Business push method and push system

Abstract

The embodiment of the present invention provides a business push method and a recommendation system, the business push method comprising: receiving a first selection instruction sent by a user's account for selecting a target object for a business migration to be performed; obtaining a target amount of the target object according to the first selection instruction; judging whether the target object meets the resource allocation limit according to the target amount, if not, in response to the first selection instruction, pushing a matching object to the account that initiates the business migration to be performed on the target object, wherein the matching object has a matching amount so that the total amount of the target amount and the matching amount meets the resource allocation limit; if satisfied, in response to the first selection instruction, pushing the associated object of the target object to the account that initiates the business migration to be performed on the target object. Through this method, matching objects that need to be pushed to users can be intelligently identified, so that the sum of the matching amount of the matching object and the target amount of the target object meets the resource allocation limit, thereby meeting the needs of users.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a service push method and a push system. Background Art

[0002] When users shop online, e-commerce platforms or stores often have certain shopping thresholds. For example, the single purchase amount must reach a certain value to be exempt from shipping charges. Therefore, after completing the original shopping needs, users usually make additional purchases to reach the free shipping threshold, that is, to complete the order. The same demand also occurs in other shopping activities that have restrictions on the single purchase amount, such as full discounts and full gifts. Therefore, e-commerce platforms will recommend some alternative products to customers, that is, push objects.

[0003] The current selection of push objects is often based on manual operations, which are difficult to identify, easy to confuse, and have high labor costs and low efficiency, making it difficult to carry out under massive data. In addition, the current push method is only based on simple rule filtering, with low accuracy and unsatisfactory recommendation effect, and it is difficult to significantly reduce the difficulty for users to select items to complete an order. Moreover, there are many types of push objects, and the user selection process is cumbersome, which increases the difficulty of selection.

[0004] Therefore, the inventors believe that the above-mentioned push method has great limitations. It is difficult to meet user needs based on simple rule filtering and manual identification recommendations. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a business push method and recommendation system, which determines whether the target amount of the target object meets the resource allocation limit. If not, a matching object is pushed to it, so that the sum of the matching amount corresponding to the matching object and the target amount corresponding to the target object meets the resource allocation limit, thereby reasonably realizing the push of the target object to the user and meeting the user's needs.

[0006] According to a first aspect of the present invention, there is provided a service push method, comprising:

[0007] Receiving a first selection instruction sent by a user account for selecting a target object for business migration to be performed;

[0008] According to the first selection instruction, obtaining the target amount of the target object;

[0009] Determine whether the target object meets the resource allocation limit according to the target amount,

[0010] If not, in response to the first selection instruction, a matching object is pushed to the account that initiates the to-be-executed business migration for the target object, wherein the matching object has a matching quota, so that the total amount of the target quota and the matching quota meets the resource allocation limit;

[0011] If satisfied, in response to the first selection instruction, the associated objects of the target object are pushed to the account that initiates the to-be-performed business migration for the target object.

[0012] Preferably, pushing the matching object to the account that initiates the to-be-performed business migration for the target object includes:

[0013] The matching objects are sorted using a sorting model, and the matching objects are pushed sequentially in the sorting order.

[0014] Preferably, after pushing the matching object to the account that initiates the to-be-executed business migration for the target object, the method further includes:

[0015] Receiving a second selection instruction sent by a user account for selecting the matching object for which business migration is to be performed;

[0016] According to the second selection instruction, obtaining the matching amount of the matching object;

[0017] Determine whether the total amount of the target amount and the matching amount meets the resource allocation limit,

[0018] If not, in response to the second selection instruction, the matching object is pushed again to the account that initiates the to-be-executed business migration for the target object until the total amount of the target amount and the matching amount meets the resource allocation limit.

[0019] Preferably, before pushing the matching object to the account that initiates the to-be-executed business migration for the target object, the method further includes:

[0020] A screening algorithm is used to screen out the matching object from a plurality of target objects to be selected according to preset screening parameters.

[0021] Preferably, the screening parameters include fusion confidence of the target object.

[0022] Preferably, using a screening algorithm to screen out the matching object from a plurality of target objects to be selected according to a preset sorting rule comprises:

[0023] Using at least one model to independently calculate the classification confidence of each of the target objects;

[0024] Integrating the classification confidences calculated under the at least one model to generate the fusion confidence corresponding to each target object; and

[0025] A confidence threshold is set, and the target objects whose fusion confidence is greater than the confidence threshold are screened out as the matching objects.

[0026] Preferably, the step of independently calculating the classification confidence of each target object using at least one model comprises:

[0027] Establishing a correlation graph model between the target objects, and calculating a first classification confidence of the target objects according to the correlation between the target objects;

[0028] Establishing a conversion rate distribution model for the target object, and calculating a second classification confidence of the target object according to a probability distribution table; and

[0029] A classifier model related to the content attribute of the target object is established, and a third classification confidence level about the target object is obtained after optimization.

[0030] Preferably, the classifier includes an integrated tree with regularization terms, logistic regression and a neural network.

[0031] Preferably, the values ​​of the classification confidence, the fusion confidence and the confidence threshold are all distributed between 0-1.

[0032] Preferably, the associated object includes the target object to be selected after removing the matching object.

[0033] According to a second aspect of the present invention, there is provided a service push system, comprising:

[0034] A first receiving unit, configured to receive a first selection instruction sent by a user's account for selecting a target object for a business migration to be performed;

[0035] A first acquisition unit, configured to acquire a target amount of the target object according to the first selection instruction;

[0036] A first judgment unit is used to judge whether the target object meets the resource allocation limit according to the target amount,

[0037] a first response unit, configured to, when not satisfied, push a matching object to the account that initiates the to-be-performed business migration for the target object in response to the first selection instruction, wherein the matching object has a matching amount so that a total amount of the target amount and the matching amount satisfies a resource allocation limit;

[0038] A second response unit is configured to, when satisfied, respond to the first selection instruction and push the associated object of the target object to the account that initiates the to-be-executed business migration for the target object.

[0039] Preferably, the first response unit is further configured to sort the matching objects using a sorting model, and push the matching objects in sequence according to the sorting order.

[0040] Preferably, the service push system further includes:

[0041] A second receiving unit, configured to receive a second selection instruction sent by a user's account for selecting the matching object for the to-be-performed business migration;

[0042] A second acquisition unit, configured to acquire a matching amount of the matching object according to the second selection instruction;

[0043] The second judgment unit is used to judge whether the total amount of the target amount and the matching amount meets the resource allocation limit.

[0044] A loop unit is used to, when not satisfied, respond to the second selection instruction and push the matching object to the account that initiates the to-be-executed business migration for the target object again until the total amount of the target amount and the matching amount meets the resource allocation limit.

[0045] Preferably, the service push system further includes:

[0046] The screening unit is used to use a screening algorithm to screen out the matching object from a plurality of target objects to be selected according to preset screening parameters.

[0047] Preferably, the screening parameters include fusion confidence of the target object.

[0048] Preferably, the screening unit comprises:

[0049] A calculation unit, configured to independently calculate the classification confidence of each target object using at least one model;

[0050] an integration unit, configured to integrate the classification confidences calculated under the at least one model to generate the fusion confidence corresponding to each of the target objects; and

[0051] The classification unit is used to set a confidence threshold and filter out the target objects whose fusion confidence is greater than the confidence threshold as the matching objects.

[0052] According to a third aspect of the present invention, there is provided a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed, the service push method as described above is implemented.

[0053] According to a fourth aspect of the present invention, there is provided a service push device, comprising: a memory for storing computer instructions; a processor coupled to the memory, the processor being configured to execute the service push method as described above based on the computer instructions stored in the memory.

[0054] The embodiments of the present invention have the following advantages or beneficial effects: first, the target amount of the target object selected by the user is obtained according to the user's selection instruction, and then it is determined whether the target amount of the target object meets the resource allocation limit. If not, a matching object is pushed to it, so that the sum of the matching amount corresponding to the matching object and the target amount corresponding to the target object meets the resource allocation limit; if so, the associated object of the target object is pushed to it, so that different selection objects can be recommended to the user in different modes, saving the user's selection time, making the user's selection more reasonable and reliable, meeting the user's different selection needs, and improving the user experience.

[0055] Another preferred embodiment of the present invention has the following advantages or beneficial effects: multiple data sources and models for independently identifying matching objects are used, classification confidences are calculated separately, and finally they are integrated to generate fusion confidence, thereby identifying matching objects. Multiple models extract information of matching objects from different angles, and after fusion, higher accuracy can be achieved, making the identification of matching objects more accurate and more in line with the actual needs of customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The above and other objects, features and advantages of the present invention will become more apparent through the description of the embodiments of the present invention with reference to the following drawings, in which:

[0057] Figure 1 A flow chart showing a service push method in an embodiment of the present invention is shown;

[0058] Figure 2 A flowchart of a service push method summarized in an embodiment of the present invention is shown;

[0059] Figure 3 Shows Figure 2 The specific flow chart of step S205 shown;

[0060] Figure 4 A structural diagram of a service push system in an embodiment of the present invention is shown;

[0061] Figure 5 A structural diagram of a service push system summarized in an embodiment of the present invention is shown;

[0062] Figure 6 A structural diagram of the screening unit 501 of the service push system in an embodiment of the present invention is shown;

[0063] Figure 7 A structural diagram of a service push device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0064] The present invention is described below based on embodiments, but the present invention is not limited to these embodiments. In the detailed description of the present invention below, some specific details are described in detail. It is possible for a person skilled in the art to fully understand the present invention without the description of these details. In order to avoid confusing the essence of the present invention, known methods, processes, and procedures are not described in detail. In addition, the drawings are not necessarily drawn to scale.

[0065] Figure 1 A flow chart of a service push method in an embodiment of the present invention is shown, and specific steps include S101-S105.

[0066] As described in the background technology, existing business push methods have low recognition and are easy to confuse. They are often based on manual operations to push matching objects, which is easy to consume manpower. In addition, the filtering rules for matching objects are relatively simple, and the recommendation effect is not ideal. Matching objects will not be filtered or otherwise processed, affecting the user experience.

[0067] Therefore, in this embodiment, it is determined whether the target amount of the target object meets the resource allocation limit. If not, a matching object is pushed to it so that the sum of the matching amount corresponding to the matching object and the target amount corresponding to the target object meets the resource allocation limit; if so, the associated objects of the target object are pushed to it, so that different selection objects can be recommended to the user in different modes, saving the user's selection time.

[0068] In step S101, a first selection instruction for selecting a target object for business migration to be performed is received from a user's account.

[0069] First, a first selection instruction sent by a user's account is received, and the instruction is used to select a target object, which is the object to be migrated. In an e-commerce system, for example, the object to be purchased added by a user's account lock is first obtained.

[0070] In step S102, according to the first selection instruction, the target amount of the target object is obtained.

[0071] Each target object has a corresponding target amount, and according to the first selection instruction, all target objects are acquired, and the target amounts corresponding to all target objects are counted. The target amount is, for example, the total price of the items to be purchased selected by the user.

[0072] In step S103, it is determined whether the target object meets the resource allocation limit according to the target amount.

[0073] Set a resource allocation limit and determine whether the target amounts corresponding to all target objects meet the resource allocation limit. For example, determine whether the total price of all items to be purchased reaches the amount for free shipping. If not, you can recommend objects to users for completing the order.

[0074] In step S104, if not satisfied, in response to the first selection instruction, a matching object is pushed to the account that initiates the pending business migration for the target object, wherein the matching object has a matching quota so that the total amount of the target quota and the matching quota meets the resource allocation limit.

[0075] Since the target quotas corresponding to all target objects do not meet the resource allocation limit, the matching objects are pushed to the account, and the matching quotas corresponding to the matching objects are combined with the target quotas corresponding to the target objects, which can provide a basis for meeting the resource allocation limit. For example, some matching objects with large sales and low prices are recommended to users so that the total price of the two objects reaches a price that can be waived or reach a discount.

[0076] In step S105 , if the conditions are met, in response to the first selection instruction, the associated objects of the target object are pushed to the account that initiates the to-be-executed business migration for the target object.

[0077] Since the target amount corresponding to all target objects has met the resource allocation limit, some related objects associated with the target object can be recommended to the user. For example, since the total price of the object selected by the user has reached the discount qualification or free shipping threshold, some non-matching objects similar to the selected object can be recommended to the user to enrich the user's selection.

[0078] The service push method of the embodiment of the present invention first obtains the target amount of the target object selected by the user according to the user's selection instruction, and then determines whether the target amount of the target object meets the resource allocation limit. If not, a matching object is pushed to it so that the sum of the matching amount corresponding to the matching object and the target amount corresponding to the target object meets the resource allocation limit; if so, the associated object of the target object is pushed to it, so that different selection objects can be recommended to the user in different modes, saving the user's selection time, making the user's selection more reasonable and reliable, meeting the user's different selection needs, and improving the user experience.

[0079] Figure 2 A flowchart of a service push method summarized in an embodiment of the present invention is shown, which specifically includes the following steps.

[0080] In step S201, a first selection instruction for selecting a target object for business migration to be performed is received from a user's account.

[0081] In step S202, according to the first selection instruction, the target amount of the target object is obtained.

[0082] In step S203, it is determined whether the target object meets the resource allocation limit according to the target amount.

[0083] In step S204, in response to the first selection instruction, the associated objects of the target object are pushed to the account that initiates the service migration to be performed on the target object.

[0084] In step S205, a screening algorithm is used to screen out matching objects from a plurality of target objects to be selected according to preset screening parameters.

[0085] In step S206, in response to the first selection instruction, a matching object is pushed to the account that initiates the to-be-executed business migration for the target object, wherein the matching object has a matching quota so that the total amount of the target quota and the matching quota meets the resource allocation limit.

[0086] In step S207, a second selection instruction for selecting a matching object for a to-be-performed business migration is received from the user's account.

[0087] In step S208, according to the second selection instruction, the matching amount of the matching object is obtained.

[0088] In step S209, it is determined whether the total amount of the target amount and the matching amount meets the resource allocation limit.

[0089] In step S210, if it is not satisfied, in response to the second selection instruction, the matching object is pushed again to the account that initiates the pending business migration for the target object until the total amount of the target amount and the matching amount meets the resource allocation limit.

[0090] This embodiment is a more complete service push method for matching objects than the previous embodiment. Figure 1 Steps S101-S105 are the same as those in the embodiment and will not be described in detail here.

[0091] In step S205, a screening algorithm is used to screen out matching objects from a plurality of target objects to be selected according to preset screening parameters.

[0092] When the target amount of the target object does not meet the resource allocation limit, the matching object is pushed to the user. This step is the matching object selection step. Some screening algorithms are used to screen out the matching object from multiple target objects to be selected in advance.

[0093] The screening parameters include the fusion confidence of the target object. For example, first select multiple objects as target objects based on historical order data, user browsing history, user praise data, etc. within a period of time. Then use a data source and model that can independently identify the target object, use certain algorithms and indicators, calculate the fusion confidence that the target object can become a matching object, and determine which target objects are suitable as matching objects for business push. Figure 3 This step is described in detail in .

[0094] In step S206, in response to the first selection instruction, the matching object is pushed to the account that initiates the business migration to be executed on the target object, including: sorting the matching objects using a sorting model, and sequentially pushing the matching objects in the sorting order.

[0095] Matching objects are pushed to users to meet their needs for completing orders due to free shipping, discounts for purchases above a certain amount, etc. In one embodiment, when pushing matching objects, selection can be made based on the attributes of the target objects and a set sorting model.

[0096] First, based on the actual scenarios of e-commerce business, multiple target objects to be recommended are filtered by rules, including price, weight, warehouse, etc., to ensure that only target objects that meet the order requirements can be displayed.

[0097] Secondly, set the sorting model, for example, adjust the sorting model according to the fusion confidence of the target object to ensure that the top-ranked matching objects can be pushed first. Specifically, when solving the sorting model, correct its gradient, that is:

[0098]

[0099] where g i is the original gradient corresponding to the i-th sample, Confidence i is the fusion confidence that the i-th sample belongs to the matching object. f is a specific weighting function, which can be taken as confidence i Itself, or the confidence after square root or logarithmic transformation, etc., this embodiment weights the gradient of the matching object.

[0100] The recommended order of matching objects is obtained according to the revised sorting model, and the matching objects are pushed to the user in the recommended order.

[0101] By correcting the gradient, the sorting model will be more inclined to fit the matching objects when solving, so that the matching objects can be effectively pushed to the user, shortening the time it takes for the user to find the matching objects. According to the sorting model corresponding to the corrected gradient, the matching objects that need to be recommended are prioritized so that the objects that are most suitable as matching objects are displayed first.

[0102] In another embodiment, the associated object includes the target object to be selected after the matching object is removed. When the associated object is pushed to the user, that is, when the matching object is not needed, the weight of the matching object in the push system and the priority in the sorting can be reduced. In a non-order-matching scenario, the matching object must be processed first and pushed last. Because although the matching object has a lower unit price and a higher conversion rate, it is not an item that the user needs on a daily basis, and the forward push position will affect the user's choice. In order to avoid too many matching objects or ranking too high, which affects the user's selection, it is necessary to downgrade the matching object and give priority to recommending the associated object.

[0103] Specifically, the adjustment methods include correcting the gradient and adding regularization items, so that when the user does not need to complete the order, the sorting priority of the matching objects is lowered, and non-matching objects are displayed to the customer first, and then a small number of high-quality matching objects are displayed.

[0104] In an optional embodiment, a sorting model is used to compress the sorting gradient of the matching object according to the fusion confidence, so as to reduce the priority of the matching object in the sorting. For example, a sorting model can be first established, and its gradient can be corrected, which can be expressed by the formula:

[0105]

[0106] in is the gradient of the modified sorting model, g i is the original gradient corresponding to the i-th sample, Confidence i is the fusion confidence that the i-th sample belongs to the matching object. f is a specific weighting function, for example, the inverse of the fusion confidence Or the value of the fusion confidence after square root or logarithmic transformation, etc. It can be seen that the non-matching recommendation adopts the method of gradient compression, which reduces the priority of the matching object in the sorting.

[0107] In one embodiment, for the matching object, an objective function with a regularization term including fusion confidence is constructed, and the regularization term is adjusted to reduce the weight of the matching object. Specifically, for example, the objective function of the model training can be established as:

[0108]

[0109] Here, an integrated tree model is used to construct K independent decision trees. The function after adding regular terms is:

[0110]

[0111] Among them, l represents the gap between the predicted value and the true value of each sample; For the model to sample x i The prediction result, f k That is, the prediction function of the kth tree trained is a weighted function of the matching objects including the fusion confidence, Ω is the penalty (regularization) term for the model complexity to prevent overfitting; Ψ is the penalty (regularization) term for the matching objects to prevent too many matching objects from appearing in the recommendation list.

[0112] In an optional embodiment, when adjusting the sorting model to optimize matching objects / non-matching objects, adjusting the second-order gradient can also achieve a similar effect. It should be noted that the present invention is not limited to the algorithm rules provided in the above embodiment, and other algorithm rules can also practice the service push method provided in the embodiment of the present invention.

[0113] Steps S207-S210 are loop steps executed after pushing the matching object to the account that initiates the business migration to be executed for the target object in response to the first selection instruction, with the purpose of making the total amount of the target quota and the matching quota meet the resource allocation limit and meet the user's needs.

[0114] In step S207, a second selection instruction for selecting a matching object for a to-be-performed business migration is received from the user's account.

[0115] When the user selects one or more matching objects from multiple matching objects, a second selection instruction is generated, and the matching object selected by the user can be obtained by receiving the second selection instruction.

[0116] In step S208, according to the second selection instruction, the matching amount of the matching object is obtained.

[0117] Get all matching amounts corresponding to the matching objects.

[0118] In step S209, it is determined whether the total amount of the target amount and the matching amount meets the resource allocation limit.

[0119] The sum of the matching amount corresponding to the matching object selected by the user and the target amount corresponding to the target object is compared with the resource allocation amount to see whether the matching amount corresponding to the matching object and the target amount corresponding to the target object meet the resource allocation amount.

[0120] In step S210, if it is not satisfied, in response to the second selection instruction, the matching object is pushed again to the account that initiates the pending business migration for the target object until the total amount of the target amount and the matching amount meets the resource allocation limit.

[0121] When the matching amount corresponding to the matching object and the target amount corresponding to the target object meet the resource allocation amount, the matching object is stopped from being pushed to the user. When the matching amount corresponding to the matching object and the target amount corresponding to the target object do not meet the resource allocation amount, the matching object is pushed to the user again, and the process is repeated until the matching amount corresponding to the matching object and the target amount corresponding to the target object meet the resource allocation amount. In this way, the push of matching objects to users is completely completed.

[0122] This embodiment provides a method for automatically identifying matching objects in a large number of objects. The accuracy and efficiency of this method are far beyond the reach of traditional methods such as manual operation or rule filtering. According to the identified matching objects, a sorting model is established and the gradient is adjusted to display more matching objects or related objects. This can meet the shopping needs of users and effectively improve the user experience.

[0123] Figure 3 Shows Figure 2 The specific flow chart of step S205 shown in FIG. specifically includes the following steps. Figure 2 In step 205, by calculating the fusion confidence of each target object, matching objects and non-matching objects are identified. This embodiment is a subdivision step of step S205. This embodiment uses at least one data source and model for independently identifying matching objects, calculates the confidence that the object is a matching object, and thus determines whether it is a matching object.

[0124] In step S2051, at least one model is used to independently calculate the classification confidence of each target object. A model is established to calculate the confidence that each target object can become a matching object. Specifically, one or more models can be used to calculate the classification confidence separately, and finally integrated into a total fusion confidence, as shown below.

[0125] Firstly, according to the historical order data, a correlation graph model between target objects is established, and the first classification confidence of the target objects is calculated according to the correlation between the target objects.

[0126] Compared with ordinary objects, matching objects should have correlations with more target objects. Therefore, a graph model of related objects can be constructed through order behavior. In this model, objects with more connected edges are possible matching objects. Accurate correlations between objects are constructed based on historical order data. Compared with traditional frequent item sets, this embodiment proposes a more robust estimation method to offset the interference of hot-selling objects and popular users. Specifically, the correlation scores of all potential related objects are calculated by the following formula:

[0127]

[0128] Among them, buyers a and buyers b Respectively represent the user sets that purchased objects a and b, n i Indicates the total number of objects purchased by user i. (a,b) Defines the degree of correlation between objects a and b in the purchase dimension. When the Score is a positive number, it means that the two objects have a positive correlation, and the larger the value, the higher the degree of correlation, and vice versa.

[0129] Select a certain relevance threshold (select all object combinations ab with relevance scores within the threshold range), take each object as a vertex, and the relevance between objects as an edge, and you can construct a relevance graph between objects. In this graph, matching objects should have high out-degree and in-degree. Because after users purchase multiple objects, they may have the need to complete the order, which creates a correlation between matching objects and multiple objects. However, matching objects such as "mobile phone cases" only have correlations in the mobile phone category, resulting in low out-degree and in-degree, so they are not suitable as matching objects. Using this model, objects with low out-degree and in-degree can be selected as non-matching objects, and objects with high out-degree and in-degree can be selected as matching objects. This can effectively identify matching objects and non-matching objects among objects with large sales, light weight, low price, and high conversion rate.

[0130] The first classification confidence that the object is a matching object is calculated. For example, the first classification confidence that an object a is a matching object can be expressed as:

[0131]

[0132] In the formula, a· and ·a represent the set related to object a and the set related to object a, respectively. Threshold is the predefined correlation score threshold, and σ is the scaling factor, which scales the first classification confidence to between 0 and 1 for subsequent operations.

[0133] Secondly, a conversion rate distribution model of the target object is established, and the second classification confidence of the target object is calculated according to the probability distribution table.

[0134] The user's feedback behavior can also be used as a judgment indicator for matching objects. The user's click rate or purchase rate for a certain object represents the probability that the user will use it as a matching object. Therefore, the user's click rate or conversion rate for a certain object can be directly used to identify matching objects. And the lower limit of the confidence interval of the conversion rate is used as another estimate of the "classification confidence".

[0135] The behavior of purchase conversion follows binomial distribution. Under the normal assumption, the revised estimate of the object conversion rate, namely the Wilson interval, is:

[0136]

[0137] Where n is the total exposure of the object. is the conversion rate of the object in a certain period of time, and z is the quantile of the normal distribution, which can be obtained from the normal distribution table. In actual calculations, the lower limit of the Wilson interval at 95% confidence level can be taken as an estimate of the confidence level of the matching object, that is,

[0138]

[0139] Similarly, the value of the second classification confidence is distributed between 0 and 1, and the closer the value is to 1, the greater the probability that the object will become a matching object.

[0140] Then, a classifier model related to the content attributes of the target object is established, and after optimization, the third classification confidence of the target object is obtained. Whether an object is likely to become a matching object also depends on the content attributes of the object, such as weight, price, purpose, size, etc. A model can be built from the content attributes of the object to predict its probability of being a matching object.

[0141] Classifiers include ensemble trees with regularization terms, logistic regression, and neural networks. Here, we use an ensemble tree model with regularization terms for prediction, that is, we build K independent decision trees, linearly add their prediction results, and get the final prediction result. That is:

[0142]

[0143] in, For the model to sample x i The prediction result, f k That is, the prediction function of the k-th tree number trained, w k is the weight of this tree. Based on this, the objective function of model training can be constructed:

[0144]

[0145] Among them, l is the defined loss function, and Ω is the regularization term to prevent the model from being too complicated and causing overfitting. Assuming that the true distribution of whether an object is a matching object is q (the distribution is obtained by manual editing and labeling), and the predicted distribution of the model is p, both of which obey the binomial distribution, then the loss of the model can be defined by cross entropy, that is: l = H (p, q) = -q·log(p) + (1-q)·log(1-p)

[0146] in, To ensure that the predicted probability p is between 0 and 1. By optimizing the objective function L to minimize it and training the ensemble tree model, we can obtain the third classification confidence of the matching object based on the object content attribute, that is:

[0147]

[0148] Similarly, the confidence level is between 0 and 1, and the closer the value is to 1, the greater the probability that the object i is a matching object.

[0149] This embodiment adopts three models to calculate three classification confidences of the target object respectively, and then integrates them to generate a fusion confidence.

[0150] In step S2052, the classification confidences calculated under at least one model are integrated to generate a fusion confidence corresponding to each target object.

[0151] The above three recognition schemes all identify (predict) whether the object is a matching object from different angles. According to the ensemble learning theory, when the three recognition schemes are all effective (i.e., accuracy>0.5), they are integrated and fused to generate the final prediction probability, which can effectively improve the accuracy.

[0152] In view of the characteristics of order-matching data, the integrated model adopted is:

[0153]

[0154] Among them, Confidence is the fusion confidence that the object after integration is a matching object. i is the confidence given by the i-th model mentioned above. a , confidence w , confidence i ).θ i is a linear addition coefficient, which can be given by empirical value or model learning. γ is used to control when a model considers an object as a matching object with extremely high confidence, directly judging the object as a matching object, that is:

[0155] γ=I(confidence1>T∨confidence2>T∨confidence3>T)

[0156] Where I is an indicative function, which takes the value of 1 when the condition is met, otherwise it takes the value of 0. T is a predefined confidence threshold.

[0157] In step S2053, a confidence threshold is set, and target objects whose fusion confidence is greater than the confidence threshold are screened out as matching objects, that is, matching objects and non-matching objects are identified by the confidence threshold.

[0158] By integrating the three confidences, we get the fusion confidence of whether an object is a matching object. Given a confidence threshold, the values ​​of the confidence threshold are distributed between 0 and 1. Objects with fusion confidence values ​​greater than the confidence threshold are considered matching objects, and objects with fusion confidence values ​​less than the confidence threshold are considered non-matching objects. Different recommendation methods will be adopted for objects that are considered matching objects and non-matching objects according to different shopping scenarios.

[0159] In this embodiment, multiple models are used to extract matching object information from different angles, and the fusion confidence level is judged after fusion to achieve a higher judgment accuracy.

[0160] Figure 4 A structural diagram of a service push system in an embodiment of the present invention is shown.

[0161] The service push system 400 includes a first receiving unit 401 , a first acquiring unit 402 , a first judging unit 403 , a first responding unit 404 and a second corresponding unit 405 .

[0162] The first receiving unit 401 is used to receive a first selection instruction sent by a user account for selecting a target object for a business migration to be performed;

[0163] The first acquisition unit 402 is used to acquire the target amount of the target object according to the first selection instruction;

[0164] The first judgment unit 403 is used to judge whether the target object meets the resource allocation limit according to the target amount.

[0165] The first response unit 404 is used to push a matching object to the account that initiates the to-be-executed business migration for the target object in response to the first selection instruction when the conditions are not met, wherein the matching object has a matching quota so that the total amount of the target quota and the matching quota meets the resource allocation limit;

[0166] The second responding unit 405 is configured to, when satisfied, respond to the first selection instruction and push the associated objects of the target object to the account that initiates the to-be-executed service migration for the target object.

[0167] Preferably, the first response unit 404 is further configured to sort the matching objects using a sorting model, and push the matching objects in sequence according to the sorting order.

[0168] In this embodiment, by calculating the fusion confidence of each selected target object, matching objects and non-matching objects are identified, and then matching objects and non-matching objects are recommended according to customer needs. Whether a target object is a matching object is determined based on its fusion confidence, and the recognition accuracy is high, which can avoid confusion caused by manual recognition. In addition, different recommendation types can be distinguished to solve different shopping needs of users, so that users can quickly find the objects they need, and improve the user's shopping experience.

[0169] Figure 5 The structure diagram of the service push system summarized in the embodiment of the present invention is shown. Figure 5 The embodiment shown in Figure 4 On the basis of the embodiment, a screening unit 501, a second receiving unit 502, a second acquiring unit 503, a second judging unit 504 and a circulating unit 505 are added.

[0170] The screening unit 501 is used to adopt a screening algorithm to screen out matching objects from a plurality of target objects to be selected according to preset screening parameters; the screening parameters include the fusion confidence of the target object.

[0171] The second receiving unit 502 is used to receive a second selection instruction sent by the user's account for selecting a matching object for the to-be-performed business migration;

[0172] The second acquisition unit 503 is used to acquire the matching amount of the matching object according to the second selection instruction;

[0173] The second judgment unit 504 is used to judge whether the total amount of the target amount and the matching amount meets the resource allocation limit.

[0174] The loop unit 505 is used for, when not satisfied, responding to the second selection instruction, again pushing the matching object to the account that initiates the pending business migration for the target object, until the total amount of the target amount and the matching amount meets the resource allocation limit.

[0175] Figure 6 It shows a structural diagram of the screening unit 501 of the service push system in the embodiment of the present invention.

[0176] like Figure 6 As shown, the identification unit 501 includes a calculation unit 5011 , an integration unit 5012 and a classification unit 5013 .

[0177] The calculation unit 5011 is used to independently calculate the classification confidence of each target object using at least one model;

[0178] The integration unit 5012 is used to integrate the classification confidences calculated under at least one model to generate a fusion confidence corresponding to each target object; and

[0179] The classification unit 5013 is used to set a confidence threshold and filter out target objects whose fusion confidence is greater than the confidence threshold as matching objects.

[0180] It should be understood that the system and method of the embodiments of the present invention correspond to each other, and therefore, the description of the system is relatively brief.

[0181] Figure 7 A structural diagram of a service push device according to an embodiment of the present invention is shown. Figure 7 The device shown is only an example and should not constitute any limitation to the functions and scope of use of the embodiments of the present invention.

[0182] refer to Figure 7 The service push device includes a processor 701, a memory 702 and an input / output device 703 connected via a bus. The memory 702 includes a read-only memory (ROM) and a random access memory (RAM). The memory 702 stores various computer instructions and data required to execute system functions. The processor 701 reads various computer instructions from the memory 702 to execute various appropriate actions and processes. The input / output device includes an input part such as a keyboard and a mouse; an output part such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage part such as a hard disk; and a communication part such as a network interface card such as a LAN card and a modem. The memory 702 also stores the following computer instructions to complete the operations specified in the business push method of an embodiment of the present invention: receiving a first selection instruction sent by a user's account for selecting a target object for business migration to be executed; obtaining a target amount of the target object according to the first selection instruction; judging whether the target object meets the resource allocation limit according to the target amount; if not, in response to the first selection instruction, pushing a matching object to the account that initiated the business migration to be executed for the target object, wherein the matching object has a matching amount so that the total amount of the target amount and the matching amount meets the resource allocation limit; if satisfied, in response to the first selection instruction, pushing an associated object of the target object to the account that initiated the business migration to be executed for the target object.

[0183] Accordingly, an embodiment of the present invention provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed, the operations specified by the above-mentioned service push method are implemented.

[0184] The flowcharts and block diagrams in the accompanying drawings illustrate the possible system frameworks, functions and operations of the systems, methods and devices of the embodiments of the present invention. The boxes on the flowcharts and block diagrams can represent a module, a program segment or just a piece of code, and the modules, program segments and codes are all executable instructions for implementing the specified logical functions. It should also be noted that the executable instructions for implementing the specified logical functions can be recombined to generate new modules and program segments. Therefore, the boxes and the order of the boxes in the accompanying drawings are only used to better illustrate the process and steps of the embodiments, and should not be used as a limitation on the invention itself.

[0185] Each module or unit of the system can be implemented by hardware, firmware or software. Software, for example, includes a coding program formed by various programming languages ​​such as JAVA, C / C++ / C#, SQL, etc. Although the steps and the order of the steps of the embodiment of the present invention are given in the method and the method legend, the executable instructions for the steps to implement the specified logical functions can be recombined to generate new steps. The order of the steps should not be limited to the order of the steps in the method and the method legend, and can be adjusted at any time according to the needs of the function. For example, some of the steps are executed in parallel or in reverse order.

[0186] The system and method according to the present invention can be deployed on a single or multiple servers. For example, different modules can be deployed on different servers to form dedicated servers. Alternatively, the same functional units, modules or systems can be deployed in a distributed manner on multiple servers to reduce load pressure. The servers include but are not limited to multiple PCs, PC servers, blade computers, supercomputers, etc. in the same local area network and connected via the Internet.

[0187] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A service push method, characterized in that: include: Receiving a first selection instruction sent by a user account for selecting a target object for business migration to be performed; According to the first selection instruction, obtaining the target amount of the target object; Determine whether the target object meets the resource allocation limit according to the target amount, If not, a screening algorithm is used to screen out a matching object from a plurality of target objects to be selected according to the fusion confidence, and in response to the first selection instruction, a matching object is pushed to the account that initiates the to-be-executed business migration for the target object, wherein the matching object has a matching amount, so that the total amount of the target amount and the matching amount meets the resource allocation limit; If satisfied, in response to the first selection instruction, pushing the associated object of the target object to the account that initiates the to-be-executed business migration for the target object, Among them, using a screening algorithm to screen out the matching object from multiple target objects to be selected according to the fusion confidence includes: using multiple models to independently calculate multiple classification confidences of each target object, and then integrating them to generate a fusion confidence; screening out the target object whose fusion confidence is greater than a set confidence threshold as the matching object, Using multiple models to independently calculate multiple classification confidences for each target object includes: establishing a correlation graph model between target objects to calculate the first classification confidence based on the correlation; establishing a conversion rate distribution model to calculate the second classification confidence based on the probability distribution table; and establishing a classifier model related to content attributes to calculate the third classification confidence.

2. The service push method according to claim 1, characterized in that: Pushing the matching object to the account that initiates the to-be-executed business migration for the target object includes: The matching objects are sorted using a sorting model, and the matching objects are pushed sequentially in the sorting order.

3. The service push method according to claim 1, characterized in that: After pushing the matching object to the account that initiates the to-be-executed business migration for the target object, the method further includes: Receiving a second selection instruction sent by a user account for selecting the matching object for which business migration is to be performed; According to the second selection instruction, obtaining the matching amount of the matching object; Determine whether the total amount of the target amount and the matching amount meets the resource allocation limit, If not, in response to the second selection instruction, the matching object is pushed again to the account that initiates the to-be-executed business migration for the target object until the total amount of the target amount and the matching amount meets the resource allocation limit.

4. The service push method according to claim 1, characterized in that: Classifiers include ensemble trees with regularization, logistic regression, and neural networks.

5. The service push method according to claim 1, characterized in that: The values ​​of the classification confidence, the fusion confidence and the confidence threshold are all distributed between 0 and 1.

6. The service push method according to claim 1, characterized in that: The associated object includes the target object to be selected after removing the matching object.

7. A service push system, characterized in that: include: A first receiving unit, configured to receive a first selection instruction sent by a user's account for selecting a target object for a business migration to be performed; A first acquisition unit, configured to acquire a target amount of the target object according to the first selection instruction; A first judgment unit, configured to judge whether the target object meets the resource allocation limit according to the target amount; A screening unit, which uses a screening algorithm to screen out matching objects from a plurality of target objects to be selected according to fusion confidence; a first response unit, configured to, when the resource allocation limit is not met, push a matching object to the account that initiates the to-be-performed business migration for the target object in response to the first selection instruction, wherein the matching object has a matching amount so that a total amount of the target amount and the matching amount meets the resource allocation limit; a second response unit, configured to, when the resource allocation limit is met, push the associated object of the target object to the account that initiates the to-be-performed service migration for the target object in response to the first selection instruction; The screening unit is used to independently calculate multiple classification confidences of each target object using multiple models, and then integrate them to generate a fusion confidence; the target object whose fusion confidence is greater than a set confidence threshold is screened out as the matching object, The screening unit is also used to establish a correlation graph model between target objects to calculate a first classification confidence based on the correlation; establish a conversion rate distribution model to calculate a second classification confidence based on a probability distribution table; and establish a classifier model related to content attributes to calculate a third classification confidence.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed, the service push method according to any one of claims 1 to 6 is implemented.

9. A service push device, characterized in that: include: Memory, for storing computer instructions; A processor is coupled to the memory, and the processor is configured to execute and implement the service pushing method according to any one of claims 1 to 6 based on computer instructions stored in the memory.

Images