one of merchant features, user features, and user-merchant cross features; The step of inputting the click sequence information into a double-layer recurrent neural network model and obtaining an output vector of the double-layer recurrent neural network model comprises: Inputting the first feature sequence into a first-layer recurrent neural network model, and obtaining a first output vector of the first-layer recurrent neural network model; Inputting the first output vector and the second feature sequence into a second-layer recurrent neural network model, and obtaining a second output vector of the second-layer recurrent neural network model as an output vector of the double-layer recurrent neural network model; The step of inputting the first output vector and the second feature sequence into the second-layer recurrent neural network model, and obtaining the second output vector of the second-layer recurrent neural network model as the output vector of the double-layer recurrent neural network model includes: Based on a preset encoding matrix, and the coordinates of each discrete feature in the first output vector and the second feature sequence relative to the encoding matrix, obtaining a first embedding vector of the discrete feature relative to the encoding matrix; According to the first embedding vector of each offline feature and the continuous features in the first output vector and the second feature sequence, obtain the second embedding vector of the first output vector and the second feature sequence; input the second embedding vector into the second-layer recurrent neural network model, and obtain the second output vector of the second-layer recurrent neural network model as the output vector of the two-layer recurrent neural network model; The recommendation score acquisition module is used to obtain the recommendation score of each candidate merchant according to the click sequence information through a preset recommendation model, including: Inputting the click sequence information into the double-layer recurrent neural network model, and obtaining an output vector of the double-layer recurrent neural network model; According to the output vector and the characteristic information of the candidate merchants, obtaining a recommendation score for each of the candidate merchants through a deep neural network model; A target recommended merchant acquisition module is used to acquire the target recommended merchant of the target user based on the recommendation score of the candidate merchant and push it to the target user; Among them, the recommendation model is a combined model obtained by jointly training a two-layer recurrent neural network model and a deep neural network model, the two-layer recurrent neural network model includes a first-layer recurrent neural network model and a second-layer recurrent neural network model, and the input data of the second-layer recurrent neural network model includes the output data of the first-layer recurrent neural network model.

6. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the merchant recommendation method according to any one of claims 1 to 4 when executing the computer program.

7. A readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the merchant recommendation method as described in any one of claims 1 to 4.

Images