Information sending method, device, equipment and readable medium

Abstract

The embodiments of this specification disclose an information sending method, device, equipment and readable medium. The scheme may include: obtaining first commodity sales log information of a target merchant; obtaining second commodity sales log information of merchants around the target merchant; based on the first commodity sales log information and the second commodity sales log information, using a multi-task model, determining first sales forecast information corresponding to the target merchant; sending the first sales forecast information to the merchant terminal of the target merchant.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an information sending method, device, equipment and computer-readable medium. Background Art

[0002] For offline merchants, due to the lack of understanding of the market sales situation of various commodities, merchants are unable to adjust their commodity purchasing strategies in a timely manner. As a result, some commodities on the market are easily out of stock, causing inconvenience to users' daily shopping. On the other hand, it also causes losses to merchants' operations.

[0003] At present, some commodity agents will provide merchants with purchase suggestions, but these purchase suggestions are often highly subjective and cannot effectively overcome the problem of commodity shortages in the market and business losses of merchants. In addition, the existing hot product lists in the hot product recommendation scenario are all based on the statistical records of historical data and are based on a global statistical indicator (for example, sales volume, price, etc.), without considering the impact of the merchant's surrounding environment. For example, the commodity sales of surrounding merchants are not considered. Therefore, for a specific merchant, such a hot product list does not have a good recommendation and practicality for the merchant to adjust its purchase strategy. Summary of the invention

[0004] The embodiments of this specification provide an information sending method, device, equipment and computer-readable medium to solve the problem in the prior art that it is impossible to provide merchants with effective information recommendations to guide merchants in purchasing.

[0005] To solve the above technical problems, the embodiments of this specification are implemented as follows:

[0006] An information sending method provided in an embodiment of the present specification includes: obtaining first commodity sales log information of a target merchant; obtaining second commodity sales log information of merchants surrounding the target merchant; based on the first commodity sales log information and the second commodity sales log information, using a multi-task model, determining first sales forecast information corresponding to the target merchant; and sending the first sales forecast information to a merchant terminal of the target merchant.

[0007] An information sending device provided in an embodiment of the present specification includes: a first acquisition module, used to obtain first commodity sales log information of a target merchant; a second acquisition module, used to obtain second commodity sales log information of merchants surrounding the target merchant; a sales forecasting module, used to determine first sales forecast information corresponding to the target merchant based on the first commodity sales log information and the second commodity sales log information using a multi-task model; an information sending module, used to send the first sales forecast information to a merchant terminal of the target merchant.

[0008] An information sending device provided in an embodiment of this specification includes:

[0009] at least one processor; and,

[0010] a memory communicatively connected to the at least one processor; wherein,

[0011] The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0012] Obtain first commodity sales log information of a target merchant; obtain second commodity sales log information of merchants surrounding the target merchant; determine first sales forecast information corresponding to the target merchant using a multi-task model based on the first commodity sales log information and the second commodity sales log information; and send the first sales forecast information to a merchant terminal of the target merchant.

[0013] An embodiment of the present specification provides a computer-readable medium on which computer-readable instructions are stored. The computer-readable instructions can be executed by a processor to implement an information sending method.

[0014] One embodiment of the present specification can at least achieve the following beneficial effects: by using a multi-task model, based on the first commodity sales log information of the target merchant and the second commodity sales log information of the surrounding merchants of the target merchant, the sales forecast information corresponding to the target merchant is obtained, and sent to the merchant terminal of the target merchant, thereby being able to predict the commodity sales of the target merchant while considering the surrounding environment of the target merchant, with high accuracy of the prediction, being able to make personalized hot product recommendations for the target merchant, and at least being able to provide a basis for the merchant to adjust its purchase, thereby achieving the purpose of guiding the merchant to gain profit, and objectively providing convenience for users who purchase commodities. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0016] Figure 1 A schematic diagram of an application scenario of a method for sending information to a merchant provided in an embodiment of this specification;

[0017] Figure 2A flowchart of a method for sending information provided in an embodiment of this specification;

[0018] Figure 3 A schematic diagram of the structure of a multi-task model provided in an embodiment of this specification;

[0019] Figure 4 A flowchart of a method for generating recommendation information provided in an embodiment of this specification;

[0020] Figure 5 A flowchart of another information sending solution provided in an embodiment of this specification;

[0021] Figure 6 A schematic diagram of a specific implementation scenario of the IODE module provided in the embodiments of this specification;

[0022] Figure 7 A flowchart of a method for determining a timing for sending information to a merchant provided in an embodiment of this specification;

[0023] Figure 8 A schematic diagram of a method for obtaining sales characteristic information at the current moment provided in an embodiment of this specification;

[0024] Fig. 9 A flowchart of a method for sending and receiving hot product recommendation information in an actual application scenario provided by an embodiment of this specification;

[0025] Fig.10 The embodiments of this specification provide corresponding to Figure 2 A structural schematic diagram of an information sending device;

[0026] Fig.11 The embodiments of this specification provide corresponding to Figure 2 A structural diagram of an information sending device. DETAILED DESCRIPTION

[0027] In actual commodity sales scenarios, merchants often lack an objective understanding of the sales situation of surrounding commodities, which leads to their inability to adjust their purchases in a timely manner, resulting in potential losses to their operations. In addition, some commodity agents’ purchase suggestions for merchants are subjective and are not based on considerations of the merchants’ interests, so they have no actual value in guiding purchases.

[0028] In order to solve this problem, the embodiments of this specification take into account the uneven distribution of merchants in offline stores and the sparse commodity sales data of each merchant, and propose a new hot product recommendation scheme. The scheme of the embodiments of this specification can at least achieve the purpose of guiding merchants to increase profits by providing merchants with relevant hot product recommendation information, and objectively provide convenience for users who purchase commodities.

[0029] In order to make the purpose, technical solutions and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of one or more embodiments of this specification.

[0030] It should be understood that although the terms first, second, third, etc. may be used in this application document to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other.

[0031] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0032] Figure 1 This is a schematic diagram of an application scenario of a method for sending information to a merchant in an embodiment of this specification.

[0033] like Figure 1 As shown, the server may send the recommendation information determined based on the solution of the embodiment of this specification to the merchant terminal, so that the merchant can learn the recommendation information through the merchant terminal. In an optional embodiment, the server may actively send the recommendation information to the merchant terminal according to a preset information sending rule. In another optional embodiment, the merchant may request the recommendation information from the server through the merchant terminal, so that the server sends the determined recommendation information to the merchant terminal in response to the request of the merchant terminal.

[0034] In an embodiment of the present specification, the server may include a commodity recommendation engine (CRE) module for determining recommendation information. Specifically, the commodity recommendation engine module may predict the sales forecast information corresponding to the target merchant based on the commodity sales log information of the target merchant and the commodity sales log information of the surrounding merchants of the target merchant, and send the sales forecast information to the merchant terminal of the target merchant as recommendation information.

[0035] Traditional hot product recommendation lists in the prior art are usually recommended based on statistical records of historical data, and do not predict the sales of goods of designated merchants under the consideration of the influence of the surrounding environment. Therefore, they do not have the recommendation characteristics for guiding purchases. Moreover, traditional solutions are all hot product recommendation lists obtained based on a global statistical indicator (for example, sales volume, price, etc.), and do not make personalized recommendations for each merchant. Therefore, they do not have the conditions to influence the purchase and sale of a merchant.

[0036] In the embodiments of the present specification, based on historical statistical data, the sales situation of surrounding merchants of each merchant is comprehensively considered, that is, the overall consumption situation of consumers in the surrounding environment of each merchant is taken into consideration, and hot product forecast recommendation information suitable for each merchant is provided, which can provide more accurate reference information for merchants' purchase and sale.

[0037] Optionally, in the embodiments of the present specification, the server may further include an Intelligent Orchestration Delivery Engine (IODE) module, which is used to determine the timing of sending the recommendation information to the merchant terminal.

[0038] although, Figure 1 A server device is shown, but it can be understood that in actual application, the server here can represent any device, equipment, platform, device cluster, etc. with computing and processing capabilities.

[0039] although, Figure 1 The merchant terminal is shown as a smart box, but in actual application, the merchant terminal used to receive recommendation information from the server to provide to the merchant, and to obtain query information from the merchant and send it to the server can also be in other forms, for example, it can be a computer, a laptop computer, a mobile terminal, etc., but not limited to this.

[0040] like Figure 1 As described in, when the merchant terminal is a smart box, the smart box may include an information output module and an information receiving module. Specifically, the information receiving module may include an information display screen, an information broadcaster, etc., and the information receiving module may include a code scanning unit, a key unit, etc. In addition, the functions of the smart box may not be limited to the above functions, for example, it may also have a clock function.

[0041] In addition, despite the Figure 1 The merchant terminal is shown as a hardware form, but it does not limit the scope of the merchant terminal. The merchant terminal in the embodiments of this specification may also include a software form, for example, it may be an application installed in a hardware device, more specifically, it may be an independent application (APP) or a small program, etc.

[0042] Next, a method for sending information provided in an embodiment of the specification will be specifically described with reference to the accompanying drawings:

[0043] Figure 2 This is a flow chart of a method for sending information provided by an embodiment of this specification. From a program perspective, the execution subject of the process can be a program installed on an application server. More specifically, it can be a CRE in the server.

[0044] like Figure 2 As shown, the process may include the following steps:

[0045] Step 202: Obtain the first commodity sales log information of the target merchant.

[0046] Step 204: Acquire the second commodity sales log information of the surrounding merchants of the target merchant.

[0047] Step 206: Based on the first commodity sales log information and the second commodity sales log information, use a multi-task model to determine first sales forecast information corresponding to the target merchant.

[0048] Step 208: Send the first sales forecast information to the merchant terminal of the target merchant.

[0049] Each step is described in detail below.

[0050] In the embodiment of this specification, before the server sends information to the target merchant, it is necessary to calculate the sales forecast information of the target merchant. Considering that the product information and product sales information of offline stores are relatively sparse, the product sales information of the target merchant and the product sales information of surrounding merchants can be comprehensively used to calculate the sales forecast information of the target merchant.

[0051] In step 202, the server may obtain commodity sales log information corresponding to the target merchant, referred to as first commodity sales log information.

[0052] Specifically, the commodity sales log information may include order information, and the first commodity sales log information may include order information corresponding to the target merchant. The order information may include merchant information (e.g., merchant identification information, merchant location information, merchant scale information, etc.), commodity information (e.g., commodity identification information, commodity name information, commodity specification information, commodity price information, commodity quantity information, etc.), order time information, order amount information, order payment method information, etc., but is not limited thereto.

[0053] Optionally, the first commodity sales log information may be obtained from a merchant terminal of the target merchant; or, may be obtained from a log system, and the log system may store the first commodity sales log information uploaded by the merchant terminal of the target merchant.

[0054] In step 204, the server may further obtain commodity sales log information corresponding to the surrounding merchants of the target merchant, which is referred to as second commodity sales log information.

[0055] Specifically, the second commodity sales log information may include order information corresponding to at least one surrounding merchant of the target merchant. The order information may include merchant information (e.g., merchant identification information, merchant location information, merchant scale information, etc.), commodity information (e.g., commodity identification information, commodity name information, commodity specification information, commodity price information, commodity quantity information, etc.), order time information, order amount information, order payment method information, etc., but is not limited thereto.

[0056] In the embodiments of the present specification, the surrounding merchants of the target merchant may be determined according to the degree of influence of the surrounding merchants on the commodity sales of the target merchant. For example, in view of the fact that the commodity sales between merchants that are geographically adjacent will affect each other, therefore, in some embodiments, the surrounding merchants of the target merchant may include merchants that are geographically adjacent to the target merchant. Optionally, the surrounding merchants of the target merchant may include merchants whose straight-line distance to the target merchant is less than a first preset distance threshold. Alternatively, the surrounding merchants of the target merchant may include merchants whose route distance to the target merchant is less than a second preset distance threshold. Furthermore, optionally, the surrounding merchants of the target merchant may include merchants that are located in the same region as the target merchant and whose straight-line distance and / or route distance is less than a preset threshold. The specific determination method of the surrounding merchants of the target merchant is not limited to the aforementioned examples and can be determined according to actual conditions.

[0057] In actual application, before obtaining the second commodity sales log information of the surrounding merchants of the target merchant, it may also include: obtaining the location information of the target merchant; and determining the surrounding merchants of the target merchant based on the location information of the target merchant. Specifically, when the first commodity sales log information contains the merchant location information, the location information of the target merchant may be directly obtained from the first commodity sales log information of the target merchant; or, when the first commodity sales log information does not contain the merchant location information, the location information of the target merchant may be indirectly obtained by querying the database based on the information in the first commodity sales log information of the target merchant (for example, the merchant identification information).

[0058] In step 206, the server can obtain the first sales forecast information corresponding to the target merchant based on the obtained first commodity sales log information of the target merchant and the second commodity sales log information of the surrounding merchants of the target merchant by using the multi-task model. At the same time, the server can also obtain the second sales forecast information corresponding to the surrounding merchants of the target merchant.

[0059] Specifically, the product recommendation engine (i.e., CRE module) in the server can perform product sales forecasting. In the embodiment of this specification, the main framework of CRE can be a multi-task learning framework, which is mainly divided into two tasks, namely, the target merchant sales forecasting task and the surrounding merchant sales forecasting task. The use of the multi-task learning framework is mainly due to the fact that offline store product information and product sales information are relatively sparse. Using the surrounding sales task as an auxiliary task can supplement the main task store sales forecasting task, which can improve the accuracy of the forecast results.

[0060] In one or more embodiments, step 206 may specifically include: step 206-1, determining a commodity sales feature vector based on the first commodity sales log information and the second commodity sales log information; step 206-2, determining the first sales forecast information corresponding to the target merchant based on the commodity sales feature vector and the gating weight corresponding to each merchant sales forecast task. It is understandable that while determining the first sales forecast information corresponding to the target merchant, the second sales forecast information corresponding to the surrounding merchants may also be determined.

[0061] Among them, in step 206-1, the merchant order information can be first parsed from the system log, and then the features required by the CRE model can be constructed. Since the order information covers information such as a person buying a certain product from a certain merchant in a certain place, the required feature information can be obtained by aggregating data at the upper layer of the data. For example, aggregation can be performed on the product dimension to obtain the number of merchants selling the product, the number of users who purchased the product, the age group ratio of users who purchased the product, etc.

[0062] In step 206-2, the commodity sales feature vector in step 206-1 can be input into the multi-task model to predict the commodity sales. The main learning tasks of the multi-task model can be divided into two, one is the sales value of the target merchant, and the other is the sales value of the surrounding merchants. These two tasks can share input data, and the output data of the multi-task model includes the sales forecast results of the target merchant and the sales forecast results of the surrounding merchants.

[0063] In an optional embodiment, the multi-task model can adopt the MMoE algorithm framework (Modeling TaskRelationships in Multi-task Learning with Multi-gate Mixture-of-Experts).

[0064] Specifically, the multi-task model of the CRE module may include an input layer, an embedding layer, a connection layer, an MMoE layer, and an output layer. Figure 3 , which shows a schematic diagram of the structure of the multi-task model used in the embodiments of this specification.

[0065] At the input layer, an original feature vector for input into a commodity sales prediction model can be constructed based on the first commodity sales log information and the second commodity sales log information; and according to the current business scenario, the commodity sales original feature vector can be divided into dense feature vectors, sparse feature vectors and sequence feature vectors.

[0066] In the embedding layer, different embedding strategies can be used to process each feature according to different feature types. The dense feature vector, the sparse feature vector and the sequence feature vector can be mapped according to their corresponding embedding methods to obtain corresponding commodity sales sub-feature vectors. The embedding methods corresponding to the dense feature vector, the sparse feature vector and the sequence feature vector can be predetermined. For example, the lookup mechanism can be used to embed sparse features, and the transformer mechanism can be used to embed sequence features.

[0067] In the connection layer, a commodity sales feature vector corresponding to the commodity sales original feature vector can be obtained based on the commodity sales sub-feature vectors corresponding to the dense feature vector, the sparse feature vector, and the sequence feature vector. Specifically, the result of processing the sparse and dense features using the DCN (Deep & Cross Network) model, the result after the lookup operation, and the transformer-based embedding result can be connected.

[0068] At the MMoE layer, the MMoE framework is mainly divided into a gated network (also known as a gated model) and an expert network (also known as an expert model), wherein each gate in the gated network is mainly used to learn the weight coefficients of each expert in the expert network. In an embodiment of the present specification, optionally, on the basis of the basic MMoE algorithm framework, an attention mechanism (i.e., Attention mechanism) can be integrated to perform weighted learning of gate structures (i.e., gate weights) for different tasks, and ultimately achieve simultaneous optimization of auxiliary tasks and main tasks. By using the attention mechanism for learning, it is possible to correct the deviation in the sales forecast for merchants and improve the accuracy of forecasts.

[0069] Specifically, the following steps can be performed at the MMoE layer: input the commodity sales feature vector into multiple expert models to obtain the latent feature vectors corresponding to each of the multiple expert models; input the commodity sales feature vector into the gating model corresponding to each merchant sales prediction task to obtain the attention gating weight of each merchant sales prediction task; for the target merchant sales prediction task in each merchant sales prediction task, weighted summation is performed based on the attention gating weight of the target merchant sales prediction task and the latent feature vectors corresponding to each of the multiple expert models to obtain the comprehensive latent feature vector of the target merchant sales prediction task.

[0070] In practical applications, since the weight selection of the model is different for different tasks, a gating model g can be equipped for each task. k (x), where k represents task k. The input of the gating model is the product sales feature vector, and the output of the gating model is the weights of all expert models.

[0071] The output of the gating model corresponding to a task is used to indicate the probability of selecting the output results of different expert models for the task. Therefore, the outputs of multiple expert models can be weighted and summed to obtain the comprehensive implicit output corresponding to the task. It can be calculated by the following formula:

[0072]

[0073] Where k represents task k; i ranges from 1 to n, and n represents n expert models; f i (x) represents the output of expert model i, g i k (x) represents the weight of expert model i for task k.

[0074] In an optional embodiment, the weight parameters of the experts learned by the attention mechanism are used to correct the deviation of the predicted sales situation of the store. The commodity sales feature vector is input into the gating model corresponding to each merchant sales prediction task to obtain the attention gating weight of each merchant sales prediction task, which may specifically include: calculating the original gating weight of each merchant sales prediction task; then using the attention function to calculate the weighted attention value vector; and then based on the original gating weight and the weighted attention value vector, obtaining the attention gating weight of each merchant sales prediction task.

[0075] Specifically, taking the main task as an example, the attention gating weight of the target merchant's sales prediction task can be calculated by the following formula:

[0076]

[0077] Among them, the gate master on the right side of the equation is the original gating weight corresponding to the original MMoE network, that is, the weight parameter of the expert obtained by learning using the fully connected network; It represents the correction of the deviation of the merchant's predicted sales situation based on the attention mechanism; the gate master on the left side of the equation is the attention gating weight proposed in the embodiment of this specification, that is, it represents the weight parameter about the expert obtained by learning using the attention mechanism.

[0078] Among them, Q represents the query vector; K represents the key vector; V represents the value vector; represents the scale factor, d k Specifically, it may be the dimension of the key vector K.

[0079] In an optional example of the present specification, the query vector Q, the key vector K, and the value vector V may be represented by the following formula:

[0080] Q=w1g'(r), k=w2g'(r), K=w3g'(r).

[0081] Among them, w1, w2 and w3 represent the weight parameters that can be learned; g'(r) represents the result of connecting the gate weight parameters of the main task and the auxiliary task, that is, it can be g'(r) = [gate 主 , gate 辅 ]. Among them, gate 主 represents the gating weight parameter corresponding to the main task, that is, the gating weight parameter corresponding to the sales volume prediction task of the target merchant; 辅 Represents the gating weight parameter corresponding to the auxiliary task, that is, the gating weight parameter corresponding to the sales volume prediction task of the surrounding merchants of the target merchant.

[0082] In the output layer, the comprehensive latent feature vector of the target merchant sales prediction task can be input into the tower model corresponding to the target merchant to obtain the first sales prediction information corresponding to the target merchant.

[0083] Specifically, after obtaining the comprehensive hidden output corresponding to each task, the comprehensive hidden output can be provided to the corresponding tower model to obtain the final output of each task. More specifically, it can be calculated by the following formula:

[0084] y k =h k (f k (x)).

[0085] Where k represents task k; h k Denotes the tower model corresponding to task k. Tower model h k The input is the output information of the MMoE layer, the tower model h k The output is the prediction information of the multi-task model for each task. For example, in the embodiment of this specification, the sales forecast information corresponding to the target merchant and the sales forecast information corresponding to the surrounding merchants can be output.

[0086] In step 208, the prediction result obtained by the multi-task model of the CRE module can be sent to the merchant terminal of the merchant as recommendation information. That is, the first sales forecast information corresponding to the target merchant can be sent to the merchant terminal of the target merchant. Thus, the merchant terminal can display, broadcast, or display and broadcast the first sales forecast information to the target merchant.

[0087] In practical applications, the information displayed or broadcasted by the merchant terminal may specifically be sales forecast list information, or may be called hot product forecast list information.

[0088] Figure 2 The method in the invention obtains the sales forecast information corresponding to the target merchant based on the first commodity sales log information of the target merchant and the second commodity sales log information of the surrounding merchants of the target merchant by using a multi-task model, and sends the sales forecast information to the merchant terminal of the target merchant. Thus, the commodity sales of the target merchant can be predicted while considering the surrounding environment of the target merchant. The prediction has high accuracy, and personalized hot product recommendations can be made for the target merchant. At least a basis can be provided for the merchant to adjust its purchase, thereby achieving the purpose of guiding the merchant to gain profit, and objectively providing convenience for users who purchase commodities.

[0089] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps can be adjusted according to actual needs, or some steps can be omitted.

[0090] based on Figure 2The method, the embodiments of this specification also provide some specific implementation methods of the method, which are described below.

[0091] In an embodiment of the present specification, while executing step 206 to obtain the first sales forecast information corresponding to the target merchant, the second sales forecast information corresponding to the surrounding merchants can also be determined based on the first commodity sales log information and the second commodity sales log information using a multi-task model.

[0092] In actual application, the first sales forecast information and the second sales forecast information should generally meet some preset constraints. For example, considering that the first sales forecast information output by the main task is for a single target merchant, and the second sales forecast information output by the auxiliary task is for multiple merchants around the target merchant, therefore, for a certain category of goods, the sales forecast value of the target merchant should be smaller than the sales forecast value of the surrounding merchants.

[0093] In an optional embodiment, before sending the first sales forecast information to the merchant terminal of the target merchant in step 208, it is also possible to: determine whether the first sales forecast information and the second sales forecast information meet preset constraints to obtain a first determination result. Then, step 208 may specifically include: if the first determination result indicates that the first sales forecast information and the second sales forecast information meet the preset constraints, sending the first sales forecast information to the merchant terminal of the target merchant.

[0094] Among them, the preset constraint condition may include: for any target category, the predicted value of the target category in the first sales forecast information is less than the predicted value of the target category in the second sales forecast information.

[0095] In actual application, the first sales forecast information provided by the CRE model may be sales forecast information corresponding to commodity categories. For example, sales forecast information (A; B; C; D; E) corresponding to categories such as (fruits; vegetables; snacks; fresh food; daily necessities) may be provided.

[0096] In this case, the historical sales ratio of each commodity in each category can be further combined to obtain a hot product recommendation list corresponding to each category of commodities. Therefore, in an optional embodiment, the sending of the first sales forecast information to the merchant terminal of the target merchant can specifically include: first determining the hot product recommendation list information corresponding to each commodity category according to the first sales forecast information and the historical sales ratio information of each commodity in each commodity category involved in the first sales forecast information; and then sending the hot product recommendation list information to the merchant terminal of the target merchant.

[0097] The historical sales volume ratio information may be obtained based on order information statistics within a predetermined area and a preset time period. For example, the predetermined area may be an area including the target merchant and surrounding merchants of the target merchant, which may be set according to actual needs. The preset time period may be within the previous 24 hours of the current time, within the previous natural day, within the previous week, within the previous natural month, etc., which may be set according to actual needs.

[0098] Continuing with the above example, we can obtain the hot product recommendation list information corresponding to each category based on the sales forecast information (A; B; C; D; E) corresponding to the categories (fruits; vegetables; snacks; fresh food; daily necessities). For example, for the fruit category, combined with the historical sales share of various fruits, and based on the sales forecast value A of the fruit category, as an example, we can obtain the sales forecast values ​​(a1; a2; a3...) corresponding to fruits such as (apples; peaches; bananas...) arranged from high to low in terms of sales. For another example, for the vegetable category, combined with the historical sales share of various vegetables, and based on the sales forecast value B of the vegetable category, as an example, we can obtain the sales forecast values ​​(b1; b2; b3...) corresponding to vegetables such as (beans; cabbage; eggplant...) arranged from high to low in terms of sales.

[0099] In an optional embodiment, if the first judgment result indicates that the first sales forecast information and the second sales forecast information do not meet the preset constraints, the historical sales statistics information is sent to the merchant terminal of the target merchant; the historical sales statistics information is based on the sales statistics of the target merchant and the surrounding merchants of the target merchant within a previously preset time period.

[0100] like Figure 4 , which shows a flow chart of the method for generating recommendation information in an embodiment of this specification.

[0101] like Figure 4 As shown, the server can first obtain the product sales log information from the log system; then perform feature analysis to construct the features required by the multi-task model; input the features into the multi-task model of the CRE module for prediction to obtain the sales forecast results corresponding to the multi-task; then determine whether the sales forecast results meet the preset constraints. If the preset constraints are met, the historical sales share information of the product in the category dimension can be combined to obtain the hot product recommendation list information corresponding to each product category, and send it to the merchant terminal as recommendation information; if the preset constraints are not met, the historical sales statistics information can be sent to the merchant terminal as recommendation information.

[0102] like Figure 4As shown, after the server sends the recommendation information to the merchant terminal, the merchant terminal can provide the recommendation information to the merchant. In order to keep the system continuously updated, in some embodiments, the merchant terminal can generate feedback log information in response to the merchant's operation on the recommendation information, and send it back to the server for updating the log information.

[0103] Specifically, after the server sends the first sales forecast information to the merchant terminal of the target merchant, the server may also include: obtaining the target merchant's feedback log information for the hot product recommendation list information; the feedback log information is generated in response to the target merchant's operation on the hot product recommendation list information; the feedback log information is used to mark the updated product sales original feature vector. In actual applications, the target merchant's operation on the hot product recommendation list information may specifically include list switching operation, list browsing operation, list attention operation, list product browsing operation, list product attention operation, etc., but is not limited to this.

[0104] In practice, when the server sends recommendation information to the merchant terminal, in order to further improve the merchant's user experience, on the one hand, it is necessary to meet the timeliness requirement of information sending, and on the other hand, it is also necessary to avoid unnecessary disturbance to the merchant, for example, it should avoid disturbing the merchant's commodity sales process. In view of this, in the embodiments of this specification, the intelligent orchestration and delivery engine (IODE) module in the server can be used to determine the timing of sending recommendation information to the merchant terminal, so as to send commodity recommendation information in time when the merchant is idle.

[0105] like Figure 5 , shows a schematic diagram of the process of improving the information sending scheme of the embodiment of this specification. Figure 5 As shown in , in the server, the CRE module first generates commodity sales forecast information, then the IODE module determines the timing of information sending, and then the commodity sales forecast information is sent to the merchant terminal according to the information sending timing determined by the IODE module, and then the merchant terminal displays and broadcasts the information to the merchant.

[0106] The scheme for determining the timing of information transmission in the IODE module is described in detail below.

[0107] In the embodiments of this specification, in order to deliver personalized and targeted information to different merchants, the concept and method of reinforcement learning can be used to determine the appropriate information delivery time for the merchant, thereby achieving a timely and low-intrusive merchant experience.

[0108] Reinforcement learning is an unlabeled learning method based on feedback from sequential behaviors, which learns strategies through continuous trial and error. The reinforcement learning system includes an agent and an execution environment. The agent continuously learns and optimizes its strategy through interaction and feedback with the execution environment. Specifically, the agent observes and obtains the state of the execution environment, and determines the behavior or action to be taken according to the current state of the execution environment based on a certain strategy. Such behavior acts on the execution environment, changes the state of the execution environment, and generates feedback to the agent, which is also called a reward score. The agent determines whether the previous behavior is correct and whether the strategy needs to be adjusted based on the reward score obtained, and then updates its strategy. By repeatedly observing the state, determining the behavior, and receiving feedback, the agent can continuously update its strategy. The ultimate goal is to be able to learn a strategy that maximizes the cumulative reward score.

[0109] Figure 6 A schematic diagram showing a specific implementation scenario of the IODE module of the embodiment of this specification.

[0110] like Figure 6 As shown, in one or more embodiments of the present specification, the sales volume information of the merchant in the time period related to the current moment can be obtained (equivalent to obtaining the state s of the execution environment); then, a certain information sending strategy is adopted to determine whether to send and execute the information at the current moment (equivalent to determining the action a to be taken and executing it); then, according to the preset reward and punishment function, the reward score r is calculated, and this is used as a learning sample to update the information sending strategy. By repeatedly learning in this way, through continuous exploration and "trial and error", the information sending strategy can be optimized. Therefore, the optimized information sending strategy can be used to determine the appropriate information sending time for each merchant.

[0111] Figure 7 A flow chart showing a method for determining the timing of sending information to a merchant according to an embodiment of the present specification. From a program perspective, the execution subject of the flow can be a program installed on an application server. More specifically, it can be an IODE in the server.

[0112] like Figure 7 As shown, the process may include the following steps 702 to 710.

[0113] Step 702: Acquire the sales characteristic information of the target merchant at the current moment; the sales characteristic information reflects the commodity sales situation at the current moment.

[0114] Specifically, the obtaining of the sales characteristic information of the target merchant at the current moment may include: obtaining first sales information of the target merchant within a first preset time period before the current moment, the first preset time period ending at the current moment; determining the historical current moment corresponding to the current moment, and obtaining second sales information within a second preset time period after the historical current moment, the second preset time period starting at the historical current moment; and then, based on the first sales information and the second sales information, obtaining the sales characteristic information corresponding to the current moment.

[0115] In actual application, the target merchant's sales characteristic information at the current moment can be obtained when the merchant terminal of the target merchant is in an on state. More specifically, since it is necessary to obtain the first sales situation information of the target merchant in a preset period before the current moment, the target merchant's sales characteristic information at the current moment can be obtained after the merchant terminal of the target merchant is in an on state for at least the first preset period.

[0116] In the embodiments of the present specification, optionally, the historical current moment corresponding to the current moment may be a historical current moment with reference to natural time. As an example, if the current moment is 8:00 today, the historical current moment may be, for example, 8:00 yesterday. In practical applications, the sales volume of goods each hour may be counted for use in determining the sales characteristic information at the current moment. Of course, using 1 hour as a unit is only an example, and may also be used in units of 0.5 hours, 15 minutes, 2 hours, etc., which may be determined according to the density of the goods sales data.

[0117] Alternatively, the historical current moment corresponding to the current moment may be a historical current moment with reference to the merchant terminal opening time. As an example, if the current moment is the moment corresponding to the merchant terminal opening for a time period this time, the historical current moment may be, for example, the moment corresponding to the merchant terminal opening for a time period last time. In actual application, the sales volume of goods each hour after each opening of the merchant terminal may be counted for use in determining the sales characteristic information at the current moment. Of course, using 1 hour as a unit is only an example, and may also be used in units of 0.5 hours, 15 minutes, 2 hours, etc., which may be determined according to the density of the goods sales data.

[0118] like Figure 8 , which shows a schematic diagram of a method for acquiring sales characteristic information at the current moment in an embodiment of this specification.

[0119] like Figure 8As shown, the current moment is recorded as A, and the first sales situation information in the first preset period before the current moment can be counted. Optionally, the first preset period may include a first sub-preset period and a second sub-preset period, thereby, the order quantity Y in the first sub-preset period before the current moment and the order quantity X in the second sub-preset period before the first sub-preset period can be counted.

[0120] Since the sales after the current moment have not occurred, the second sales information within the second preset period after the historical current moment corresponding to the current moment can be obtained. Optionally, the second preset period can include a third sub-preset period and a fourth sub-preset period, thereby, the order quantity M within the third sub-preset period after the historical current moment and the order quantity N within the fourth sub-preset period after the third sub-preset period can be counted.

[0121] In actual application, optionally, the durations of the first sub-preset period, the second sub-preset period, the third sub-preset period, and the fourth sub-preset period may be the same. For example, they may all be 0.5 hours. In another optional embodiment, one or more of the durations of the first sub-preset period, the second sub-preset period, the third sub-preset period, and the fourth sub-preset period may also be different from each other.

[0122] When the plan is implemented, sales characteristic information such as order quantity X, order quantity Y, order quantity M and order quantity N can be statistically obtained when the current time is reached, and the information can be uploaded to IODE.

[0123] The obtained commodity sales feature information is equivalent to the current state s of the execution environment in the reinforcement learning system. Next, the action a to be taken in the current state s can be determined according to the information sending strategy. That is, step 704 is entered.

[0124] Step 704: Based on the sales characteristic information and according to the information sending strategy, determine information sending decision information; the information sending decision information includes sending the first sales forecast information at the current moment, or not sending the first sales forecast information at the current moment.

[0125] In the embodiments of the present specification, the purpose of making a decision based on the information sending strategy at any current moment is to determine whether the current moment is suitable for sending sales forecast information to the merchant.

[0126] In actual application, the information sending strategy can be implemented through various machine learning models, such as logistic regression model, gradient boosting decision tree GBDT model, deep neural network model, etc.

[0127] Initially, the information sending strategy can be a random strategy, which randomly decides whether to send information at the current moment. Of course, the random strategy is a temporary strategy during the initial "trial and error" and strategy exploration process. The information sending strategy will be continuously optimized and updated through environmental feedback.

[0128] After the relevant decision is generated in step 704, the process proceeds to step 706 to execute the decision.

[0129] Step 706: Execute the operation corresponding to the information sending decision information.

[0130] In step 706, the execution operation of the decision is equivalent to issuing an execution action a to the execution environment.

[0131] Step 708: Calculate reward and punishment information based on a first time length from the current time to the power-on time of the merchant terminal of the target merchant, and a second time length from a sales time after the current time that meets the specified conditions to the current power-on time.

[0132] The reward and punishment information may be a reward score r.

[0133] It can be understood that the reward score r in reinforcement learning is a measure of whether the previous action a is correct, and it is also the main basis for subsequent strategy adjustments. The goal of strategy adjustment is to maximize the accumulation of reward scores obtained from the action a.

[0134] In the scenario of sending recommended information in the embodiment of this specification, the purpose of strategy training is to enable the information sending strategy to send recommended information in a timely manner without disturbing the merchant. Therefore, the reward score can be set to reflect the timeliness of information sending through the first time length from the current moment to the power-on moment of the merchant terminal. It can be understood that the shorter the first time length, the more timely it is; and reflect the disturbance of information sending to the merchant through the second time length from the sales moment after the current moment to the current power-on moment. It can be understood that the longer the second time length, the less disturbing it is.

[0135] In a specific embodiment, the reward score of the information sending decision result can be determined as the weighted sum of the first duration and the second duration. Figure 8 , the calculation of reward and punishment information based on a first time length from the current time to the power-on time of the merchant terminal of the target merchant, and a second time length from the current time to the current power-on time when the selling time that meets the specified conditions is after the current time, may specifically include calculating the reward and punishment information based on the following reward and punishment function:

[0136] f = α(AS) + β(KA).

[0137] Wherein, f represents the reward score; A represents the current time, S represents the power-on time of the merchant terminal of the target merchant, K represents the sales time after the current time that meets the specified conditions; α and β are weighting coefficients, where α is less than 0 and β is greater than 0. The sales time that meets the specified conditions can be the first order after the current time.

[0138] According to the above reward and punishment function, it can be understood that AS can reflect the timeliness of information sending, and the smaller the difference, the better; while KA can reflect the low-intrusiveness of information sending, that is, try not to disturb the merchant's commodity sales process, and the larger the difference, the better. Combined with the weighted coefficient α is less than 0 and β is greater than 0, the more timeliness and low-intrusiveness are met, the higher the reward score.

[0139] Step 710: Based on at least the reward and punishment information, a reinforcement learning algorithm is used to update the information sending strategy; the updated information sending strategy is used to determine information sending decision information.

[0140] According to the concept of strategy updating in reinforcement learning, in the embodiments of this specification, the information sending strategy is continuously updated according to the feedback information (i.e., the reward score r), and the goal of the update is to ensure that, for the same commodity sales feature information (environmental state s), the information sending decision information (action a) determined by the information sending strategy can obtain more expected cumulative reward scores.

[0141] In order to achieve such a goal, in an optional embodiment, a strategy update may be performed each time a reward score corresponding to a merchant is obtained. In another optional embodiment, feedback information corresponding to multiple information sending decision information from multiple merchants may be collected as a training sample set, and the information sending strategy may be trained using the training sample set to update the information sending strategy.

[0142] Specifically, for a target merchant, after executing the information sending decision information, the sales information of the target merchant can be obtained and the reward score corresponding to this information sending decision information can be calculated. Then, the target merchant's commodity sales feature information (equivalent to the environment state s), the information sending decision information (equivalent to action a) and the reward score (r) can be taken as a training sample and added to the training sample set. In this way, the training sample set can contain multiple<s,a,r> The training samples in the form of .

[0143] Optionally, in the scenario of continuous interaction, the new commodity sales feature information obtained by the current merchant when executing the corresponding operation of the information sending decision information can also be obtained, which is equivalent to obtaining the new state s' to which the execution environment migrates after executing action a.<s,a,r,s'> The training samples in the form of .

[0144] Based on the training samples obtained above, a reinforcement learning algorithm can be used to update the information sending strategy using the training sample set until the model converges.

[0145] In the embodiments of this specification, considering that the order quantity is a quantity with a large difference in value, the calculation complexity is extremely high, therefore, the reinforcement learning algorithm can adopt a reinforcement learning algorithm based on a neural network. For example, it can include a Q-learning algorithm based on a fuzzy neural network, a Sarsa algorithm based on a BP neural network, etc.

[0146] Figure 7 The method in the paper uses an intelligent information delivery scheme based on deep reinforcement learning to display / broadcast product recommendation information in a timely manner when the merchant is idle. Based on this scheme, with very few parameters, the process of information recommendation for merchants is timely and low-intrusive, which can improve the merchant's user experience.

[0147] According to the above description, the present specification provides a flowchart of a method for sending and receiving hot product recommendation information in an actual application scenario, as shown in FIG. Fig. 9 shown.

[0148] Return to reference Figure 1 , the server can provide software as a service (SaaS), and in practice, merchants can obtain the services corresponding to the solutions of the embodiments of this specification by subscription. In the embodiments of this specification, the SaaS service may include a commodity recommendation engine (CRE) and an intelligent orchestration and delivery engine (IODE), and in actual application, the recommendation information may be delivered to the hardware merchant terminal and / or the software merchant terminal.

[0149] The merchant terminal may include a hardware terminal and a software terminal. Among them, the hardware terminal may include a smart box. For example, the smart box may have both the function of a clock and the function of information display / broadcasting. Optionally, it may also have the function of receiving information from merchants. In the embodiments of this specification, by improving and enriching the functions of the existing smart box, merchants can use the services corresponding to the solutions of the embodiments of this specification without updating the terminal hardware devices, which provides convenience and saves costs for merchants. The software terminal may include a mobile phone APP or applet installed in a mobile terminal. In the embodiments of this specification, the push information sent by the server is received through the software terminal, which can overcome the defects of the traditional inconvenience of information reception caused by the fixed position of the smart box, so that merchants can receive push information anytime and anywhere, and also provide guarantee for the timeliness of information reception.

[0150] like Fig. 9As shown in , in an optional example, when the merchant opens the merchant terminal (hardware; for example, a product clock), the product clock can display the clock. In actual application, the display screen of the product clock can be used to display clock information at ordinary times, and when receiving the hot product recommendation information sent by the server, the hot product recommendation information can be displayed. The product clock can also have an intelligent broadcast function for voice broadcasting the hot product recommendation information sent by the server.

[0151] Optionally, the product clock may also have a code scanning area for obtaining product information.

[0152] In a specific application scenario, merchants can perform product query operations in the scanning area according to the product's unique identification code (for example, barcode, QR code, etc.), and the product clock can submit the product's unique identification code (or the product name obtained based on the product's unique identification code) and the location information of the product clock to the SaaS service. Subsequently, the SaaS service can return relevant product information to the product clock (for example, product peripheral sales ranking, product sales category ranking, product sales category rising list, product sales rising list and other hot product recommendation information), and then the product clock can broadcast and display the relevant product information to the merchant. In actual application, the queried product can be specially marked and displayed in the relevant product information obtained based on the merchant's active query.

[0153] In an optional embodiment of the present specification, before obtaining the first commodity sales log information of the target merchant, the server may also include: obtaining the location information of the target merchant and the target commodity information queried by the target merchant; thereby, the first sales forecast information obtained may include hot product recommendation list information corresponding to the commodity category to which the target commodity belongs.

[0154] In another specific application scenario, when a merchant opens the corresponding terminal software bound to the hardware merchant terminal account, the terminal software can then send the merchant location information to the SaaS service so that the SaaS service can provide hot product recommendation information. Specifically, the hot product recommendation information can include a variety of product recommendation schemes for different categories of goods. Merchants can switch between multiple schemes on the mobile terminal, and can also comment, recommend, collect and other operations. In addition, in actual application, after the merchant clicks to open the corresponding terminal software, it can automatically receive the hot product recommendation information issued by the SaaS service; or, it can also actively obtain the hot product recommendation information provided by the SaaS service through functions such as product search.

[0155] In another specific application scenario, when the SaaS service determines that the current time is a suitable time for information delivery, the SaaS service can proactively arrange and deliver the recommendation plan, so as to display and broadcast the recommendation plan with the help of the product clock.

[0156] Based on the above method provided in the embodiments of this specification, on the one hand, a multi-task learning framework based on Attention is proposed. Taking into account the problems of commodity distribution and sparse sales in offline stores, the sales information of commodities is predicted by collaborative optimization through a combination of main and auxiliary tasks, which can eventually achieve the effect of "thousands of stores with thousands of faces" and meet the needs of offline merchants to adjust their purchasing plans in a timely manner. On the other hand, a scheduling and delivery scheme based on deep reinforcement learning is proposed, which can achieve a timely and low-intrusive user experience with very few parameters. On the other hand, a scheme for making commodity recommendations based on the commodity clock is proposed, which can not only complete the function of clock display but also broadcast and display commodity information, thus enhancing the function of the clock.

[0157] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method. Fig.10 The embodiments of this specification provide corresponding to Figure 2 A structural diagram of an information sending device. Fig.10 As shown, the device may include:

[0158] The first acquisition module 1002 is used to acquire the first commodity sales log information of the target merchant;

[0159] The second acquisition module 1004 is used to acquire the second commodity sales log information of the surrounding merchants of the target merchant;

[0160] The sales volume forecasting module 1006 is used to determine the first sales volume forecasting information corresponding to the target merchant based on the first commodity sales log information and the second commodity sales log information by using a multi-task model;

[0161] The information sending module 1008 is used to send the first sales forecast information to the merchant terminal of the target merchant.

[0162] It is understood that the above modules refer to computer programs or program segments for executing one or more specific functions. In addition, the distinction between the above modules does not mean that the actual program codes must also be separated.

[0163] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method.

[0164] Fig.11 The embodiments of this specification provide corresponding to Figure 2 A structural diagram of an information sending device. Fig.11 As shown, the device 1100 may include:

[0165] at least one processor 1110; and,

[0166] A memory 1130 is communicatively connected to the at least one processor; wherein,

[0167] The memory 1130 stores instructions 1120 executable by the at least one processor 1110. The instructions are executed by the at least one processor 1110 to enable the at least one processor 1110 to:

[0168] Obtain the first product sales log information of the target merchant;

[0169] Acquire second commodity sales log information of surrounding merchants of the target merchant;

[0170] Based on the first commodity sales log information and the second commodity sales log information, using a multi-task model, determining first sales volume forecast information corresponding to the target merchant;

[0171] The first sales forecast information is sent to a merchant terminal of the target merchant.

[0172] Based on the same idea, the embodiment of this specification also provides a computer-readable medium corresponding to the above method. The computer-readable medium stores computer-readable instructions, which can be executed by a processor to implement the following method:

[0173] The above describes specific embodiments of the present specification. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0174] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other.

[0175] The devices, equipment and methods provided in the embodiments of this specification correspond to each other, and therefore, the devices and equipment also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the corresponding devices and equipment will not be repeated here.

[0176] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0177] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0178] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0179] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0180] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0181] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0182] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0183] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0184] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0185] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0186] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0187] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0188] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0189] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for sending information, comprising: Obtain the first product sales log information of the target merchant; Acquire second commodity sales log information of surrounding merchants of the target merchant; Based on the first commodity sales log information and the second commodity sales log information, using a multi-task model, determining first sales volume forecast information corresponding to the target merchant; sending the first sales forecast information to a merchant terminal of the target merchant; The sending the first sales forecast information to the merchant terminal of the target merchant specifically includes: Acquire the sales characteristic information of the target merchant at the current moment; the sales characteristic information reflects the sales situation of the commodity at the current moment; Based on the sales characteristic information, according to the information sending strategy, determining information sending decision information; the information sending decision information includes: sending the first sales forecast information at the current moment, or not sending the first sales forecast information at the current moment; the information sending strategy is implemented by a machine learning model; Execute the operation corresponding to the information sending decision information; Calculate reward and punishment information based on a first time length from the current time to the power-on time of the merchant terminal of the target merchant, and a second time length from the current time to a sales time meeting a specified condition after the current time; Based on at least the reward and punishment information, a reinforcement learning algorithm is used to update the information sending strategy.

2. The method according to claim 1, before obtaining the second commodity sales log information of the surrounding merchants of the target merchant, further comprising: Get the location information of the target merchant; Based on the location information of the target merchant, surrounding merchants of the target merchant are determined.

3. The method according to claim 1, wherein the determining the first sales volume forecast information corresponding to the target merchant based on the first commodity sales log information and the second commodity sales log information using a multi-task model specifically comprises: determining a commodity sales feature vector according to the first commodity sales log information and the second commodity sales log information; Based on the commodity sales feature vector and the gating weights corresponding to the sales volume prediction tasks of each merchant, first sales volume prediction information corresponding to the target merchant is determined.

4. The method according to claim 3, wherein determining the commodity sales feature vector according to the first commodity sales log information and the second commodity sales log information comprises: Constructing an original feature vector for inputting into a commodity sales prediction model according to the first commodity sales log information and the second commodity sales log information; Dividing the original feature vector of commodity sales into a dense feature vector, a sparse feature vector and a sequence feature vector; Determine the embedding modes corresponding to the dense feature vector, the sparse feature vector and the sequence feature vector respectively; According to the respective corresponding embedding modes, the dense feature vector, the sparse feature vector and the sequence feature vector are mapped respectively to obtain corresponding commodity sales sub-feature vectors; Based on the commodity sales sub-feature vectors corresponding to the dense feature vector, the sparse feature vector and the sequence feature vector, a commodity sales feature vector corresponding to the commodity sales original feature vector is obtained.

5. The method according to claim 3, wherein determining the first sales forecast information corresponding to the target merchant based on the commodity sales feature vector and the gating weight corresponding to each merchant sales forecast task specifically comprises: Inputting the commodity sales feature vector into multiple expert models to obtain latent feature vectors corresponding to each of the multiple expert models; Inputting the commodity sales feature vector into the gating model corresponding to each merchant sales prediction task to obtain the attention gating weight of each merchant sales prediction task; For the target merchant sales prediction task in each merchant sales prediction task, weighted summation is performed based on the attention gating weight of the target merchant sales prediction task and the latent feature vectors corresponding to each of the multiple expert models to obtain a comprehensive latent feature vector of the target merchant sales prediction task; The comprehensive latent feature vector of the target merchant sales prediction task is input into the tower model corresponding to the target merchant to obtain the first sales prediction information corresponding to the target merchant.

6. The method according to claim 5, wherein the step of inputting the commodity sales feature vector into a gating model corresponding to each merchant sales prediction task to obtain the attention gating weight of each merchant sales prediction task specifically comprises: Calculate the original gating weights of each merchant's sales forecasting task; Use the attention function to calculate the weighted attention value vector; Based on the original gating weight and the weighted attention value vector, the attention gating weight of each merchant sales prediction task is obtained.

7. The method according to claim 1, after obtaining the second commodity sales log information of the surrounding merchants of the target merchant, further comprising: Based on the first commodity sales log information and the second commodity sales log information, using a multi-task model, determining second sales forecast information corresponding to the surrounding merchants; Before sending the first sales forecast information to the merchant terminal of the target merchant, the method further includes: Determine whether the first sales forecast information and the second sales forecast information meet preset constraints to obtain a first determination result; The sending the first sales forecast information to the merchant terminal of the target merchant specifically includes: If the first judgment result indicates that the first sales forecast information and the second sales forecast information meet the preset constraint conditions, the first sales forecast information is sent to the merchant terminal of the target merchant.

8. The method according to claim 7, wherein the preset constraint condition comprises: For any target category, the predicted value of the target category in the first sales forecast information is less than the predicted value of the target category in the second sales forecast information.

9. The method according to claim 7, wherein the first sales forecast information is sales forecast information corresponding to a commodity category; The sending the first sales forecast information to the merchant terminal of the target merchant specifically includes: Determine hot product recommendation list information corresponding to each commodity category according to the first sales forecast information and historical sales ratio information of each commodity in each commodity category involved in the first sales forecast information; The hot product recommendation list information is sent to the merchant terminal of the target merchant.

10. The method according to claim 9, after sending the first sales forecast information to the merchant terminal of the target merchant, further comprising: Obtain feedback log information of the target merchant regarding the hot product recommendation list information; The feedback log information is generated in response to the target merchant's operation on the hot product recommendation list information; the feedback log information is used to mark the updated product sales original feature vector.

11. The method according to claim 1, wherein obtaining the sales characteristic information of the target merchant at the current moment specifically comprises: Obtaining first sales information of the target merchant within a first preset time period before the current moment; The first preset time period ends at the current moment; Determine a historical current moment corresponding to the current moment; Acquire second sales situation information within a second preset time period after the current historical moment; The second preset time period takes the historical current moment as a starting point; Based on the first sales situation information and the second sales situation information, sales characteristic information corresponding to the current moment is obtained.

12. The method of claim 1, wherein the calculation of reward and punishment information based on a first time length from the current time to the power-on time of the merchant terminal of the target merchant and a second time length from the current time to the selling time that meets the specified conditions after the current time, specifically comprises: The reward and punishment information is calculated based on the following reward and punishment function: f=α(AS)+β(KA) Among them, f represents the reward score; A represents the current time, S represents the power-on time of the merchant terminal of the target merchant, K represents the sales time that meets the specified conditions after the current time; α and β are weighting coefficients, where α is less than 0 and β is greater than 0.

13. The method of claim 1, wherein the reinforcement learning algorithm comprises a reinforcement learning algorithm based on a neural network.

14. The method according to claim 1, before obtaining the first commodity sales log information of the target merchant, further comprising: Obtaining location information of a target merchant and target product information queried by the target merchant; The first sales forecast information includes hot product recommendation list information corresponding to the product category to which the target product belongs.

15. An information sending device, comprising: A first acquisition module, used to acquire the first commodity sales log information of the target merchant; A second acquisition module is used to acquire second commodity sales log information of surrounding merchants of the target merchant; a sales volume forecasting module, configured to determine first sales volume forecasting information corresponding to the target merchant using a multi-task model based on the first commodity sales log information and the second commodity sales log information; An information sending module, used for sending the first sales forecast information to a merchant terminal of the target merchant; The information sending module is specifically used for: Acquire the sales characteristic information of the target merchant at the current moment; the sales characteristic information reflects the sales situation of the commodity at the current moment; Based on the sales characteristic information and according to the information sending strategy, determining information sending decision information; The information sending decision information includes: sending the first sales forecast information at the current moment, or not sending the first sales forecast information at the current moment; the information sending strategy is implemented by a machine learning model; Execute the operation corresponding to the information sending decision information; Calculate reward and punishment information based on a first time length from the current time to the power-on time of the merchant terminal of the target merchant, and a second time length from the current time to a sales time meeting a specified condition after the current time; Based on at least the reward and punishment information, a reinforcement learning algorithm is used to update the information sending strategy.

16. An information sending device, comprising: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Obtain the first product sales log information of the target merchant; Acquire second commodity sales log information of surrounding merchants of the target merchant; Based on the first commodity sales log information and the second commodity sales log information, using a multi-task model, determining first sales volume forecast information corresponding to the target merchant; sending the first sales forecast information to a merchant terminal of the target merchant; The sending the first sales forecast information to the merchant terminal of the target merchant specifically includes: Acquire the sales characteristic information of the target merchant at the current moment; the sales characteristic information reflects the sales situation of the commodity at the current moment; Based on the sales characteristic information, according to the information sending strategy, determining information sending decision information; the information sending decision information includes: sending the first sales forecast information at the current moment, or not sending the first sales forecast information at the current moment; the information sending strategy is implemented by a machine learning model; Execute the operation corresponding to the information sending decision information; Calculate reward and punishment information based on a first time length from the current time to the power-on time of the merchant terminal of the target merchant, and a second time length from the current time to a sales time meeting a specified condition after the current time; Based on at least the reward and punishment information, a reinforcement learning algorithm is used to update the information sending strategy.

17. A computer-readable medium having computer-readable instructions stored thereon, wherein the computer-readable instructions can be executed by a processor to implement the information sending method according to any one of claims 1 to 14.

Images