Information prediction method and device, equipment and storage medium

Abstract

The embodiment of the invention discloses an information prediction method and device, equipment and a storage medium. The method comprises the following steps: obtaining returned goods in a warehouse; target article attributes and a target classification network model corresponding to the current iteration period are obtained, and the target article attributes are article attributes determined from to-be-selected article attributes based on to-be-selected article attribute information and actual damage levels corresponding to sample articles in the current iteration period; the target classification network model is obtained by training based on target article attribute information corresponding to the sample article and an actual damage level; and based on the target classification network model and the target article attribute information corresponding to the returned article, carrying out classification prediction on the damage degree of the returned article, and determining a target damage level corresponding to the returned article. Through the technical scheme of the embodiment of the invention, automatic prediction and accurate prediction of the damage grade of the returned goods can be realized, manual participation is not needed, and the efficiency and accuracy of damage grade prediction are improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to computer technology, and in particular to an information prediction method, apparatus, device and storage medium. Background Art

[0002] With the rapid development of the e-commerce industry, more and more users like to buy items on e-commerce platforms, and if users are not satisfied with the purchased items, they can return them, thereby improving the user's purchasing experience. Reverse logistics serves more e-commerce platforms to handle items returned by users.

[0003] At present, e-commerce platforms usually manually judge the damage of returned items and make subsequent processing based on the manual judgment results. E-commerce platforms can also directly return the returned items to the supplier of the items for processing.

[0004] However, in the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art:

[0005] The existing method of judging the damage of returned items based on manual experience is time-consuming and labor-intensive, and cannot effectively guarantee the accuracy of damage judgment. Summary of the invention

[0006] The embodiments of the present invention provide an information prediction method, device, equipment and storage medium to realize automatic and accurate prediction of the damage grade of returned items without human intervention, thereby improving the efficiency and accuracy of damage grade prediction.

[0007] In a first aspect, an embodiment of the present invention provides an information prediction method, comprising:

[0008] Get returned items in the warehouse;

[0009] Obtaining the target item attribute and the target classification network model corresponding to the current iteration cycle, wherein the target item attribute is an item attribute determined from the attributes of the items to be selected based on the attribute information and the actual damage level of the items to be selected corresponding to the sample items in the current iteration cycle; and the target classification network model is obtained by training based on the target item attribute information and the actual damage level corresponding to the sample items;

[0010] Based on the target classification network model and the target item attribute information corresponding to the returned item, a classification prediction of the damage degree of the returned item is performed to obtain a target damage level corresponding to the returned item.

[0011] In a second aspect, an embodiment of the present invention further provides an information prediction device, including:

[0012] The return item acquisition module is used to obtain the return items in the warehouse;

[0013] A model acquisition module, used to acquire the target item attributes and target classification network model corresponding to the current iteration cycle, wherein the target item attributes are item attributes determined from the attributes of the items to be selected based on the attribute information and actual damage levels of the items to be selected corresponding to the sample items in the current iteration cycle; the target classification network model is obtained by training based on the target item attribute information and actual damage levels corresponding to the sample items;

[0014] The information prediction module is used to classify and predict the degree of damage of the returned items based on the target classification network model and the target item attribute information corresponding to the returned items, so as to obtain the target damage level corresponding to the returned items.

[0015] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0016] one or more processors;

[0017] A memory for storing one or more programs;

[0018] When the one or more programs are executed by the one or more processors, the one or more processors implement the information prediction method provided by any embodiment of the present invention.

[0019] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an information prediction method as provided in any embodiment of the present invention.

[0020] One embodiment of the above invention has the following advantages or beneficial effects:

[0021] By determining the target item attributes that can affect the damage degree of the items from the attributes of the selected items based on the attribute information of the selected items and the actual damage level corresponding to the sample items in the current iteration cycle in advance, and training based on the target item attribute information and the actual damage level corresponding to the sample items, a target classification network model that can accurately predict information in the current iteration cycle is obtained. Based on the target classification network model and the target item attribute information of the returned items in the warehouse, the damage degree of the returned items is classified and predicted, and the accurate target damage level can be automatically obtained. Therefore, the automatic and accurate prediction of the damage level of the returned items can be achieved by using the periodically iterated target item attributes and target classification network model, without the need for human intervention, thereby improving the efficiency and accuracy of damage level prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 is a flow chart of an information prediction method provided by an embodiment of the present invention;

[0024] Figure 2 is a flow chart of another information prediction method provided by an embodiment of the present invention;

[0025] Figure 3 This is an example of a network architecture of a composite decision tree involved in an embodiment of the present invention;

[0026] Figure 4 is a flow chart of another information prediction method provided by an embodiment of the present invention;

[0027] Figure 5 is a schematic diagram of the structure of an information prediction device provided by an embodiment of the present invention;

[0028] Figure 6 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0030] Figure 1 This is a flow chart of an information prediction method provided by an embodiment of the present invention. This embodiment can be applied to the case of predicting the damage level of returned items in a warehouse. The method can be executed by an information prediction device, which can be implemented by software and / or hardware and integrated into an electronic device. Figure 1 As shown, the method specifically comprises the following steps:

[0031] S110. Obtain returned items in the warehouse.

[0032] Among them, the returned items can refer to any items that are successfully returned by the user after being purchased from the e-commerce platform. For example, when a user wants to return the purchased items, he can apply for a return on the e-commerce platform. The e-commerce platform can review the return description information applied for. If the return conditions are met, it is determined that the review is passed and the user is allowed to return the goods. The logistics will mail the returned items to the reverse warehouse of the e-commerce platform. The reverse warehouse will unpack and inspect the returned items received to determine whether the returned items meet the preset return conditions, such as whether the returned items are consistent with the user's description, whether there are missing parts or other damages, etc. If the returned items meet the preset return conditions, the returned items are allowed to be received and stored, and the user will be refunded, thereby realizing a closed loop process for item returns.

[0033] Specifically, the returned items in the warehouse whose damage degree needs to be predicted can be obtained periodically, or, based on the item return operation, the returned items stored in the warehouse can be obtained in real time.

[0034] S120. Obtain the target item attributes and target classification network model corresponding to the current iteration cycle, wherein the target item attributes are item attributes determined from the attributes of the candidate items based on the attribute information and actual damage levels of the candidate items corresponding to the sample items in the current iteration cycle; and the target classification network model is obtained by training based on the target item attribute information and actual damage levels corresponding to the sample items.

[0035] Among them, the iteration cycle may refer to the selection and update cycle of item attributes and the training and update cycle of the network model. For example, if the item is not currently in the promotion stage, the item attributes and network model may be iterated in a cycle of 20 to 30 days. If the item is currently in the promotion stage, the item attributes and network model may be iterated in a cycle of 5 to 10 days. The current iteration cycle may refer to the iteration cycle in which the current time is located, that is, the most recent iteration cycle. It should be noted that due to the constant changes in the surrounding environment, such as item prices, consumables prices, and labor costs will change over time, etc., resulting in changes in the degree of influence of each item attribute on item damage in different cycles. Therefore, it is necessary to periodically iterate and update the selection of item attributes and the training of network models in order to ensure the accuracy of information prediction.

[0036] Among them, the target item attribute may refer to the item attribute that has a greater impact on the damage of the item in the current iteration cycle, that is, there is a correlation between the target item attribute and the degree of damage to the item. The attribute of the selected item may refer to the attribute of the item that may be related to the degree of damage to the item. For example, the attributes of the selected item may include but are not limited to: the degree of damage to the packaging, the degree of damage to the appearance, the completeness of the accessories, the remaining degree of the warranty period, the remaining degree of the shelf life, the functional completeness, the degree of safety, the category of consumables, the storage conditions, and whether it is a luxury item. The attribute information of the selected item may refer to the specific attribute value of the attribute of the selected item. Since the attribute of the selected item is a classification parameter, it needs to be digitized so that it can be directly input into the model. For example, Table 1 gives the information assignment results of the attributes of the selected item:

[0037] Table 1 Information assignment results of attributes of selected items

[0038]

[0039] Exemplarily, based on the category of the sample item, the item attributes that may affect the degree of damage of the item can be screened out from all the item attributes of the sample item, and the item attributes that will definitely not affect the damage can be removed. For example, if the sample item is food, there will be no attribute of warranty period, so the attribute of the remaining warranty period can be directly removed. In other words, all the item attributes except the remaining warranty period can be used as the item attributes to be selected, thereby further improving the efficiency of attribute selection.

[0040] The sample items in the current iteration cycle may refer to historical returned items used for training in the current iteration cycle. For example, the actual returned items in the previous iteration cycle may be used as sample items in the current iteration cycle for iterative update. The sample items in the current iteration cycle and the returned items obtained in step S110 may belong to the same category to further ensure the accuracy of information prediction. The actual damage level may be the actual damage level of the sample items calibrated in advance.

[0041] Among them, the target classification network model may refer to the classification network model obtained by training in the current iteration cycle. The target classification network model may be a multi-classification network model for predicting the damage level of returned items. For example, random forests may be used for prediction and classification, that is, the target classification network model is a target random forest model. The damage level of returned items may be divided to obtain multiple damage levels. For example, the damage level may be divided into four damage levels, namely, the first damage level, the second damage level, the third damage level, and the fourth damage level. Among them, the first damage level may refer to a level with a damage level greater than 45%, that is, a scrap level. The second damage level may refer to a level with a damage level less than or equal to 45% and greater than 25%, that is, a recycled level. The third damage level may refer to a level with a damage level less than or equal to 25% and greater than 10%, that is, a damaged level. The fourth damage level may refer to a level with a damage level less than or equal to 10%, that is, a defective level.

[0042] Specifically, the target item attributes in each iteration cycle can be determined from the attributes of the selected items based on the attribute information of the selected items and the actual damage level corresponding to the sample items in each iteration cycle in advance, and training can be performed based on the target item attribute information and the actual damage level corresponding to the sample items to obtain the target classification network model in each iteration cycle. In actual application, the target item attributes and target classification network model in the current iteration cycle that are predetermined can be directly obtained so as to use the latest target item attributes and target classification network model for information prediction, which can effectively ensure the accuracy of information prediction.

[0043] S130: Based on the target classification network model and the target item attribute information corresponding to the returned items, classify and predict the damage degree of the returned items to obtain the target damage level corresponding to the returned items.

[0044] The target item attribute information corresponding to the returned item may refer to the specific attribute value of the target item attribute of the returned item. The degree of damage may refer to the ratio between the repair cost of the returned item and the value attribute value of the returned item. The repair cost may refer to the cost of refurbishing and repairing the returned item. The value attribute value of the returned item may refer to the original price of the returned item. The greater the degree of damage, the higher the damage level.

[0045] Specifically, the target item attribute information corresponding to the returned item can be obtained. At this time, the target item attribute information obtained is the attribute information after the assignment, so the target item attribute information corresponding to the returned item can be directly input into the target classification network model for classification prediction of the damage degree. The target classification network model can output the predicted target damage level, so that the target classification network model with periodic iteration can automatically and accurately predict the damage level of the returned item, thereby improving the accuracy and efficiency of information prediction.

[0046] The technical solution of this embodiment determines the target item attributes that can affect the damage degree of the items from the attributes of the items to be selected based on the attribute information of the items to be selected and the actual damage level corresponding to the sample items in the current iteration cycle in advance, and performs training based on the target item attribute information and the actual damage level corresponding to the sample items, so as to obtain a target classification network model that can accurately predict information in the current iteration cycle. Based on the target classification network model and the target item attribute information of the returned items in the warehouse, the damage degree of the returned items is classified and predicted, and the accurate target damage level can be automatically obtained, so that the automatic prediction and accurate prediction of the damage level of the returned items can be realized by using the periodically iterated target item attributes and target classification network model, without the need for manual participation, thereby improving the efficiency and accuracy of the damage level prediction.

[0047] Figure 2 A flowchart of another information prediction method provided by an embodiment of the present invention. Based on the above embodiments, this embodiment describes in detail the process of determining the attributes of the target item in the current iteration cycle, and also describes in detail the training process when the target classification network model is a target random forest model. The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.

[0048] See also Figure 2 Another information prediction method provided in this embodiment specifically includes the following steps:

[0049] S210: Determine a target connection function corresponding to the current iteration cycle from a plurality of candidate connection functions.

[0050] Among them, the candidate link function may refer to a link function that can be selected. The target link function may refer to the most suitable link function in the current iteration cycle. The link function may refer to a function used in logistic regression to convert continuous regression values ​​into discrete category values, so that logistic regression can be fitted for categorical data. For example, the link function may include but is not limited to: logit function, Probit function, supplementary log-log function, negative log-log function and Cauchit function. Different link functions can be used to describe different dependent variable distributions. For example, the logit function is used when the dependent variable is evenly distributed or the number of categories is small; the Probit function is used when the dependent variable is close to a normal distribution; the supplementary log-log function is used when the probability of occurrence of categories with high levels of the dependent variable is high and the number of categories is large; the negative log-log function is used when the probability of occurrence of categories with low levels of the dependent variable is high and the number of categories is large; the Cauchit function is used when there are extreme values ​​in the dependent variable.

[0051] Specifically, based on the usage scope of each candidate connection function and the distribution of actual damage levels corresponding to sample items in the current iteration cycle, the most suitable target connection function in the current iteration cycle can be determined from multiple candidate connection functions.

[0052] Exemplarily, S210 may include: testing the logistic regression model corresponding to each candidate connection function based on a preset hypothesis testing method to determine the hypothesis testing parameter value corresponding to each candidate connection function; determining a candidate connection function from multiple candidate connection functions based on the hypothesis testing parameter value; and determining the target connection function corresponding to the current iteration cycle from the candidate connection functions based on the distribution information of the actual damage levels corresponding to the sample items.

[0053] Specifically, a corresponding logistic regression model can be constructed based on each candidate connection function, and a hypothesis test can be performed on each constructed logistic regression model based on a preset hypothesis test method, such as a Pearson test or a De-viance test method, to obtain the hypothesis test parameter value corresponding to each logistic regression model, that is, the P value, so as to obtain the hypothesis test parameter value corresponding to each candidate connection function. The candidate connection function with a hypothesis test parameter value less than or equal to 0.05 can be determined as a candidate connection function, that is, the candidate connection function with a significant relationship in which the hypothesis is established is determined as a candidate connection function. After determining the candidate connection function, the most matching target connection function can be determined from the candidate connection function based on the scope of use of each candidate connection function and the distribution information of the actual damage level corresponding to the sample items in the current iteration cycle. By testing the logistic regression model corresponding to the candidate connection function, a more matching target connection function can be determined, further improving the accuracy of attribute selection.

[0054] S220, performing a multivariate ordered logistic regression based on the target connection function, the attribute information of the candidate items corresponding to the sample items in the current iteration cycle, and the actual damage level, to determine the regression index information corresponding to each attribute of the candidate items.

[0055] Among them, multivariate ordered logistic regression is a logistic regression method used when the dependent variable (i.e., the damage level) is multi-classified data and there is an order relationship between the numerical values. The regression index information can be an index information used to measure the degree of influence of the attributes of the selected items on the damage of the items. The regression index information can also be used to characterize the importance of the attributes of the selected items. For example, the regression index information can include: at least one of the regression coefficient, the hypothesis test parameter value (i.e., the P value), and the attribute change amplitude index value (i.e., the OR value). Among them, the regression coefficient can be used to characterize the degree of influence of the attributes of the selected items on the damage of the items. The larger the regression coefficient, the greater the degree of influence. The hypothesis test parameter value can be used to characterize the significance of the influence of the attributes of the selected items on the damage of the items. The smaller the hypothesis test parameter value, the more significant the influence. The attribute change amplitude index value refers to the change amplitude of the dependent variable after the independent variable increases by one unit. If the attribute change amplitude index value is greater than 1, it means that the dependent variable increases with the increase of the independent variable, and vice versa, it decreases with the increase of the independent variable.

[0056] Specifically, the multivariate ordered logistic regression model can be expressed as follows:

[0057]

[0058]

[0059]

[0060] Among them, δ1 is a constant term; k is the number of damage levels, for example, k is equal to 4; π1~π k-1 is the probability of 1 to k-1 level; p k-1 is the ratio of the probability of the k-1th type of result appearing to the probability of not appearing; π1+…+π k-1 is the cumulative probability of occurrence of k = 1 to k = k-1; X1 to X m is the attribute information of the selected item; β1 to β m is the regression coefficient; OR is used to measure the magnitude of the increase in the dependent variable when the independent variable changes by one unit.

[0061] Through the above-mentioned multivariate ordered logistic regression model, multivariate ordered logistic regression is performed based on the target connection function, the attribute information of the candidate items corresponding to the sample items in the current iteration cycle, and the actual damage level, so that the attribute information of the candidate items and the actual damage level can be fitted to obtain the regression index information corresponding to each attribute of the candidate items in the final fitted multivariate ordered logistic regression model.

[0062] S230: Determine the target item attributes from the attributes of the items to be selected based on the regression index information.

[0063] Specifically, the regression index information corresponding to each candidate item attribute can be compared with the index information threshold, and the candidate item attribute having a significant relationship with the loss degree can be determined as the target item attribute. For example, based on the hypothesis test parameter value corresponding to each candidate item attribute, the candidate item attribute can be determined from the candidate item attributes, and based on the regression coefficient or attribute change amplitude index value corresponding to each candidate item attribute, the target item attribute can be determined from the candidate item attributes. For example, the candidate item attribute whose hypothesis test parameter value is less than the first threshold (such as 0.05) can be determined as the candidate item attribute, or the candidate item attribute whose hypothesis test parameter value is less than the first threshold (such as 0.05) and greater than the second threshold (such as 0.01) can be determined as the candidate item attribute, and the candidate item attribute whose hypothesis test parameter value is less than the second threshold (such as 0.01) can be directly determined as the target item attribute, so that the candidate item attribute having a significant relationship can be directly determined as the target item attribute based on the hypothesis test parameter value. The candidate item attributes whose regression coefficients or attribute change amplitude index values ​​are greater than a preset threshold are determined as target item attributes, so that the target item attributes that are significantly correlated with item damage can be accurately determined using the regression index information.

[0064] S240. Sample the target item attributes corresponding to the sample item to determine the sample data corresponding to each composite decision tree, wherein the sample data includes: partial item attribute information and actual damage level corresponding to the sample item, and each composite decision tree includes at least two decision trees with a hierarchical structure.

[0065] The random forest model can include multiple composite decision trees. Each composite decision tree has the same network structure, but each composite decision tree is trained with different sample data, so that multiple different composite decision trees can be trained. The composite decision tree is a network model used to perform multi-classification of the degree of damage of items. Usually a decision tree is a binary classification network model, so at least two decision trees need to be connected to achieve multi-classification prediction. For example, Figure 3An example of a network architecture of a composite decision tree is given. The damage degree is divided into four damage levels, namely: the first damage level with a damage degree F greater than 45%, the second damage level with a damage degree F less than or equal to 45% and greater than 25%, the third damage level with a damage degree F less than or equal to 25% and greater than 10%, and the fourth damage level with a damage degree F less than or equal to 10%. The composite decision tree may include three decision trees with a hierarchical structure, namely tree 1, tree 2 and tree 3. Among them, tree 1 is located at the top layer, which is used for binary classification of damage degree F greater than 25% and F less than or equal to 25%; tree 2 and tree 3 are both located at the bottom layer, and tree 2 is used for binary classification of damage degree F greater than 45% and F less than or equal to 45%, thereby realizing the classification of the first damage level and the second damage level. Tree 3 is used for binary classification of damage degree F greater than 10% and F less than or equal to 10%, thereby realizing the classification of the third damage level and the fourth damage level. By using the leaf nodes of tree 1 as the root nodes of tree 2 and tree 3, a combination of multiple decision trees is achieved, and a composite decision tree for multi-classification is constructed.

[0066] Specifically, based on the number of damage levels and the damage degree cutoff value corresponding to each damage level, a composite decision tree with a hierarchical structure is constructed. There are multiple composite decision trees, and the specific number can be determined based on business needs. Based on the number of composite decision trees, random sampling with replacement can be performed on the target item attributes corresponding to the sample items, and the partial item attributes sampled for each composite decision tree are determined, and the partial item attribute information and the actual damage level are used as sample data of the corresponding composite decision tree.

[0067] S250, training each composite decision tree based on the sample data, and constructing a target random forest model based on the trained composite decision tree.

[0068] Specifically, the training of each composite decision tree can be independent of each other. Based on the partial item attribute information and the actual damage level corresponding to each composite decision tree, each decision tree in the composite decision tree is trained to obtain a composite decision tree after training, and multiple composite decision trees after training are combined to obtain the trained target random forest model in the current iteration cycle.

[0069] Exemplarily, "training each compound decision tree based on sample data" in S250 may include: for each compound decision tree, based on the classification result of each decision tree in the compound decision tree, processing the sample data corresponding to the compound decision tree to obtain sub-sample data corresponding to each decision tree in the compound decision tree, the sub-sample data including partial item attribute information and damage grade labels corresponding to the sample items; training each decision tree based on the sub-sample data to obtain a trained compound decision tree.

[0070] Specifically, for each composite decision tree, based on the classification results corresponding to each decision tree in the composite decision tree (i.e., the damage levels that each decision tree can classify), the sample data corresponding to the composite decision tree can be selected and the actual damage level can be adjusted to obtain the damage level labels and corresponding partial item attribute information corresponding to the sample items that each decision tree can predict. For example, see Figure 2 For tree 1, all sample items corresponding to the composite decision tree can be used as sample items of tree 1, and the actual damage level corresponding to each sample item can be adjusted to a matching damage level label. For example, when the actual damage level is the second damage level with a damage degree F greater than 45%, the adjusted damage level label is a damage level with a damage degree F greater than 25%. Sample items corresponding to damage levels with a damage degree F greater than 25% can be used as sample items of tree 2, and sample items corresponding to damage levels with a damage degree F less than or equal to 25% can be used as sample items of tree 3. Since trees 2 and 3 are the bottom-level decision trees, the actual damage level corresponding to the sample items can be directly determined as the damage level label. Similarly, subsample data corresponding to each decision tree in the composite decision tree can be obtained. The training process of each decision tree in the composite decision tree is also the same. For example, the training process of each decision tree can be: generate a node based on sub-sample data, and detect whether the node is a leaf node. If not, divide the attributes for the current node, and divide the current node to generate branches, and determine whether there is sample data in each branch. If there is sample data, continue to generate nodes based on the sample data. If there is no sample data, convert the branch into a leaf node, until all the generated nodes are leaf nodes, thereby constructing a decision tree.

[0071] Among them, the leaf node is the node that finally determines the damage level category. If the data of the current node all belongs to the same damage level, the current node is a leaf node, and its label is the damage level label of the current node data. In order to prevent infinite iteration, the maximum depth of the decision tree can be set. If the constructed decision tree reaches the maximum depth, the current node is a leaf node, and the majority of the labels in the current node are leaf node labels.

[0072] As the data is continuously divided, the purity of the data in the node becomes higher and higher, which can be achieved by selecting the optimal division attribute from the attribute set to divide the data. In this embodiment, the information entropy gain rate can be used as a label to determine the optimal attribute. Among them, information entropy is an indicator used to measure the purity of data. Information entropy is negatively correlated with the purity of the data set. Assume that the current data set is D, and the proportion of the i-th type of data is p i (i=1,2,…|γ|), then the information entropy of the current data set D is:

[0073]

[0074] Information gain is a measure of the improvement in purity of the current data set by using attribute a to divide it. Assuming that attribute a has m values, dividing the data set by this attribute will produce m subsets (D1, D2, ... D m ). Based on the above formula, the information entropy of each subset can be obtained. The information entropy of each subset is weighted according to the amount of subset data, and the information gain of dividing the current data set D by attribute a is obtained, that is:

[0075]

[0076] The greater the information gain, the greater the improvement in data purity by using attribute a to divide the data. The information entropy gain rate can be expressed as:

[0077]

[0078] The denominator of this formula will increase with the increase of the number of subsets m, so the information entropy gain rate can be used to balance the defects of information gain.

[0079] After determining the optimal partitioning attribute of the current node, assuming that the current attribute has m values, the attribute is used to partition the data set D, and m branches (D1, D2, ... D m ). For each branch, if there is no sample data in the current branch, the branch is a leaf node, and its label is the label with the largest number in D; otherwise, the branch is used as a new node to continue to generate a tree. It should be noted that since the random forest method is used for classification prediction, there is no need to pre-prune or post-prune the decision tree, which can also enhance the generalization of classification prediction.

[0080] S260. Obtain returned items from the warehouse.

[0081] S270: Based on the target random forest model and the target item attribute information corresponding to the returned items, classify and predict the damage degree of the returned items to obtain the target damage level corresponding to the returned items.

[0082] Specifically, the target item attribute information corresponding to the returned items can be input into the target random forest model. Each compound decision tree in the target random forest model performs classification prediction of the degree of damage based on the input target item attribute information, obtains the predicted damage grade output by each compound decision tree, and determines the predicted damage grade with the highest number of occurrences among all predicted damage grades as the final target damage grade. The target random forest model can be used to realize automatic and accurate prediction of the damage grade of returned items, thereby improving the efficiency and accuracy of damage grade prediction.

[0083] The technical solution of this embodiment can accurately determine the target item attributes from the attributes of the items to be selected, thereby improving the accuracy of item attribute selection, by performing multivariate ordered logistic regression based on the target connection function, the attribute information of the items to be selected corresponding to the sample items in the current iteration cycle, and the actual damage level, and based on the regression index information corresponding to each attribute of the items to be selected. By constructing a composite decision tree including at least two decision trees with a hierarchical structure, and constructing a target random forest model based on multiple composite decision trees, multi-classification prediction of the target random forest model can be achieved, and the accuracy of classification prediction can be guaranteed.

[0084] Figure 4 This is a flowchart of another information prediction method provided by an embodiment of the present invention. Based on the above embodiments, this embodiment adds a processing operation for transferring the ownership of returned items. The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.

[0085] See also Figure 4 Another information prediction method provided in this embodiment specifically includes the following steps:

[0086] S410. Obtain returned items in the warehouse.

[0087] S420. Obtain the target item attributes and target classification network model corresponding to the current iteration cycle, wherein the target item attributes are item attributes determined from the attributes of the candidate items based on the attribute information of the candidate items and the actual damage levels corresponding to the sample items in the current iteration cycle; and the target classification network model is obtained by training based on the target item attribute information and the actual damage levels corresponding to the sample items.

[0088] S430: Based on the target classification network model and the target item attribute information corresponding to the returned items, classify and predict the damage degree of the returned items to obtain the target damage level corresponding to the returned items.

[0089] S440: Based on the target damage level, the returned items are processed for ownership transfer, and an ownership transfer task corresponding to the returned items is generated.

[0090] The ownership transfer process may refer to the e-commerce platform transferring the ownership of the returned item to another party other than the supplier of the item. For example, the ownership transfer process may refer to the resale process of the returned item. The ownership transfer task may refer to the task of transferring the ownership of the returned item to another party other than the supplier of the item. For example, the ownership transfer task may refer to the order after resale, that is, the purchase order of the returned item.

[0091] Specifically, the target damage level corresponding to the returned items can be displayed on the e-commerce platform to encourage users who are interested in the returned items to purchase them, thereby generating the corresponding ownership transfer task for the returned items, thus realizing the ownership transfer of the returned items. The returned items can be resold through the e-commerce platform without returning the returned items to the supplier, thereby achieving the goal of reducing losses and generating revenue, and avoiding the risk of secondary damage in the process of returning the returned items to the supplier, further increasing the stickiness of the e-commerce platform to the supplier.

[0092] Exemplarily, S440 may include: determining the target value attribute value corresponding to the returned item based on the target damage level, and displaying the target item information corresponding to the returned item, the target item information including the target value attribute value and the target damage level; in response to the item acquisition operation triggered by the target object for the displayed target item information, generating an ownership transfer task corresponding to the returned item based on the object information of the target object.

[0093] The target value attribute value may refer to the price of the returned item when it is resold. The smaller the target damage level and the greater the damage, the lower the corresponding target value attribute value. The target object may refer to the object that owns the returned item after resale. The target object may be an individual user or an enterprise. The object information of the target object may include the contact information and delivery address of the target object.

[0094] Specifically, the original value attribute value corresponding to the returned item can be processed based on the target damage level to obtain a target value attribute value that is less than the original value attribute value. The target item information including the target value attribute value and the target damage level can be displayed on the second-hand resale platform in the e-commerce platform. If the target object wants to purchase the returned item based on the displayed target item information, the target object can trigger an item acquisition operation, such as clicking a purchase button, so that the ownership transfer task corresponding to the returned item can be generated based on the object information of the target object, thereby realizing the secondary sale of the returned item without returning the returned item to the item supplier, thereby realizing the supplier's loss-stopping and cost-reduction, and increasing the revenue of the e-commerce platform.

[0095] For example, the e-commerce platform can send the target damage level, target value attribute value and the grading video of the returned items to the supplier of the returned items to support the supplier to review and determine whether to allow the e-commerce platform to transfer the ownership. If the review is passed, the ownership of the returned items will be transferred, otherwise the returned items will be returned to the supplier, thereby meeting the personalized needs of the supplier.

[0096] The technical solution of this embodiment, by transferring the ownership of returned items based on the target damage level, generates ownership transfer tasks corresponding to the returned items, thereby enabling secondary sales of returned items without returning the returned items to the supplier of the items, thereby avoiding the risk of secondary damage in the process of returning the returned items to the supplier, and further increasing the stickiness of the e-commerce platform to the supplier.

[0097] Based on the above technical solution, step S440 may include: based on the ownership transfer configuration information of the item supplier corresponding to the returned item, detecting whether the returned item meets the preset ownership transfer conditions; if the returned item meets the preset ownership transfer conditions, then based on the target damage level, performing ownership transfer processing on the returned item, and generating an ownership transfer task corresponding to the returned item.

[0098] The ownership transfer configuration information may be used to represent the information of items that the supplier allows to transfer ownership. For example, the ownership transfer configuration may be performed according to the category dimension or the supplier dimension. The ownership transfer configuration information may include at least one category that the supplier allows to transfer ownership.

[0099] Specifically, the ownership transfer configuration information of each item supplier can be pre-configured on the e-commerce platform. In actual application, the ownership transfer configuration information of the item supplier corresponding to the returned item can be obtained, and based on the ownership transfer configuration information, it is detected whether the returned item meets the preset ownership transfer conditions, such as detecting whether the category of the returned item belongs to the configured category. If so, it indicates that the returned item is allowed to transfer ownership. At this time, the returned item is processed for ownership transfer based on the target damage level. Otherwise, the returned item is returned to the item supplier, thereby realizing the configurability of ownership transfer and meeting the resale needs of different items of the supplier.

[0100] The following is an embodiment of an information prediction device provided by an embodiment of the present invention. The device and the information prediction methods of the above-mentioned embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the information prediction device, reference can be made to the embodiments of the above-mentioned information prediction methods.

[0101] Figure 5 This is a schematic diagram of the structure of an information prediction device provided by an embodiment of the present invention. This embodiment is applicable to the case of predicting the degree of damage of returned items in a warehouse. Figure 5 As shown, the device specifically includes: a returned item acquisition module 510, a model acquisition module 520 and an information prediction module 530.

[0102] Among them, the returned item acquisition module 510 is used to acquire the returned items in the warehouse; the model acquisition module 520 is used to acquire the target item attributes and the target classification network model corresponding to the current iteration cycle, wherein the target item attributes are item attributes determined from the attributes of the selected items based on the attribute information of the selected items and the actual damage level corresponding to the sample items in the current iteration cycle; the target classification network model is obtained by training based on the target item attribute information and the actual damage level corresponding to the sample items; the information prediction module 530 is used to classify and predict the degree of damage of the returned items based on the target classification network model and the target item attribute information corresponding to the returned items, and obtain the target damage level corresponding to the returned items.

[0103] The technical solution of this embodiment determines the target item attributes that can affect the damage degree of the items from the attributes of the items to be selected based on the attribute information of the items to be selected and the actual damage level corresponding to the sample items in the current iteration cycle in advance, and performs training based on the target item attribute information and the actual damage level corresponding to the sample items, so as to obtain a target classification network model that can accurately predict information in the current iteration cycle. Based on the target classification network model and the target item attribute information of the returned items in the warehouse, the damage degree of the returned items is classified and predicted, and the accurate target damage level can be automatically obtained, so that the automatic prediction and accurate prediction of the damage level of the returned items can be realized by using the periodically iterated target item attributes and target classification network model, without the need for manual participation, thereby improving the efficiency and accuracy of the damage level prediction.

[0104] Optionally, the device further comprises:

[0105] A target connection function determination module is used to determine a target connection function corresponding to a current iteration cycle from a plurality of connection functions to be selected;

[0106] A regression index information determination module is used to perform multivariate ordered logistic regression based on the target connection function, the attribute information of the candidate items corresponding to the sample items in the current iteration cycle, and the actual damage level to determine the regression index information corresponding to each candidate item attribute;

[0107] The target item attribute determination module is used to determine the target item attribute from the attributes of the items to be selected based on the regression index information.

[0108] Optionally, the target connection function determination module is specifically used to:

[0109] Based on a preset hypothesis testing method, the logistic regression model corresponding to each candidate connection function is tested to determine the hypothesis testing parameter value corresponding to each candidate connection function; based on the hypothesis testing parameter value, a candidate connection function is determined from multiple candidate connection functions; based on the distribution information of the actual damage levels corresponding to the sample items, the target connection function corresponding to the current iteration cycle is determined from the candidate connection functions.

[0110] Optionally, the regression index information includes: at least one of a regression coefficient, a hypothesis test parameter value, and an attribute change amplitude index value.

[0111] Optionally, the target classification network model is a target random forest model; the device also includes:

[0112] A sample data determination module, used to sample the target item attributes corresponding to the sample items, and determine the sample data corresponding to each composite decision tree, wherein the sample data includes: partial item attribute information and actual damage level corresponding to the sample items, and each composite decision tree includes at least two decision trees with a hierarchical structure;

[0113] The compound decision tree training module is used to train each compound decision tree based on the sample data, and to construct a target random forest model based on the trained compound decision tree.

[0114] Optionally, the composite decision tree training module is specifically used for:

[0115] For each compound decision tree, based on the classification results of each decision tree in the compound decision tree, the sample data corresponding to the compound decision tree is processed to obtain sub-sample data corresponding to each decision tree in the compound decision tree, wherein the sub-sample data includes partial item attribute information and damage grade labels corresponding to the sample items; each decision tree is trained based on the sub-sample data to obtain a trained compound decision tree.

[0116] Optionally, the device further comprises:

[0117] The ownership transfer module is used to, after obtaining the target damage level corresponding to the returned item, perform ownership transfer processing on the returned item based on the target damage level, and generate an ownership transfer task corresponding to the returned item.

[0118] Optionally, the ownership transfer module is specifically used to:

[0119] Based on the target damage level, determine the target value attribute value corresponding to the returned item, and display the target item information corresponding to the returned item, the target item information including the target value attribute value and the target damage level;

[0120] In response to an item acquisition operation triggered by a target object for displayed target item information, an ownership transfer task corresponding to the returned item is generated based on the object information of the target object.

[0121] Optionally, the ownership transfer module is specifically used to:

[0122] Based on the ownership transfer configuration information of the item supplier corresponding to the returned item, detect whether the returned item meets the preset ownership transfer conditions; if the returned item meets the preset ownership transfer conditions, perform ownership transfer processing on the returned item based on the target damage level, and generate an ownership transfer task corresponding to the returned item.

[0123] The information prediction device provided in the embodiment of the present invention can execute the information prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the information prediction method.

[0124] It is worth noting that in the embodiment of the above-mentioned information prediction device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0125] Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 6 A block diagram of an exemplary electronic device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 6 The electronic device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0126] like Figure 6 As shown, the electronic device 12 is in the form of a general purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0127] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0128] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0129] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 6 not shown, usually called a "hard drive"). Although Figure 6 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0130] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0131] The electronic device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the electronic device 12, and / or may communicate with any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the electronic device 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0132] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing the steps of an information prediction method provided by any embodiment of the present invention, the method comprising:

[0133] Acquire returned items in the warehouse; acquire target item attributes and a target classification network model corresponding to the current iteration cycle, wherein the target item attributes are item attributes determined from the attributes of the candidate items based on the candidate item attribute information and the actual damage level corresponding to the sample items in the current iteration cycle; the target classification network model is obtained by training based on the target item attribute information and the actual damage level corresponding to the sample items; based on the target classification network model and the target item attribute information corresponding to the returned items, classify and predict the degree of damage of the returned items to obtain the target damage level corresponding to the returned items.

[0134] Of course, those skilled in the art can understand that the processor can also implement the technical solution of the information prediction method provided by any embodiment of the present invention.

[0135] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the information prediction method provided by any embodiment of the present invention are implemented.

[0136] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0137] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0138] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0139] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0140] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0141] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. An information prediction method, characterized in that: include: Get returned items in the warehouse; Obtaining the target item attribute and the target classification network model corresponding to the current iteration cycle, wherein the target item attribute is an item attribute determined from the attributes of the items to be selected based on the attribute information and the actual damage level of the items to be selected corresponding to the sample items in the current iteration cycle; and the target classification network model is obtained by training based on the target item attribute information and the actual damage level corresponding to the sample items; Based on the target classification network model and the target item attribute information corresponding to the returned item, a classification prediction of the damage degree of the returned item is performed to obtain a target damage level corresponding to the returned item.

2. The method according to claim 1, characterized in that Based on the candidate item attribute information and the actual damage level corresponding to the sample items in the current iteration cycle, the target item attributes are determined from the candidate item attributes, including: Determine a target connection function corresponding to a current iteration cycle from a plurality of candidate connection functions; Perform multivariate ordered logistic regression based on the target connection function, the attribute information of the candidate items corresponding to the sample items in the current iteration cycle, and the actual damage level to determine the regression index information corresponding to each candidate item attribute; Based on the regression index information, the target item attribute is determined from the attributes of the items to be selected.

3. The method according to claim 2, characterized in that Determine the target connection function corresponding to the current iteration cycle from multiple candidate connection functions, including: Testing the logistic regression model corresponding to each candidate connection function based on a preset hypothesis testing method, and determining the hypothesis testing parameter value corresponding to each candidate connection function; Determining a candidate connection function from a plurality of candidate connection functions based on the hypothesis testing parameter value; Based on the distribution information of the actual damage levels corresponding to the sample items, a target connection function corresponding to the current iteration cycle is determined from the candidate connection functions.

4. The method according to claim 2, characterized in that: The regression index information includes at least one of a regression coefficient, a hypothesis test parameter value, and an attribute change range index value.

5. The method according to claim 1, characterized in that The target classification network model is a target random forest model; Training is performed based on the target item attribute information and the actual damage level corresponding to the sample item to obtain a target classification network model, including: Sampling the target item attributes corresponding to the sample items, and determining sample data corresponding to each composite decision tree, wherein the sample data includes: partial item attribute information and actual damage level corresponding to the sample items, and each composite decision tree includes at least two decision trees with a hierarchical structure; Each composite decision tree is trained based on the sample data, and a target random forest model is constructed based on the trained composite decision trees.

6. The method according to claim 5, characterized in that Training each composite decision tree based on the sample data includes: For each compound decision tree, based on the classification results of each decision tree in the compound decision tree, the sample data corresponding to the compound decision tree is processed to obtain sub-sample data corresponding to each decision tree in the compound decision tree, wherein the sub-sample data includes partial item attribute information and damage grade labels corresponding to the sample items; Each decision tree is trained based on the sub-sample data to obtain a trained composite decision tree.

7. The method according to any one of claims 1 to 6, characterized in that: After obtaining the target damage level corresponding to the returned item, the method further includes: Based on the target damage level, the ownership transfer process is performed on the returned items, and an ownership transfer task corresponding to the returned items is generated.

8. The method according to claim 7, characterized in that Based on the target damage level, the returned item is subjected to ownership transfer processing, and an ownership transfer task corresponding to the returned item is generated, including: Based on the target damage level, determine the target value attribute value corresponding to the returned item, and display the target item information corresponding to the returned item, the target item information including the target value attribute value and the target damage level; In response to an item acquisition operation triggered by a target object for displayed target item information, an ownership transfer task corresponding to the returned item is generated based on the object information of the target object.

9. The method according to claim 7, characterized in that: Based on the target damage level, the returned item is subjected to ownership transfer processing, and an ownership transfer task corresponding to the returned item is generated, including: Based on the ownership transfer configuration information of the item supplier corresponding to the returned item, detecting whether the returned item meets the preset ownership transfer conditions; If the returned item meets the preset ownership transfer conditions, the returned item is processed for ownership transfer based on the target damage level, and an ownership transfer task corresponding to the returned item is generated.

10. An information prediction device, characterized in that: include: The return item acquisition module is used to obtain the return items in the warehouse; A model acquisition module, used to acquire the target item attributes and target classification network model corresponding to the current iteration cycle, wherein the target item attributes are item attributes determined from the attributes of the items to be selected based on the attribute information and actual damage levels of the items to be selected corresponding to the sample items in the current iteration cycle; the target classification network model is obtained by training based on the target item attribute information and actual damage levels corresponding to the sample items; The information prediction module is used to classify and predict the degree of damage of the returned items based on the target classification network model and the target item attribute information corresponding to the returned items, so as to obtain the target damage level corresponding to the returned items.

11. An electronic device, characterized in that: The electronic device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the information prediction method as described in any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the information prediction method as described in any one of claims 1 to 9 is implemented.

Images