Sample construction method, article attribute early warning method and device

Abstract

The invention discloses a sample construction method, an article attribute early warning method and a device thereof, and relates to the technical field of computers. A specific embodiment of the method comprises the steps of classifying a first target attribute of an article to obtain a classification target attribute of the article, and further determining a reference target attribute of the article; monitoring each service interface to obtain an attribute feedback message of a second target attribute in the order log of the article, and performing sample marking on the second target attribute according to the attribute feedback message to obtain a target attribute feedback sample containing a positive sample and a negative sample; and forming a target attribute sample of the article by using the reference target attribute and the target attribute feedback sample. According to the embodiment, accurate early warning of the object target attribute is realized, the target attribute sample is constructed by marking the target attribute through the attribute feedback message, the performance of training an object attribute early warning model by using the sample subsequently is ensured, and the accuracy of early warning of the object target attribute is further ensured.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a sample construction method, an object attribute early warning method and a device thereof. Background Art

[0002] In the field of item promotion business, among the various attributes of items, the determination of target attributes (such as price, quantity, etc.) with floating numerical characteristics is directly related to the promotion effect of items and the interests of merchants. In order to take into account both the promotion effect of items and the interests of merchants, item attribute warnings are performed in advance for the target attributes of items before the promotion activities. Existing item attribute warning methods mainly determine the corresponding warning threshold based on the historical target attributes of items to issue warnings.

[0003] In the process of implementing the present invention, the inventors found that the prior art has the following problems:

[0004] The existing method of determining the warning threshold based on historical target attributes has a relatively fixed warning threshold, which cannot match the volatility of the market well and is prone to misjudgment of warnings, resulting in low accuracy of warnings, and thus cannot well protect the interests of the platform and merchants. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a sample construction method, an item attribute warning method and a device thereof, which realize an accurate warning of the target attribute of an item, and use a target attribute sample constructed by sample marking of a second target attribute based on an attribute feedback message, thereby ensuring the performance of an item attribute warning model subsequently trained using the target attribute sample, thereby ensuring the accuracy of the warning of the target attribute of the item.

[0006] To achieve the above object, according to one aspect of an embodiment of the present invention, a method for constructing a sample is provided, comprising:

[0007] Classifying the first target attribute of the item to obtain classified target attributes of the item, and determining the reference target attribute of the item according to each of the classified target attributes;

[0008] Monitor each business interface to obtain an attribute feedback message of a second target attribute in the order log of the item, and perform sample marking on the second target attribute according to the attribute feedback message to obtain a target attribute feedback sample including positive samples and negative samples;

[0009] The reference target attribute and the target attribute feedback sample are combined to form a target attribute sample of the item.

[0010] Optionally, according to each of the classification target attributes, the baseline target attribute of the object is determined, including: determining the priority of each of the classification target attributes, and judging the numerical stability of each of the classification target attributes according to the priority; and obtaining the baseline target attribute of the object according to the first classification target attribute that meets the preset numerical stability standard.

[0011] Optionally, judging the numerical stability of each of the classification target attributes includes: judging the numerical stability index of each of the classification target attributes by calculating at least one of the numerical standard deviation, numerical variance, numerical mean absolute deviation, numerical moving average, numerical moving standard deviation, numerical volatility, and numerical average true amplitude of each of the classification target attributes.

[0012] Optionally, monitoring each business interface to obtain the attribute feedback message of the second target attribute in the order log of the item includes: determining each business interface corresponding to the second target attribute in the order log, the business interface being deployed in an external business system and used to receive the attribute feedback message of the second target attribute uploaded by the user; monitoring each of the business interfaces to obtain each attribute feedback message of the second target attribute; and assigning a corresponding feedback weight to each of the attribute feedback messages according to the pre-configured feedback weights of each of the business interfaces.

[0013] Optionally, according to the attribute feedback message, the second target attribute is sample marked, including: determining whether the second target attribute has a risk according to the attribute feedback message and the feedback weight corresponding to the attribute feedback message; in response to the second target attribute not having a risk and no attribute feedback message indicating that the second target attribute has a risk is heard within a preset approval time, marking the second target attribute as a negative sample representing safety; in response to the second target attribute having a risk, performing a risk verification on the second target attribute, and marking the second target attribute as a negative sample representing safety or a positive sample representing risk according to the risk verification result.

[0014] Optionally, the second target attribute is subjected to risk verification, including: verifying whether the second target attribute has risks based on whether the second target attribute is affected by the superposition of promotional resources; and verifying whether the second target attribute has risks based on the historical second target attributes of the item in response to the second target attribute not being affected by the superposition of promotional resources.

[0015] According to a second aspect of an embodiment of the present invention, a method for training an item attribute warning model is provided, comprising:

[0016] Inputting a target attribute sample into a predefined item attribute warning model; the target attribute sample is constructed by any method provided by the first aspect of the embodiment of the present invention;

[0017] Based on the maximum likelihood method and the target attribute samples, the parameters of the item attribute warning model are adjusted by calculating the positive sample loss function and the negative sample loss function of the item attribute warning model to obtain a trained item attribute warning model with attribute warning capability.

[0018] According to a third aspect of an embodiment of the present invention, there is provided an item attribute warning method, comprising:

[0019] Inputting the target attribute to be determined of the item into an item attribute warning model, wherein the item attribute warning model is trained according to the method provided in the second aspect of the embodiment of the present invention;

[0020] The probability that the target attribute to be determined of the item has a risk output by the item attribute warning model is obtained.

[0021] Optionally, after obtaining the probability that the target attribute to be judged of the item is at risk output by the item attribute warning model, the method further includes: in response to the probability that the target attribute to be judged is at risk being lower than a preset risk threshold, determining that the target attribute to be judged is a safe target attribute, adding the safe target attribute to the order log of the item to obtain an attribute feedback message of the safe target attribute; in response to the probability that the target attribute to be judged is at risk reaching the risk threshold, determining that the target attribute to be judged is a risk target attribute, and adding the risk target attribute as a positive sample to the target attribute sample.

[0022] According to a fourth aspect of an embodiment of the present invention, there is provided a sample construction device, comprising:

[0023] A first sample determination module is used to classify the first target attribute of the item to obtain the classified target attribute of the item, and determine the reference target attribute of the item according to each of the classified target attributes;

[0024] A second sample determination module is used to monitor each business interface to obtain an attribute feedback message of a second target attribute in the order log of the item, and to perform sample marking on the second target attribute according to the attribute feedback message to obtain a target attribute feedback sample including positive samples and negative samples;

[0025] The target attribute sample determination module is used to combine the reference target attribute and the target attribute feedback sample to form a target attribute sample of the item.

[0026] According to a fifth aspect of an embodiment of the present invention, there is provided an item attribute warning device, comprising:

[0027] A data input module, used to input the target attribute of the item to be judged into the item attribute warning model, wherein the item attribute warning model is trained according to the method of claim 7;

[0028] The attribute warning module is used to obtain the probability of the existence of risk in the target attribute to be judged of the item output by the item attribute warning model.

[0029] According to a sixth aspect of an embodiment of the present invention, there is provided an electronic device for warning of item attributes, including:

[0030] one or more processors;

[0031] a storage device for storing one or more programs,

[0032] When the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the third aspect of the embodiment of the present invention.

[0033] According to a seventh aspect of an embodiment of the present invention, there is provided a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method provided in the third aspect of the embodiment of the present invention is implemented.

[0034] According to an eighth aspect of an embodiment of the present invention, there is provided a computer program product, including a computer program, wherein when the computer program is executed by a processor, the method provided in the third aspect of the embodiment of the present invention is implemented.

[0035] An embodiment of the invention has the following advantages or beneficial effects: by classifying the first target attribute of the item, the classified target attribute of the item is obtained, and according to each classified target attribute, the baseline target attribute of the item is determined; each business interface is monitored to obtain the attribute feedback message of the second target attribute in the order log of the item, and according to the attribute feedback message, the second target attribute is sampled and labeled to obtain the target attribute feedback sample containing positive samples and negative samples; the technical solution of combining the baseline target attribute and the target attribute feedback sample into the target attribute sample of the item realizes the construction of a complete, accurate and dynamic target attribute sample, and the target attribute sample constructed by sample labeling the second target attribute based on the attribute feedback message ensures the performance of the item attribute warning model subsequently trained using the target attribute sample, thereby ensuring the accuracy of the warning of the item target attribute. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0037] Figure 1 is a schematic diagram of the main process of the method for constructing a sample according to an embodiment of the present invention;

[0038] Figure 2 is a flow chart of a method for constructing a sample according to a reference embodiment of the present invention;

[0039] Figure 3 is a schematic diagram of a method for determining a reference target attribute according to an embodiment of the present invention;

[0040] Figure 4 is a flow chart of a method for constructing a sample according to another reference embodiment of the present invention;

[0041] Figure 5 is a schematic diagram of constructing a target attribute feedback sample according to an embodiment of the present invention;

[0042] Figure 6 is a schematic diagram of an item attribute early warning method according to an embodiment of the present invention;

[0043] Figure 7 is a schematic diagram of main modules of a device for constructing a sample according to an embodiment of the present invention;

[0044] Figure 8 2 is a schematic diagram of main modules of an object attribute early warning device according to an embodiment of the present invention;

[0045] Fig. 9 is an exemplary system architecture diagram to which embodiments of the present invention may be applied;

[0046] Fig.10 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0047] It should be noted that in the technical solution of the present invention, the collection / collection, update, analysis, use, transmission, storage and other aspects of user personal information involved are in compliance with the provisions of relevant laws and regulations, are used for legal and reasonable purposes, are not shared, leaked or sold outside of these legal uses, and are subject to supervision and management by national regulatory authorities. Necessary measures should be taken for user personal information to selectively block the use or access to personal information data to prevent illegal access to such personal information data, ensure that persons who have access to personal information data comply with the provisions of relevant laws and regulations, and ensure the security of user personal information. In addition, once these user personal information data are no longer needed, the risks should be minimized by limiting or even prohibiting data collection and / or deleting data.

[0048] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0049] The existing method of determining the warning threshold based on historical target attributes has a relatively fixed warning threshold, which cannot match the volatility of the market well and is prone to misjudgment, resulting in low accuracy of warnings, and thus cannot well protect the interests of the platform and merchants.

[0050] In order to solve the above problems existing in the prior art, the present invention proposes a sample construction method and an item attribute warning method, which uses a machine learning model to perform item attribute warning, mainly from the aspect of constructing the target attribute samples required for model training, using attribute feedback messages to mark the positive and negative samples of the second target attribute in the order log, and then combining the benchmark target attributes that characterize the target attribute baseline determined according to multiple classification target attributes to construct a complete, accurate and dynamic target attribute sample, thereby ensuring the performance of the item attribute warning model subsequently trained using the target attribute sample, and further ensuring the accuracy of the warning of the item target attribute using the trained item attribute warning model.

[0051] Figure 1 is a schematic diagram of the main process of the method for constructing a sample according to an embodiment of the present invention, such as Figure 1 As shown, the sample construction method of the embodiment of the present invention includes the following steps S101 to S103.

[0052] Step S101: classify the first target attribute of the object to obtain the classified target attribute of the object, and determine the reference target attribute of the object according to each of the classified target attributes.

[0053] Specifically, when promoting items, it is necessary to determine the promotion details of the items in advance, especially the target attributes with floating changes among the various attributes of the items, such as item price, item quantity, etc. Given that other promotions are inevitably superimposed during the promotion business activities, such as the superposition of platform full discount activities and store full discount activities, brand discount activities and platform full discount activities, etc., early warning of the pre-determined target attributes of items is a necessary measure to prevent merchants from suffering losses due to unreasonable target attributes of items.

[0054] In the early stages of the entire life cycle of the promotion business, due to the lack of risk feedback, it is necessary to collect the first target attributes of the items stored in the background system within a specified historical time period and classify the first target attributes. For example, for the historical prices of the items stored in the background system, the corresponding categories are purchase price, transaction price, official price, etc. Different classification target attributes have different degrees of influence on the interests of merchants. From each classification target attribute, a classification target attribute that can be used as a benchmark for the target attribute is selected, and the selected classification target attribute is used as the benchmark target attribute of the item. The benchmark target attribute of the embodiment of the present invention is equivalent to the baseline of the target attribute, which provides a reasonable basis for determining the target attributes of similar items.

[0055] Step S102: monitor each business interface to obtain the attribute feedback message of the second target attribute in the order log of the item, and perform sample marking on the second target attribute according to the attribute feedback message to obtain a target attribute feedback sample including positive samples and negative samples.

[0056] Specifically, the historical target attribute of an item facing consumers on a trading platform for business transactions is defined as a second target attribute in an embodiment of the present invention, such as the sales price of an item, which is usually stored and recorded in an order log. In view of the volatility of the second target attribute of an item and the high sensitivity of various related business interfaces to the volatility of the second target attribute, the embodiment of the present invention abandons the traditional method of marking the second target attribute by dedicated labeling personnel, and utilizes the feedback of various business interfaces related to the second target attribute on the second target attribute to mark the second target attribute, thereby dynamically and more accurately constructing target attribute feedback samples.

[0057] Furthermore, based on the above concept, the sample construction system monitors each business interface, obtains the attribute feedback message about the second target attribute received by each business interface, parses the attribute feedback message according to the preset message structure, and obtains the information whether the second target attribute has risk. If the second target attribute is analyzed to be risky, then the second target attribute is marked as a risk positive sample; if the second target attribute is not risky, then the second target attribute is marked as a risk negative sample, and finally the target attribute feedback sample with positive sample mark and negative sample mark composed of each second target attribute in the order log of the item is obtained.

[0058] It can be understood that the embodiment of the present invention can also be simplified to obtain the attribute feedback message only when the second target attribute is risky. Accordingly, when the sample construction system obtains the attribute feedback message, the second target attribute is marked as a risk positive sample. If the attribute feedback message is not obtained, it indicates that the second target attribute is not risky, then the second target attribute is marked as a risk negative sample.

[0059] Step S103: Combining the reference target attribute and the target attribute feedback sample into a target attribute sample of the item.

[0060] Specifically, the benchmark target attributes of the object determined from the multiple classification target attributes and the target attribute feedback samples constructed by marking the second target attributes using the received attribute feedback message constitute the target attribute samples for model training.

[0061] The embodiment of the present invention utilizes the sensitivity of each business interface related to the target attribute to the target attribute to mark the second target attribute to obtain a more accurate and real-time attribute feedback message, thereby ensuring the accuracy of sample marking, constructing high-quality target attribute feedback samples, and then combining the benchmark target attributes determined based on multiple classification target attributes to ensure the completeness, accuracy and real-time dynamics of the target attribute samples, so as to improve the performance of subsequent training models.

[0062] Figure 2 is a flow chart of a method for constructing a sample according to a reference embodiment of the present invention. As another embodiment of the present invention, Figure 2 As shown, the sample construction method may include:

[0063] Step S201: classify the first target attribute of the object to obtain the classified target attribute of the object.

[0064] Step S202, determine the priority of each of the classification target attributes, and judge the numerical stability of each of the classification target attributes according to the priority; obtain the reference target attribute of the object according to the first classification target attribute that meets the preset numerical stability standard.

[0065] Specifically, although different classification target attributes have different degrees of influence on the interests of merchants, one classification target attribute that can be used as a target attribute benchmark can be selected from each classification target attribute. However, preferably, it is also necessary to screen each classification target attribute from the perspective of numerical stability, and it is more reasonable to select a classification target attribute with stable values. Therefore, the embodiment of the present invention determines the priority of each classification target attribute based on the difference in the degree of influence of the classification target attribute on the interests of merchants. Then, according to the order of priority, the numerical stability of each classification target attribute is judged in turn, and the numerical value corresponding to the first classification target attribute that meets the numerical stability standard requirements in the judgment process is used as the numerical value of the benchmark target attribute of the item.

[0066] For example, taking the first target attribute of an item as the price of the item stored in the background system, Figure 3Schematic diagram of a method for determining the benchmark target attribute of an embodiment of the present invention. The classification target attributes of an item include official price, platform purchase price, platform sales price, and average transaction price, and the priority of each classification target attribute is defined from high to low as official price, platform purchase price, platform sales price, and platform transaction price. It can be understood that, given that the official price is a fixed price, in order to improve the efficiency of determining the benchmark target attribute, it is only necessary to determine whether there is an official mark, and there is no need to judge the numerical stability.

[0067] The specific process of determining the benchmark target attribute is as follows: First, determine whether there is an official price in the historical price of the item obtained from the specified historical period. If so, the official price is used as the value of the benchmark target attribute of the item. If not, determine whether the platform purchase price of the item is stable within a specified period of time. If stable, the stable purchase price is used as the value of the benchmark target attribute of the item. If not, determine the next priority, that is, whether the platform sales price of the item is stable within a specified period of time. If stable, the stable platform sales price is used as the value of the benchmark target attribute of the item. If not, determine the next priority, whether the average transaction price of the item is stable within a specified period of time. If stable, the stable average transaction price is used as the value of the benchmark target attribute of the item. If it is still unstable, the preset fallback logic is used, and the guide price specified by the business module is used as the value of the benchmark target attribute of the item.

[0068] According to one embodiment of the present invention, judging the numerical stability of each of the classification target attributes includes: judging the numerical stability index of each of the classification target attributes by calculating at least one of the numerical standard deviation, numerical variance, numerical mean absolute deviation, numerical moving average, numerical moving standard deviation, numerical volatility, and numerical average true amplitude of each of the classification target attributes.

[0069] Specifically, the embodiments of the present invention provide a variety of methods for determining numerical stability. For each classification target attribute, the numerical standard deviation between the numerical values ​​can be calculated. The standard deviation is used to measure the degree of deviation between the numerical value and the mean. The larger the standard deviation, the greater the volatility. Specifically, the numerical standard deviation can be calculated using the following formula: , in, is the standard deviation, N is the number of data points, is the i-th value in the categorical target attribute, and μ is the mean.

[0070] The variance between the values ​​in each categorical target attribute can also be calculated. The variance is the square of the standard deviation of the values ​​mentioned above, which can also be used to measure stability. It is calculated using the following formula: .

[0071] You can also calculate the numerical mean absolute deviation between the values ​​in each categorical target attribute, which is the average of the absolute deviations of the values ​​from the mean and is an alternative to the numerical standard deviation: .

[0072] You can also calculate the numerical moving average between the values ​​in each categorical target attribute. The numerical moving average is used to smooth time series data and reduce short-term fluctuations, so as to better observe the long-term trend of the values ​​in each categorical target attribute: , Where k is the size of the moving window and t is the current time point.

[0073] You can also calculate the moving standard deviation of the values ​​in each categorical target attribute. The moving standard deviation is a moving window version of the standard deviation and is used to analyze the volatility of time series data: .

[0074] You can also calculate the value volatility between the values ​​in each categorical target attribute. Value volatility usually refers to the value standard deviation or value variance based on past values.

[0075] Of course, you can also calculate the average true amplitude of the values ​​in each categorical target attribute, using the formula: , , in, is the true volatility of the ith day, It is the high value of the day. This is the low value for the day. It is the average value of yesterday.

[0076] It can be understood that, depending on the specific computing resources, one of the above can be selected to calculate the numerical stability index of each classification target data, or multiple combinations or weighted combinations can be selected to calculate the numerical stability index, which is not specifically limited in the embodiments of the present invention.

[0077] Step S203: monitor each business interface to obtain the attribute feedback message of the second target attribute in the order log of the item, and perform sample marking on the second target attribute according to the attribute feedback message to obtain the target attribute feedback sample including positive samples and negative samples.

[0078] Step S204: Combining the reference target attribute and the target attribute feedback sample into a target attribute sample of the item.

[0079] The method for determining the reference target attribute based on the stability between the values ​​in each classification target attribute in the embodiment of the present invention improves the accuracy and objectivity of the reference target attribute determination, thereby ensuring the construction quality of the target attribute sample.

[0080] Figure 4 is a flow chart of a method for constructing a sample according to another reference embodiment of the present invention. As another embodiment of the present invention, Figure 4 As shown, the sample construction method may include:

[0081] Step S401: classify the first target attribute of the item to obtain the classified target attribute of the item, and determine the reference target attribute of the item according to each of the classified target attributes.

[0082] Step S402, determine each business interface corresponding to the second target attribute in the order log, where the business interface is deployed in an external business system and is used to receive attribute feedback messages of the second target attribute uploaded by users; monitor each of the business interfaces to obtain each attribute feedback message of the second target attribute; and assign corresponding feedback weights to each of the attribute feedback messages according to pre-configured feedback weights of each of the business interfaces.

[0083] Specifically, the embodiment of the present invention pre-counts the business systems related to the historical target attributes of the goods used for business on the trading platform, and counts the external business interfaces of these business systems. For example, taking the second target attribute as the sales price of the goods as an example, the business systems related to the sales price include the settlement business system, the logistics business system, and the operation business system, then the external business interfaces of the settlement business system, the logistics business system, and the operation business system are taken as monitoring objects.

[0084] It can be understood that when there is a business hierarchy relationship between various business interfaces, and the number of business interfaces exceeds the quantity threshold, in order to ensure efficiency, the various business interfaces with business hierarchy relationships can be deleted, and the highest-level business interfaces can be retained. For example, business system 1 and business system 2 corresponding to business interface 1 and business interface 2 are subordinate systems of business system 3. Accordingly, business interface 3 of business system 3 is located at the upper layer of business interface 1 and business interface 2. In this case, business interface 1 and business interface 2 can be deleted, and business interface 3 can be retained to reduce the system consumption of subsequent processing of attribute feedback messages.

[0085] Further, each service interface is monitored to obtain each attribute feedback message about the second target attribute. In addition, the embodiment of the present invention pre-configures a corresponding feedback weight for each service interface according to the correlation between the service system to which each service interface belongs and the second target attribute, and the feedback weight indicates the acceptance degree of the sample construction system to the attribute feedback message. Finally, according to the configured feedback weight, a corresponding feedback weight is assigned to each attribute feedback message.

[0086] Step S403: determine whether the second target attribute has a risk based on the attribute feedback message and the feedback weight corresponding to the attribute feedback message; in response to the second target attribute not having a risk and no attribute feedback message indicating that the second target attribute has a risk is heard within a preset approval time, mark the second target attribute as a negative sample representing safety; in response to the second target attribute having a risk, perform a risk verification on the second target attribute, and mark the second target attribute as a negative sample representing safety or a positive sample representing risk according to the risk verification result, to obtain a target attribute feedback sample containing positive samples and negative samples.

[0087] Specifically, when the attribute feedback message is obtained, the risk level of the second target attribute is calculated according to the feedback weight corresponding to the attribute feedback message, and whether the second target attribute has a risk is determined by comparing whether the risk level reaches a preset threshold. For the calculation of the risk level, when multiple attribute feedback messages are obtained through multiple business interfaces, each attribute feedback message and the corresponding feedback weight can be weighted and summed to obtain the risk level of the target attribute.

[0088] Furthermore, if there is no risk in the second target attribute, in order to ensure the accuracy as much as possible, an approval time can be preset, such as m minutes or h hours, and the attribute feedback messages about the second target attribute received by each business interface within this approval time are monitored. If the content in the attribute feedback message indicates that the second target attribute is safe, the second target attribute is marked as a negative sample representing safety; if the content in the attribute feedback message indicates that there is a risk in the second target attribute, then the second target attribute is determined to be risky and is marked as a positive sample representing risk.

[0089] If the second target attribute has risks, in order to ensure its accuracy as much as possible, the embodiment of the present invention needs to further perform risk verification on the second target attribute, and obtain the second target attributes of other items similar to the item, and further determine whether the second target attribute really has risks by comparing the differences. Finally, according to the risk verification result, the second target attribute is marked as a negative sample representing safety or a positive sample representing risk, and a target attribute feedback sample is obtained.

[0090] According to one embodiment of the present invention, the risk verification is performed on the second target attribute, including: verifying whether the second target attribute has risks based on whether the second target attribute is affected by the superposition of promotional resources; in response to the second target attribute not being affected by the superposition of promotional resources, verifying whether the second target attribute has risks based on the historical second target attributes of the item.

[0091] Specifically, the embodiment of the present invention also provides a method for risk verification, which is mainly based on two levels. First, considering that the second target attribute of the item may be affected by the superposition of various promotional resources, for example, when the second target attribute is the sales price of the item on the trading platform, the promotional price determined by the current item for the full reduction activity is affected by the superposition of the promotional resources of the store coupon, resulting in the actual transaction price being lower than the promotional price during the life cycle of the full reduction activity of the promotional price. In this case, the monitored risk is essentially due to the influence of the superposition of promotional resources, and the value of the second target attribute itself does not have a risk. At this time, the above-mentioned determined risk needs to be corrected to a negative sample where the second target attribute is safe.

[0092] Of course, if the second target attribute is not affected by the superposition of promotional resources, that is, the monitored risk is the risk feedback on the value of the second target attribute, then it is also necessary to compare the historical second target attribute of the item with the current second target attribute. If the second target attribute is smaller than the historical second target attribute, it means that the second target attribute is indeed risky and is marked as a risky positive sample; otherwise, if the historical second target attribute is smaller than the second target attribute, under the premise that the historical second target attributes are all safe, the risk judgment of the second target attribute is unreasonable and the above-mentioned determined risks need to be corrected to safe negative samples.

[0093] Figure 5 It is a schematic diagram of constructing a target attribute feedback sample of an embodiment of the present invention. For the second target attribute in the order log of the item, first monitor each business interface to obtain the attribute feedback message whether the second target attribute has a risk. If there is no risk, then within the preset approval time, continue to monitor the attribute feedback message whether the second target attribute has a risk. If there is still no risk, then the second target attribute can be marked as a safe negative sample; on the contrary, if an attribute feedback message indicating that the second target attribute has a risk is received within the approval time, then the second target attribute is marked as a risky positive sample.

[0094] Furthermore, if there is risk feedback in the first attribute feedback message, first determine whether the risk target attribute corresponding to the current risk feedback is the second target attribute in the order log. If not, it means that the second target attribute is affected by the superposition of promotional resources and is mistakenly considered to be risky, and the second target attribute is marked as a corrected negative sample. For example, the sales price of item A is p yuan, and during the effective life cycle of p yuan, there is a superposition of promotional resources for full reduction, causing p yuan to drop to d yuan, and the risk feedback is for the final value d, causing the sales price of p yuan to be misjudged as risky. At this time, it is necessary to correct this risk feedback and correct p yuan to a safe negative sample.

[0095] Furthermore, if the second target attribute is not affected by the superposition of promotional resources, that is, the risk target attribute is the second target attribute in the order log, then it is also necessary to retrieve the historical second target attribute of the item, compare the second target attribute with the historical second target attribute, and determine whether the second target attribute is smaller than the historical second target attribute. If the second target attribute is smaller than the historical second target attribute, then there is a risk in the second target attribute and it is marked as a positive sample. Otherwise, correct the risk feedback of the second target attribute, determine that there is no risk in the second target attribute, and mark it as a corrected negative sample.

[0096] Step S404: Combining the reference target attribute and the target attribute feedback sample into a target attribute sample of the item.

[0097] The embodiment of the present invention marks the second target attribute through attribute feedback messages, and when initially receiving the risk of the second target attribute, accurately performs risk verification based on two levels: whether the second target attribute is affected by the superposition of promotional resources and the historical second target attribute of the item, so as to further improve the accuracy of sample marking, construct accurate target attribute feedback samples, and obtain high-quality target attribute samples.

[0098] According to the target attribute samples constructed as above, an embodiment of the present invention further provides a method for training an item attribute warning model, comprising: inputting the target attribute samples into a predefined item attribute warning model; the target attribute samples are constructed by any one of the methods for constructing the target attribute samples as mentioned above; based on the maximum likelihood method and the target attribute samples, the parameters of the item attribute warning model are adjusted by calculating the positive sample loss function and the negative sample loss function of the item attribute warning model to obtain a trained item attribute warning model with attribute warning capability.

[0099] Specifically, the target attribute samples constructed above are used to train the item attribute warning model. The embodiment of the present invention adopts the maximum likelihood method, and considers both positive samples and negative samples, decouples the positive samples and negative samples into positive sample loss function and negative sample loss function, and constructs the loss function of the item attribute warning model: , Among them, p(y|s) represents the estimated probability of positive feedback based on the target attribute s, which can be calculated by the activation function softmax.

[0100] The decoupled positive sample loss function and negative sample loss function constructed in the embodiment of the present invention not only facilitates the adjustment of the proportion of the positive sample and negative sample loss functions, but also can introduce multiple negative samples and assign differentiated sample weights.

[0101] According to the item attribute warning model constructed by the above training, an embodiment of the present invention also provides an item attribute warning method, including: inputting the target attribute to be judged of the item into the item attribute warning model, the item attribute warning model is obtained according to the training method of the above item attribute warning model; obtaining the probability of the existence of risk of the target attribute to be judged of the item output by the item attribute warning model.

[0102] Specifically, based on the item attribute warning model trained using the target attribute samples, the target attribute to be judged of the item is input into the item attribute warning model, and the features of the target attribute to be judged are extracted and the ability to judge the target attribute learned by the item attribute warning model is used to calculate the probability that there is a risk in the target attribute to be judged of the item.

[0103] According to one embodiment of the present invention, after obtaining the probability of the existence of risk in the target attribute to be judged of the item output by the item attribute warning model, the method further includes: in response to the probability of the existence of risk in the target attribute to be judged being lower than a preset risk threshold, judging the target attribute to be judged to be a safe target attribute, adding the safe target attribute to the order log of the item to obtain an attribute feedback message of the safe target attribute; in response to the probability of the existence of risk in the target attribute to be judged reaching the risk threshold, judging the target attribute to be judged to be a risk target attribute, and adding the risk target attribute as a positive sample to the target attribute sample.

[0104] Specifically, after using the item attribute warning model to obtain the probability that the target attribute to be judged has a risk, the target attribute feedback sample can also be updated according to the specific risk probability value. If the risk probability of the target attribute output by the item attribute warning model is lower than the risk threshold, it means that the target attribute is safe, then the target attribute can be added to the order log of the item to further obtain the attribute feedback message of each business interface about the target attribute, and then marked as a positive sample or a negative sample according to the attribute feedback message. If the risk probability of the target attribute output by the item attribute warning model reaches the risk threshold, it means that the target attribute to be judged has a risk, and then the target attribute is added to the target attribute sample as a positive sample.

[0105] Figure 6 : is a schematic diagram of an item attribute warning method according to an embodiment of the present invention. The figure integrates the training of the item attribute warning model and the online prediction of the item attribute warning model. In the training stage, the target attribute samples including the benchmark target attribute of the item, the target attribute feedback samples of the positive sample, and the target attribute feedback samples of the negative sample are used to train the item attribute warning model, so as to obtain the item attribute warning model with the attribute warning capability. Then the item attribute warning model is put online for use. The item attribute warning model receives the target attribute to be judged of the item, and calculates the probability that the target attribute to be judged has a risk through feature extraction and recall of similar samples. If the probability is lower than the risk threshold, it is determined that the target attribute does not have a risk, the target attribute passes the warning, and the target attribute is added to the order log of the item. If the probability reaches the risk threshold, it means that there is a risk. In order to ensure the accuracy, this step can also be added with a secondary check, and the target attribute is manually checked for a secondary check. If the target attribute is checked to have no risk, the target attribute is corrected to pass the warning and added to the order log; if the target attribute is checked to have a risk, the target attribute is intercepted and marked as a positive sample to update the original positive sample.

[0106] The embodiment of the present invention applies the above-constructed item attribute warning model to the warning of the target attributes of items. The item attribute warning model obtained by training based on the above-constructed target attribute samples ensures the accuracy of the judgment results, and the judgment of the target attributes in the item dimension can provide warning services for each item on the platform, thereby protecting the interests of merchants.

[0107] Figure 7 Schematic diagram of the main modules of the sample construction device according to the embodiment of the present invention. Figure 7 As shown, the sample construction device 700 mainly includes a first sample determination module 701 , a second sample determination module 702 and a target attribute sample determination module 703 .

[0108] The first sample determination module 701 is used to classify the first target attribute of the item to obtain the classified target attribute of the item, and determine the reference target attribute of the item according to each of the classified target attributes;

[0109] The second sample determination module 702 is used to monitor each business interface to obtain the attribute feedback message of the second target attribute in the order log of the item, and perform sample marking on the second target attribute according to the attribute feedback message to obtain a target attribute feedback sample including positive samples and negative samples;

[0110] The target attribute sample determination module 703 is used to combine the reference target attribute and the target attribute feedback sample to form a target attribute sample of the item.

[0111] According to one embodiment of the present invention, the first sample determination module 701 is also used to: determine the priority of each of the classification target attributes, and judge the numerical stability of each of the classification target attributes according to the priority; and obtain the baseline target attribute of the object based on the first classification target attribute that meets the preset numerical stability standard.

[0112] According to another embodiment of the present invention, the first sample determination module 701 is further used to judge the numerical stability index of each of the classification target attributes by calculating at least one of the numerical standard deviation, numerical variance, numerical mean absolute deviation, numerical moving average, numerical moving standard deviation, numerical volatility, and numerical average true amplitude of each of the classification target attributes.

[0113] According to another embodiment of the present invention, the second sample determination module 702 is further used to: determine each business interface corresponding to the second target attribute in the order log, the business interface is deployed in an external business system, and is used to receive attribute feedback messages of the second target attribute uploaded by users; monitor each of the business interfaces to obtain each attribute feedback message of the second target attribute; and assign corresponding feedback weights to each of the attribute feedback messages according to pre-configured feedback weights of each of the business interfaces.

[0114] According to another embodiment of the present invention, the second sample determination module 702 is further used to: determine whether the second target attribute has a risk based on the attribute feedback message and the feedback weight corresponding to the attribute feedback message; in response to the second target attribute not having a risk and no attribute feedback message indicating that the second target attribute has a risk is heard within a preset approval time, mark the second target attribute as a negative sample representing safety; in response to the second target attribute having a risk, perform a risk verification on the second target attribute, and mark the second target attribute as a negative sample representing safety or a positive sample representing risk according to the risk verification result.

[0115] According to another embodiment of the present invention, the second sample determination module 702 is further used to: verify whether there is a risk in the second target attribute based on whether the second target attribute is affected by the superposition of promotional resources; and verify whether there is a risk in the second target attribute based on the historical second target attribute of the item in response to the second target attribute not being affected by the superposition of promotional resources.

[0116] Figure 8 Schematic diagram of the main modules of the object attribute warning device according to an embodiment of the present invention. Figure 8 As shown, the item attribute warning device 800 mainly includes a data input module 801 and an attribute warning module 802 .

[0117] A data input module 801, used to input the target attribute of the item to be determined into the item attribute warning model, wherein the item attribute warning model is trained according to the method of claim 7;

[0118] The attribute warning module 802 is used to obtain the probability of the target attribute to be judged of the item having a risk output by the item attribute warning model.

[0119] According to one embodiment of the present invention, the item attribute warning device 800 also includes an update module (not shown in the figure), which is used to: after obtaining the probability of the existence of risk in the target attribute to be judged of the item output by the item attribute warning model, in response to the probability of the existence of risk in the target attribute to be judged being lower than a preset risk threshold, determine that the target attribute to be judged is a safe target attribute, add the safe target attribute to the order log of the item to obtain an attribute feedback message of the safe target attribute; in response to the probability of the existence of risk in the target attribute to be judged reaching the risk threshold, determine that the target attribute to be judged is a risk target attribute, and add the risk target attribute as a positive sample to the target attribute sample.

[0120] Fig. 9 is an exemplary system architecture diagram to which embodiments of the present invention can be applied.

[0121] like Fig. 9 As shown, system architecture 900 may include terminal devices 901, 902, 903, network 904 and server 905. Network 904 is used to provide a medium for communication links between terminal devices 901, 902, 903 and server 905. Network 904 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0122] Users can use terminal devices 901, 902, 903 to interact with server 905 through network 904 to receive or send messages, etc. Terminal devices 901, 902, 903 can be installed with various communication client applications, such as sample construction applications, etc. (only for example).

[0123] The terminal devices 901 , 902 , and 903 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers, etc.

[0124] The server 905 may be a server that provides various services, such as a background management server (only an example) that supports the construction of samples by users using the terminal devices 901, 902, and 903. The background management server may classify the first target attribute of the item to obtain the classified target attribute of the item, and determine the reference target attribute of the item according to each of the classified target attributes; monitor each business interface to obtain the attribute feedback message of the second target attribute in the order log of the item, and perform sample marking on the second target attribute according to the attribute feedback message to obtain the target attribute feedback sample containing positive samples and negative samples; form the target attribute sample of the item with the reference target attribute and the target attribute feedback sample, and feed back the processing result (such as the target attribute sample, etc. - only an example) to the terminal device.

[0125] It should be noted that the sample construction method and item attribute warning method provided in the embodiments of the present invention are generally executed by the server 905 , and accordingly, the sample construction device and item attribute warning device are generally set in the server 905 .

[0126] It should be understood that Fig. 9 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0127] Reference below Fig.10 , Fig.10 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. Fig.10The terminal device or server 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.

[0128] like Fig.10 As shown, the computer system 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage part 1008 into a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the system 1000 are also stored. The CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0129] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.

[0130] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the central processing unit (CPU) 1001, the above-mentioned functions defined in the system of the present invention are executed.

[0131] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may 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 of computer-readable storage media may include, but are not limited to: 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 the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal 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 computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination thereof.

[0132] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0133] The units involved in the embodiments of the present invention may be implemented by software or hardware. The units described may also be set in a processor, for example, it may be described as: a processor includes: a first sample determination module, a second sample determination module and a target attribute sample determination module.

[0134] Among them, the names of these modules do not constitute a limitation of the modules themselves in some cases. For example, the target attribute sample determination module can also be described as "a module for combining the baseline target attribute and the target attribute feedback sample to form a target attribute sample of the object."

[0135] On the other hand, the present invention also provides a computer-readable medium, which may be included in the device described in the embodiment; or may exist independently without being assembled into the device. The computer-readable medium carries one or more programs, and when the one or more programs are executed by a device, the device includes: classifying the first target attribute of the item to obtain the classified target attribute of the item, and determining the reference target attribute of the item according to each of the classified target attributes; monitoring each business interface to obtain the attribute feedback message of the second target attribute in the order log of the item, and marking the second target attribute according to the attribute feedback message to obtain the target attribute feedback sample containing positive samples and negative samples; and composing the reference target attribute and the target attribute feedback sample into the target attribute sample of the item.

[0136] The technical solution according to the embodiment of the present invention has the following advantages or beneficial effects: by classifying the first target attribute of the item, the classified target attribute of the item is obtained, and the baseline target attribute of the item is determined according to each classified target attribute; each business interface is monitored to obtain the attribute feedback message of the second target attribute in the order log of the item, and the second target attribute is sampled according to the attribute feedback message to obtain the target attribute feedback sample containing positive samples and negative samples; the technical solution of combining the baseline target attribute and the target attribute feedback sample to form the target attribute sample of the item realizes the construction of a complete, accurate and dynamic target attribute sample, and the target attribute sample constructed by sample labeling the second target attribute based on the attribute feedback message ensures the performance of the item attribute warning model subsequently trained using the target attribute sample, thereby ensuring the accuracy of the warning of the item target attribute.

[0137] The specific implementation methods described herein do not constitute limitations on the scope of protection of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A method for constructing a sample, characterized in that: include: Classifying the first target attribute of the item to obtain classified target attributes of the item, and determining the reference target attribute of the item according to each of the classified target attributes; Monitor each business interface to obtain an attribute feedback message of a second target attribute in the order log of the item, and perform sample marking on the second target attribute according to the attribute feedback message to obtain a target attribute feedback sample including positive samples and negative samples; The reference target attribute and the target attribute feedback sample are combined to form a target attribute sample of the item.

2. The method according to claim 1, characterized in that: Determining the reference target attribute of the item according to each of the classification target attributes includes: Determining the priority of each of the classification target attributes, and judging the numerical stability of each of the classification target attributes according to the priority; A baseline target attribute of the object is obtained based on the first classification target attribute that meets the preset numerical stability standard.

3. The method according to claim 2, characterized in that Determining the numerical stability of each of the classification target attributes includes: By calculating at least one of the numerical standard deviation, numerical variance, numerical mean absolute deviation, numerical moving average, numerical moving standard deviation, numerical volatility, and numerical average true amplitude of each of the classification target attributes, the numerical stability index of each of the classification target attributes is judged.

4. The method according to claim 1, characterized in that: Monitoring each business interface to obtain the attribute feedback message of the second target attribute in the order log of the item, including: Determine each business interface corresponding to the second target attribute in the order log, where the business interface is deployed in an external business system and is used to receive an attribute feedback message of the second target attribute uploaded by a user; Monitoring each of the service interfaces to obtain each attribute feedback message of the second target attribute; According to the pre-configured feedback weights of the service interfaces, corresponding feedback weights are allocated to the attribute feedback messages.

5. The method according to claim 4, characterized in that According to the attribute feedback message, sample marking is performed on the second target attribute, including: determining whether the second target attribute has a risk according to the attribute feedback message and the feedback weight corresponding to the attribute feedback message; In response to the second target attribute not having risk and no attribute feedback message indicating that the second target attribute has risk is heard within a preset approval time, marking the second target attribute as a negative sample indicating safety; In response to the existence of risk in the second target attribute, a risk check is performed on the second target attribute, and the second target attribute is marked as a negative sample representing safety or a positive sample representing risk according to a risk check result.

6. The method according to claim 5, characterized in that Performing risk verification on the second target attribute includes: Based on whether the second target attribute is affected by the superposition of promotion resources, verifying whether the second target attribute has risks; In response to the second target attribute not being affected by the superposition of promotional resources, based on the historical second target attribute of the item, it is verified whether the second target attribute has risks.

7. A method for training an item attribute warning model, characterized in that: include: Inputting a target attribute sample into a predefined item attribute warning model; the target attribute sample is constructed by any method in claims 1 to 6; Based on the maximum likelihood method and the target attribute samples, the parameters of the item attribute warning model are adjusted by calculating the positive sample loss function and the negative sample loss function of the item attribute warning model to obtain a trained item attribute warning model with attribute warning capability.

8. An object attribute early warning method, characterized in that: include: Inputting the target attribute of the item to be judged into an item attribute warning model, wherein the item attribute warning model is trained according to the method of claim 7; The probability that the target attribute to be determined of the item has a risk output by the item attribute warning model is obtained.

9. The method according to claim 8, characterized in that After obtaining the probability output by the item attribute warning model that the target attribute to be determined of the item has a risk, the method further includes: In response to the probability that the target attribute to be determined has a risk lower than a preset risk threshold, determining that the target attribute to be determined is a safety target attribute, adding the safety target attribute to the order log of the item, and obtaining an attribute feedback message of the safety target attribute; In response to the probability that the target attribute to be determined has a risk reaching the risk threshold, the target attribute to be determined is determined to be a risky target attribute, and the risky target attribute is added to the target attribute sample as a positive sample.

10. A sample construction device, characterized in that: include: A first sample determination module is used to classify the first target attribute of the item to obtain the classified target attribute of the item, and determine the reference target attribute of the item according to each of the classified target attributes; A second sample determination module is used to monitor each business interface to obtain an attribute feedback message of a second target attribute in the order log of the item, and to perform sample marking on the second target attribute according to the attribute feedback message to obtain a target attribute feedback sample including positive samples and negative samples; The target attribute sample determination module is used to combine the reference target attribute and the target attribute feedback sample to form a target attribute sample of the item.

11. An object attribute warning device, characterized in that: include: A data input module, used to input the target attribute of the item to be judged into the item attribute warning model, wherein the item attribute warning model is trained according to the method of claim 7; The attribute warning module is used to obtain the probability of the existence of risk in the target attribute to be judged of the item output by the item attribute warning model.

12. A mobile electronic device terminal, characterized in that: include: one or more processors; a storage device 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 method according to any one of claims 1 to 9.

13. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images