Product anomaly prediction model training, product recommendation method, device and equipment
By cold-start processing and iterative update of the initial data features, a commodity abnormality prediction model is formed, which solves the problem of low product recommendation accuracy in the prior art, and achieves higher prediction accuracy and recommendation accuracy.
Patent Information
- Application Number
- CN202210068302.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-20
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-01-20
AI Technical Summary
In the prior art, the accuracy of product recommendation is low, which leads to the high probability of users purchasing poor products and is unable to make targeted recommendations.
By obtaining the initial data characteristics and performing cold start processing, target data characteristics are generated, input into the preset identification model for iterative updates until the predicted loss value reaches the convergence condition, and a product abnormality prediction model is formed to filter recommended products.
It improves the prediction accuracy and recommendation accuracy of the product anomaly prediction model, and can effectively screen out products with high complaint rates or high complaint levels to improve user experience.
Smart Images

Figure CN114511379B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of prediction model technology, and in particular to a product anomaly prediction model training, product recommendation method, device, equipment and storage medium. Background Art
[0002] With the development of science and technology, product selection and purchase on online platforms has gradually become mainstream. Major platforms often display all product information matching the user's search terms on the display interface, or recommend products based on the user's historical purchase records. However, these recommendation methods often push all products to users without making targeted recommendations, resulting in uneven products pushed to users, a high probability of users purchasing inferior products, and a low accuracy rate of product recommendations. Summary of the invention
[0003] The embodiments of the present invention provide a product anomaly prediction model training, product recommendation method, device, equipment and storage medium to solve the problem of low product recommendation accuracy in the prior art.
[0004] A commodity anomaly prediction model training method, comprising:
[0005] Acquire a data feature set; the data feature set includes at least one initial data feature and a commodity abnormality label corresponding to the initial data feature one by one;
[0006] Performing cold start processing on each of the initial data features to obtain newly added data features corresponding to each of the initial data features, and determining target data features corresponding to each of the initial data features based on the initial data features and the newly added data features;
[0007] Inputting the target data features into a preset recognition model including initial parameters to determine a predicted anomaly label for each of the initial data features;
[0008] Determining a predicted loss value corresponding to the preset recognition model according to the product abnormality label and the predicted abnormality label;
[0009] When the predicted loss value does not reach the preset convergence condition, the initial parameters in the preset recognition model are iteratively updated until the predicted loss value reaches the convergence condition, and the preset recognition model after convergence is recorded as a commodity abnormality prediction model.
[0010] A product recommendation method, comprising:
[0011] Receiving a product recommendation instruction sent by a target user; the product recommendation instruction includes required product information;
[0012] Determine, from a preset commodity information storage platform, preset commodity information matching the required commodity information and record it as commodity information to be screened; the preset commodity information storage platform includes at least one preset commodity information;
[0013] Inputting the product information to be screened into a product anomaly prediction model to determine the product anomaly type and product anomaly level corresponding to each product information to be screened;
[0014] According to the product abnormality type and the product abnormality level, recommended product information is selected from all the product information to be screened, and the recommended product information is pushed to the target user.
[0015] A commodity anomaly prediction model training device, comprising:
[0016] A data feature acquisition module, used to acquire a data feature set; the data feature set includes at least one initial data feature and a commodity abnormality label corresponding to the initial data feature one by one;
[0017] A cold start processing module, used to perform cold start processing on each of the initial data features to obtain a newly added data feature corresponding to each of the initial data features, and determine a target data feature corresponding to each of the initial data features based on the initial data features and the newly added data features;
[0018] A data anomaly prediction module, used for inputting the target data features into a preset recognition model including initial parameters to determine a predicted anomaly label of each of the initial data features;
[0019] A predicted loss value determination module, used to determine a predicted loss value corresponding to the preset recognition model according to the commodity abnormality label and the predicted abnormality label;
[0020] The model training module is used to iteratively update the initial parameters in the preset recognition model when the predicted loss value does not reach the preset convergence condition, until the predicted loss value reaches the convergence condition, and record the preset recognition model after convergence as the commodity abnormality prediction model.
[0021] A commodity recommendation device, comprising:
[0022] An instruction receiving module, used to receive a product recommendation instruction sent by a target user; the product recommendation instruction includes the required product information;
[0023] A product matching module, used to determine preset product information matching the required product information from a preset product information storage platform and record it as product information to be screened; the preset product information storage platform includes at least one preset product information;
[0024] A commodity prediction module, used for inputting the commodity information to be screened into a commodity anomaly prediction model to determine a commodity anomaly type and a commodity anomaly level corresponding to each commodity information to be screened;
[0025] The product recommendation module is used to select recommended product information from all the product information to be screened according to the product anomaly type and the product anomaly level, and push the recommended product information to the target user.
[0026] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned commodity abnormality prediction model training method when executing the computer program, or implements the above-mentioned commodity recommendation method when executing the computer program.
[0027] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the above-mentioned product anomaly prediction model training method, or the computer program, when executed by a processor, implements the above-mentioned product recommendation method.
[0028] The above-mentioned product anomaly prediction model training, product recommendation method, device, equipment and storage medium, the product anomaly prediction model training method performs cold start processing on the initial data features, so that each target data feature after the cold start processing has the attribute of the corresponding data classification, that is, the target data feature is increased by one more dimension, thereby improving the model training efficiency of the preset recognition model, and also improving the prediction accuracy of the trained product anomaly prediction model, so that when making product recommendations, the effectiveness of the products can be screened by predicting anomaly labels, thereby improving the accuracy of product recommendations.
[0029] The product recommendation method of the present invention determines the product anomaly type and product anomaly level of each product according to the product information to be screened through a product anomaly prediction model, so that the products can be screened according to the product anomaly type and product anomaly level, so that products with high complaint rates or high complaint levels can be eliminated, thereby improving the accuracy of product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0031] Figure 1Schematic diagram of an application environment of a commodity anomaly prediction model training method in one embodiment of the present invention;
[0032] Figure 2 is a flow chart of a method for training a commodity anomaly prediction model in one embodiment of the present invention;
[0033] Figure 3 is a flow chart of step S20 in the commodity anomaly prediction model training method in one embodiment of the present invention;
[0034] Figure 4 is a flow chart of a commodity recommendation method in one embodiment of the present invention;
[0035] Figure 5 This is a principle block diagram of a commodity anomaly prediction model training device in one embodiment of the present invention;
[0036] Figure 6 It is a principle block diagram of a cold start processing module in a commodity abnormality prediction model training device in one embodiment of the present invention;
[0037] Figure 7 is a principle block diagram of a commodity recommendation device in one embodiment of the present invention;
[0038] Figure 8 is a schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION
[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] The commodity abnormality prediction model training method provided by the embodiment of the present invention can be applied as follows: Figure 1 Specifically, the commodity anomaly prediction model training method is applied in a commodity anomaly prediction model training system, and the commodity anomaly prediction model training system includes: Figure 1The client and server shown in the figure communicate with each other through the network to solve the problem of low accuracy of product recommendation in the prior art. The client, also known as the user end, refers to a program corresponding to the server that provides local services to customers. The client can be installed on, but not limited to, various personal computers, laptops, smart phones, tablets, and portable wearable devices. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.
[0041] In one embodiment, if Figure 2 As shown, a commodity abnormality prediction model training method is provided, and the method is applied in Figure 1 The server in the example is used as an example to illustrate the process, including the following steps:
[0042] S10: Acquire a data feature set; the data feature set includes at least one initial data feature and a commodity abnormality label corresponding to the initial data feature one by one.
[0043] It can be understood that the initial data features are constructed based on different data, and the initial data features are different in different application scenarios. For example, in the shopping application scenario, the initial data features can be features constructed based on different users' purchase records, purchase evaluation records, product use records, product refund records, product compensation records, etc.; in the insurance risk control application scenario, the initial data features can be features constructed based on user attribute information, user behavior information, insurance product information, premium payment information, etc. Among them, one initial data feature corresponds to a product abnormality label. Exemplarily, in the shopping application scenario, the product abnormality label can be generated based on product evaluation records, product refund records, or product compensation records; in the insurance risk control application scenario, it can be generated based on premium payment information, insurance product information, etc. The product abnormality label characterizes the product abnormality type and product abnormality level. In the shopping application scenario, the product abnormality type can be that the product has been complained about, or that the product has not been complained about; the product abnormality level can be determined based on the amount of the user's claim after the product has been complained about; in the insurance risk control application scenario, the product abnormality type can be determined based on the probability of the user's insurance after the insurance, and the product abnormality level can be determined based on the amount of compensation paid when the user's insurance is insured.
[0044] S20: performing cold start processing on each of the initial data features to obtain newly added data features corresponding to each of the initial data features, and determining target data features corresponding to each of the initial data features based on the initial data features and the newly added data features.
[0045] It is understandable that, for example, in a shopping application scenario, when a user initially purchases a product, it may not be confirmed whether the product is defective or has not yet been used. However, returns, exchanges or claims for different products have a certain period, and users generally perform usage tests on the products within the prescribed period. For another example, in an insurance risk control application scenario, when a customer initially purchases insurance, there is no claim. Therefore, for this type of product, there may not be any abnormalities in the initial stage (such as the product being defective or the insurance not being claimed). In this embodiment, all initial data features are cold-started, so that the same type of data features can be used to provide more feature information for data features that may not have abnormalities in the initial stage, thereby improving the accuracy and efficiency of the model when training the model.
[0046] S30: Inputting the target data features into a preset recognition model including initial parameters to determine a predicted abnormal label for each of the initial data features.
[0047] Specifically, after determining the target data features corresponding to each of the initial data features according to the initial data features and the newly added data features, the target data features are input into a preset recognition model, and then the initial data features are recognized by the preset recognition model to determine the predicted abnormality labels corresponding to the initial data features. It can be understood that since each data to be input is generated according to the initial data features and the newly added data features corresponding thereto, the predicted abnormality labels output by the preset recognition model according to the target data features are associated with the initial data features corresponding to the target data features.
[0048] Among them, the predicted anomaly label is generated according to the predicted classification label and the predicted level label. The predicted classification label is the preset recognition model that determines the abnormal classification according to the target data characteristics (such as in the shopping application scenario, determine whether it is a bad product, whether the user will complain, etc.; in the insurance risk control scenario, determine whether there is an accident, that is, the accident rate); the predicted level label is the preset recognition model that determines the abnormal level according to the target data characteristics (such as in the shopping application scenario, if the user will complain, the merchant needs to claim the specific amount; in the insurance risk control scenario, determine the amount of compensation). The preset recognition model includes an abnormal type prediction module and an abnormal level prediction module. The abnormal type prediction module is used to perform abnormal classification on the target data characteristics, and the abnormal level prediction module is used to predict the abnormal level of the target data characteristics. Among them, the abnormal type prediction module can adopt a binary classification model, and the abnormal level prediction module can adopt a regression model.
[0049] S40: Determine a predicted loss value corresponding to the preset recognition model according to the product abnormality label and the predicted abnormality label.
[0050] It can be understood that there is an error between the predicted abnormal label and the product abnormal label obtained when the untrained preset recognition model predicts the target data feature. Then, the prediction loss value of the preset recognition model can be determined based on the product abnormal label and the predicted abnormal label corresponding to the same initial data feature. In this way, the preset recognition model can be updated according to the prediction loss value, thereby improving the prediction accuracy of the preset recognition model.
[0051] Specifically, after the target data features are input into a preset recognition model containing initial parameters to determine the predicted anomaly labels corresponding to each initial data feature through the preset recognition model, the predicted loss value of the preset recognition model can be determined based on the product anomaly labels and predicted anomaly labels corresponding to the same initial data feature, such as determining whether the product anomaly labels and predicted anomaly labels are the same.
[0052] S50: When the predicted loss value does not reach a preset convergence condition, iteratively update the initial parameters in the preset recognition model until the predicted loss value reaches the convergence condition, and record the preset recognition model after convergence as a commodity abnormality prediction model.
[0053] It can be understood that the convergence condition can be the condition that the predicted loss value is less than the set threshold, that is, when the predicted loss value is less than the set threshold, the training is stopped; the convergence condition can also be the condition that the predicted loss value is very small after 10,000 calculations and will not decrease anymore, that is, when the predicted loss value is very small after 10,000 calculations and will not decrease, the training is stopped and the preset recognition model after convergence is recorded as the product abnormality prediction model.
[0054] Furthermore, after determining the predicted loss value corresponding to the preset recognition model based on the product abnormality label and the predicted abnormality label corresponding to the same initial data feature, when the predicted loss value does not reach the preset convergence condition, the initial parameters of the preset recognition model are adjusted according to the predicted loss value, and the target data feature is re-input into the preset recognition model after the initial parameters are adjusted, so that when the predicted loss value of the target data feature reaches the preset convergence condition, another target data feature is selected, and the above steps S30 to S50 are executed to obtain the predicted loss value corresponding to the target data feature, and when the predicted loss value does not reach the preset convergence condition, the initial parameters of the preset recognition model are adjusted again according to the predicted loss value, so that the predicted loss value of the target data feature reaches the preset convergence condition.
[0055] In this way, after the preset recognition model is trained through all target data features, the results output by the preset recognition model can continue to approach the accurate results, making the recognition accuracy higher and higher, until the predicted loss values of all target data features reach the preset convergence conditions, and the preset recognition model after convergence is recorded as the product abnormality prediction model.
[0056] In this embodiment, by performing cold start processing on the initial data features, each target data feature after the cold start processing has the attributes of the corresponding data classification, that is, the target data feature is given an additional dimension, thereby improving the model training efficiency of the preset recognition model, and also improving the prediction accuracy of the trained product anomaly prediction model, so that when making product recommendations, the effectiveness of the products can be screened by predicting anomaly labels, thereby improving the accuracy of product recommendations.
[0057] In one embodiment, before obtaining the data feature set, the process includes:
[0058] At least one sample data group is obtained; one of the sample data groups includes at least one sample data.
[0059] It is understandable that different sample data groups can be sample data of different users. In the shopping application scenario, the sample data can be basic user information (such as user age, gender, etc.), user product behavior information (such as product purchase records, product evaluation records, product use records, product refund records, and product compensation records), etc. In the insurance risk control application scenario, the sample data can be user attribute information (such as user name, age, gender, etc.), user behavior information (whether there has been any risk behavior, etc.), insurance product information, premium payment information, etc.
[0060] Data processing is performed on all the sample data to obtain initial data features corresponding to each of the sample data groups.
[0061] It can be understood that the data processing in this embodiment may include data preprocessing (such as null value processing, outlier processing, etc.), normalization processing, feature screening, etc. Then, by performing data processing on the sample data in each sample data group, the initial data features corresponding to each sample data group can be obtained.
[0062] Generate commodity abnormality labels corresponding to each of the initial data features based on all the sample data.
[0063] It is understandable that the above description points out that the product anomaly label represents the product anomaly type and product anomaly level. In the shopping application scenario, the product anomaly type can be a product that has been complained about, or a product that has not been complained about; the product anomaly level can be determined based on the amount of compensation claimed by the user after the product has been complained about; in the insurance risk control application scenario, the product anomaly type can be determined based on the probability of the user being insured, and the product anomaly level can be determined based on the amount of compensation paid when the user is insured.
[0064] Therefore, in the shopping application scenario, it is possible to determine whether the user has ever complained about a product based on the sample data of different users, and then determine the product abnormality type; and if the user has complained about a product, the claim amount after the product complaint is obtained, and then the product abnormality level is determined based on the claim amount. In the insurance risk control scenario, it is possible to determine whether the user has ever experienced an accident based on the sample data of different users, and then determine the product abnormality type; and if the user has experienced an accident, the amount of compensation paid when the user is insured is obtained, and then the product abnormality level is determined based on the amount of compensation paid.
[0065] The data feature set is generated according to the initial data features and the commodity abnormality label.
[0066] Specifically, after the initial data features and the commodity anomaly labels corresponding to each sample data group are determined, the sample data group can be associated with the initial data features and the commodity anomaly labels to generate a data feature set.
[0067] In one embodiment, the data processing of the sample data in each of the sample data groups to obtain initial data features corresponding to each of the sample data groups includes:
[0068] All the sample data are preprocessed to obtain preprocessed data features corresponding to each of the sample data groups.
[0069] It can be understood that the preprocessing in this embodiment may include abnormal value processing (such as data format does not meet the requirements, for example, the purchase amount of the product is a negative number, then the purchase amount of the product is an abnormal value. Then all data associated with the purchase amount of the product can be deleted), null value processing (for example, there is a product purchase record, but the purchase amount of the product is a null value, then all data associated with the purchase record of the product can be deleted), category information conversion into numerical processing (for example, if the sample data is that there is no product refund record, then the value corresponding to the non-existent product refund record can be recorded as 0, and the value corresponding to the existing product refund record can be recorded as 1), etc. Then, after preprocessing the sample data in each sample data group, the preprocessed data features corresponding to each sample data group can be obtained.
[0070] Normalization processing is performed on each of the preprocessed data features to obtain a normalized data feature corresponding to each of the preprocessed data features.
[0071] It can be understood that the normalization process is to eliminate the adverse effects caused by the different dimensions between the preprocessed data features, so that the expressions of all the preprocessed data features after the normalization process are the same, so that the convergence speed of the model can be accelerated when the model is trained through the initial data features. Specifically, after the sample data in each sample data group is preprocessed to obtain the preprocessed data features corresponding to each sample data group, each preprocessed data feature can be normalized, so that each preprocessed data feature is converted into a dimensionless expression, that is, the format of each preprocessed data feature is unified, and then the normalized data features corresponding to each preprocessed data feature are obtained.
[0072] Feature screening is performed on each of the normalized data features to obtain the initial data features corresponding to each of the sample data groups.
[0073] Specifically, after normalizing each preprocessed data feature to obtain a normalized data feature corresponding to each preprocessed data feature, feature screening is performed on each normalized data feature, that is, irrelevant features or redundant features are selected from the normalized data features. For example, if there are multiple repeated normalized data features, only any one normalized data feature can be retained and other identical normalized data features can be eliminated, thereby reducing the impact of noise in the data, reducing the time required for model training, and improving the accuracy of model training.
[0074] In one embodiment, if Figure 3 As shown, in step S20, that is, the cold start processing is performed on each of the initial data features to obtain the newly added data features corresponding to each of the initial data features, and the target data features corresponding to each of the initial data features are determined according to the initial data features and the newly added data features, including:
[0075] S201: performing feature clustering on each of the initial data features to obtain at least one data feature group; one of the data feature groups includes at least one initial data feature.
[0076] It can be understood that feature clustering is the process of dividing data with similar or identical features into the same category. In the above description, it is pointed out that in the shopping application scenario, the initial data features can be features constructed based on different users' purchase records, purchase evaluation records, product usage records, product refund records, product compensation records, etc.; therefore, in the shopping application scenario, feature clustering can be performed based on the type of goods purchased by the user, so that the initial data features of the same or similar goods are classified into one category. In the insurance risk control application scenario, the initial data features can be features constructed based on user attribute information, user behavior information, insurance product information, premium payment information, etc. Therefore, in the insurance risk control application scenario, feature clustering can be performed based on the type of insurance product insured by the user, so that the initial data features of the same insurance product are classified into one category.
[0077] Specifically, after obtaining the data feature set, each initial data feature can be clustered to divide the data according to the preset division requirements. For example, in the above description, the initial data features of the same or similar commodities are classified into one category in the shopping application scenario; in the insurance risk control application scenario, the initial data features of the same insurance product are classified into one category, thereby obtaining at least one data feature group, and a data feature group includes at least one initial data feature. For example, when an initial data feature does not match all other initial data features, the initial data feature can be classified into one data feature group; when there are multiple identical or similar initial data features, the identical or similar initial data features can be classified into one data feature group.
[0078] S202: Determine attribute data features corresponding to each data feature group according to all initial data features in each data feature group.
[0079] S203: Perform dimensionality reduction processing on each of the attribute data features to obtain new data features corresponding to each of the initial data features.
[0080] Specifically, after performing feature clustering on each of the initial data features to obtain at least one data feature group, the mean or mode of the initial data features in each of the data feature groups is determined, and the mean or mode in each of the data feature groups is used as the attribute data feature of the data feature group. The attribute data features can then be subjected to dimensionality reduction processing by algorithms such as PCA to obtain new data features corresponding to each of the initial data features.
[0081] S204: Perform feature mapping on the initial data features and the newly added data features to obtain target data features corresponding to the initial data features.
[0082] Specifically, after the dimension reduction processing is performed on each of the attribute data features to obtain the newly added data features corresponding to each of the initial data features, feature mapping can be performed on the initial data features and the newly added data features corresponding to the initial data features, that is, on the basis of keeping the original initial data features unchanged, the newly added data features corresponding to the initial data features are additionally added to obtain the target data features. In this way, after the dimension reduction processing is performed on the group attribute features of the data feature group corresponding to the initial data features (such as the above-mentioned attribute data features) through the above steps, they are added to the initial data features, so that the dimension of the initial data features is increased, which can improve the accuracy of model training.
[0083] In one embodiment, the preset recognition model includes an abnormality type prediction module and an abnormality level prediction module; the inputting the target data feature into the preset recognition model including initial parameters to determine the predicted abnormality label of each of the initial data features includes:
[0084] The target data feature is input into the preset recognition model so as to perform feature processing on the target data feature through the abnormal type prediction module to obtain a feature classification vector corresponding to the target data feature and a predicted classification label.
[0085] It can be understood that the abnormal type module in the preset recognition model has a bidirectional neural network and a classification network. The bidirectional neural network can learn all the characteristic information in the target data features (since the target data features are derived from the sample data corresponding to the initial data features, and the sample data is the user's basic information, product purchase information or complaint information, etc., the characteristic information here refers to text information), so that the accuracy of abnormal type prediction can be improved; the classification network is used to classify the abnormal category of the target data features according to the vector output by the bidirectional neural network, and then obtain the predicted classification label. Among them, the predicted classification label is the abnormal type of the target data feature predicted by the emotion prediction module, such as in the shopping application scenario, determining whether it is a bad product, whether the user will complain, etc.; in the insurance risk control scenario, determining whether there is an accident, that is, the accident rate, the predicted classification label may be the same as or different from the product abnormal label.
[0086] Specifically, after determining the target data features corresponding to each initial data feature based on the initial data features and the newly added data features, the target data features are input into the abnormal type prediction module in the preset recognition model, and then the predicted classification label corresponding to the target data features and the feature classification vector are determined through the bidirectional neural network and classification network in the abnormal type prediction module. Among them, the feature classification vector is obtained by the abnormal type prediction module performing vector processing on the target data features.
[0087] The target data features and feature classification vectors are input into the abnormality level prediction module to obtain a predicted level label corresponding to the target data features.
[0088] It can be understood that in the process of the abnormality type prediction module predicting the abnormality type according to the target data characteristics, a feature classification vector for abnormality classification will be generated, and then the feature classification vector and the target data characteristics can be used to make predictions when determining the predicted level label of the target data characteristics, thereby improving the accuracy of the abnormality level prediction.
[0089] Specifically, after the target data feature is input into the preset recognition model, and the target data feature is feature processed by the abnormal type prediction module to obtain a feature classification vector and a predicted classification label corresponding to the target data feature, the feature classification vector obtained by the abnormal type prediction module based on the target data feature is obtained, and then the predicted level label corresponding to the target data feature is determined by the abnormal level prediction module according to the feature classification vector and the target data feature.
[0090] The predicted abnormal label is determined according to the predicted classification label and the predicted grade label.
[0091] Specifically, after the predicted classification label and the predicted grade label corresponding to the target data feature are determined, the predicted classification label and the predicted grade label can be associated and recorded as the predicted abnormality label corresponding to the target data feature.
[0092] In one embodiment, if Figure 4 As shown, a commodity recommendation method is provided, comprising:
[0093] S60: receiving a product recommendation instruction sent by a target user; the product recommendation instruction includes required product information.
[0094] It is understandable that the commodity recommendation instruction can be sent through the mobile terminal of the target user, or can be automatically generated after the target user sends the required commodity information, wherein the required commodity information represents the type of commodity that the target user needs to recommend.
[0095] S70: Determine preset product information matching the required product information from a preset product information storage platform and record it as product information to be screened; the preset product information storage platform includes at least one preset product information.
[0096] It can be understood that the preset product information storage platform includes at least one preset product information, which represents the specific information of the product (such as product specifications, type, etc.), product evaluation information, compensation record information, etc. One preset product information corresponds to one product. After receiving the product recommendation instruction sent by the target user, each preset product information can be matched with the required product information from the preset product information storage platform, so as to determine the preset product information that matches the required product information.
[0097] S80: Inputting the product information to be screened into a product anomaly prediction model to determine the product anomaly type and product anomaly level corresponding to each piece of product information to be screened.
[0098] Specifically, after determining the preset product information matching the demand product information in the preset product information storage platform and recording it as the product information to be screened, the product information to be screened is input into the product anomaly prediction model, so as to determine the product anomaly type and product anomaly level corresponding to each product information to be screened through the product anomaly prediction model. Among them, the product anomaly type indicates whether the product corresponding to the product information to be screened will be complained, that is, the complaint rate; the product anomaly level indicates the claim amount after the product corresponding to the product information to be screened is complained.
[0099] S90: Select recommended product information from all the product information to be screened according to the product abnormality type and the product abnormality level, and push the recommended product information to the target user.
[0100] Specifically, after the product information to be screened is input into the product anomaly prediction model, and the product anomaly type and product anomaly level corresponding to each of the product information to be screened are determined by the product anomaly prediction model, one or more recommended product information can be selected from all the product information to be screened according to the product anomaly type and product anomaly level corresponding to the product information to be screened, and the recommended product information can be pushed to the target user. For example, the product information to be screened with a product anomaly type of no anomaly can be selected as the recommended product information. If there is no product information to be screened with no anomaly, the product information to be screened with the lowest product anomaly level can be selected as the recommended product information.
[0101] In this embodiment, the product anomaly prediction model is used to determine the product anomaly type and product anomaly level of each product based on the product information to be screened. In this way, the products can be screened according to the product anomaly type and product anomaly level, so that products with high complaint rates or high complaint levels can be eliminated, thereby improving the accuracy of product recommendations.
[0102] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.
[0103] In one embodiment, a product anomaly prediction model training device is provided, and the product anomaly prediction model training device corresponds to the product anomaly prediction model training method in the above embodiment. Figure 5 As shown, the device includes a data feature acquisition module 10, a cold start processing module 20, a data anomaly prediction module 30, a prediction loss value determination module 40 and a model training module 50. Each functional module is described in detail as follows:
[0104] The data feature acquisition module 10 is used to acquire a data feature set; the data feature set includes at least one initial data feature and a commodity abnormality label corresponding to the initial data feature one by one;
[0105] A cold start processing module 20 is used to perform cold start processing on each of the initial data features to obtain a newly added data feature corresponding to each of the initial data features, and determine a target data feature corresponding to each of the initial data features based on the initial data features and the newly added data features;
[0106] A data anomaly prediction module 30, for inputting the target data features into a preset recognition model including initial parameters to determine a predicted anomaly label of each of the initial data features;
[0107] A predicted loss value determination module 40, configured to determine a predicted loss value corresponding to the preset recognition model according to the commodity abnormality label and the predicted abnormality label;
[0108] The model training module 50 is used to iteratively update the initial parameters in the preset recognition model when the predicted loss value does not reach the preset convergence condition, until the predicted loss value reaches the convergence condition, and record the preset recognition model after convergence as the commodity abnormality prediction model.
[0109] In one embodiment, the commodity anomaly prediction model training device further includes:
[0110] A sample data acquisition module, used to acquire a sample data set; the sample data set includes at least one sample data group; one of the sample data groups includes at least one sample data;
[0111] A data processing module, used for performing data processing on all the sample data to obtain initial data features corresponding to each of the sample data groups;
[0112] A label generation module, used to generate a commodity abnormality label corresponding to each of the initial data features according to all the sample data;
[0113] The data feature set construction module is used to generate the data feature set according to the initial data features and the commodity abnormality label.
[0114] In one embodiment, the data processing module includes:
[0115] A data processing unit, used for preprocessing all the sample data to obtain preprocessed data features corresponding to each of the sample data groups;
[0116] A normalization processing unit, used to perform normalization processing on each of the preprocessed data features to obtain a normalized data feature corresponding to each of the preprocessed data features;
[0117] The feature screening unit is used to perform feature screening on each of the normalized data features to obtain the initial data features corresponding to each of the sample data groups.
[0118] In one embodiment, if Figure 6 As shown, the cold start processing module 20 includes:
[0119] A feature clustering unit 201 is used to perform feature clustering on each of the initial data features to obtain at least one data feature group; one of the data feature groups includes at least one initial data feature;
[0120] An attribute feature confirmation unit 202, configured to determine an attribute data feature corresponding to each data feature group according to all initial data features in each data feature group;
[0121] A feature dimension reduction unit 203 is used to perform dimension reduction processing on each of the attribute data features to obtain a newly added data feature corresponding to each of the initial data features;
[0122] The feature mapping unit 204 is used to perform feature mapping on the initial data features and the newly added data features to obtain target data features corresponding to the initial data features.
[0123] In one embodiment, the data anomaly prediction module 30 includes:
[0124] A feature processing unit, used for inputting the target data feature into the preset recognition model, so as to perform feature processing on the target data feature through the abnormal type prediction module, and obtain a feature classification vector corresponding to the target data feature and a predicted classification label;
[0125] An abnormality level prediction unit, used for inputting the target data feature and the feature classification vector into the abnormality level prediction module to obtain a prediction level label corresponding to the target data feature;
[0126] The abnormal label generating unit is used to determine the predicted abnormal label according to the predicted classification label and the predicted grade label.
[0127] For the specific limitations of the commodity anomaly prediction model training device, please refer to the limitations of the commodity anomaly prediction model training method above, which will not be repeated here. Each module in the above-mentioned commodity anomaly prediction model training device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0128] In one embodiment, if Figure 7 As shown, a commodity recommendation device is provided, comprising:
[0129] The instruction receiving module 60 is used to receive a product recommendation instruction sent by a target user; the product recommendation instruction includes the required product information;
[0130] The product matching module 70 is used to determine the preset product information matching the required product information from the preset product information storage platform and record it as the product information to be screened; the preset product information storage platform includes at least one preset product information;
[0131] The commodity prediction module 80 is used to input the commodity information to be screened into the commodity anomaly prediction model to determine the commodity anomaly type and commodity anomaly level corresponding to each commodity information to be screened;
[0132] The product recommendation module 90 is used to select recommended product information from all the product information to be screened according to the product anomaly type and the product anomaly level, and push the recommended product information to the target user.
[0133] For the specific definition of the product recommendation device, please refer to the definition of the product recommendation method above, which will not be repeated here. Each module in the above-mentioned product recommendation device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0134] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the commodity anomaly prediction model training method or the commodity recommendation method in the above-mentioned embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a commodity anomaly prediction model training method is implemented, or when the computer program is executed by the processor, a commodity recommendation method is implemented.
[0135] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the product anomaly prediction model training method in the above embodiment is implemented, or when the processor executes the computer program, the product recommendation method in the above embodiment is implemented.
[0136] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the product anomaly prediction model training method in the above embodiment is implemented, or when the computer program is executed by a processor, the product recommendation method in the above embodiment is implemented.
[0137] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0138] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0139] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.
Claims
1. A method for training a commodity anomaly prediction model. It is characterized in that include: Acquire a data feature set; the data feature set includes at least one initial data feature and a commodity abnormality label corresponding to the initial data feature one by one; Performing cold start processing on each of the initial data features to obtain newly added data features corresponding to each of the initial data features, and determining target data features corresponding to each of the initial data features based on the initial data features and the newly added data features; Inputting the target data features into a preset recognition model including initial parameters to determine a predicted anomaly label for each of the initial data features; Determining a predicted loss value corresponding to the preset recognition model according to the product abnormality label and the predicted abnormality label; When the predicted loss value does not reach the preset convergence condition, iteratively updating the initial parameters in the preset recognition model until the predicted loss value reaches the convergence condition, and recording the preset recognition model after convergence as the commodity abnormality prediction model; Before obtaining the data feature set, the following steps are included: Acquire at least one sample data group; one of the sample data groups includes at least one sample data; Performing data processing on all the sample data to obtain initial data features corresponding to each of the sample data groups; Generate a commodity abnormality label corresponding to each of the initial data features according to all the sample data; Generate the data feature set according to the initial data features and the commodity abnormality label; The step of performing data processing on all the sample data to obtain initial data features corresponding to each of the sample data groups includes: Preprocessing all the sample data to obtain preprocessed data features corresponding to each of the sample data groups; Normalizing each of the preprocessed data features to obtain a normalized data feature corresponding to each of the preprocessed data features; Performing feature screening on each of the normalized data features to obtain the initial data features corresponding to each of the sample data groups; The cold start processing is performed on each of the initial data features to obtain a newly added data feature corresponding to each of the initial data features, and the target data feature corresponding to each of the initial data features is determined according to the initial data features and the newly added data features, including: Performing feature clustering on each of the initial data features to obtain at least one data feature group; one of the data feature groups includes at least one initial data feature; Determine the attribute data features corresponding to each data feature group according to all the initial data features in each data feature group; Performing dimensionality reduction processing on each of the attribute data features to obtain newly added data features corresponding to each of the initial data features; Feature mapping is performed on the initial data features and the newly added data features to obtain target data features corresponding to each initial data feature.
2. The commodity abnormality prediction model training method according to claim 1, It is characterized in that The preset recognition model includes an abnormal type prediction module and an abnormal level prediction module; the inputting the target data feature into the preset recognition model including initial parameters to determine the predicted abnormal label of each of the initial data features includes: Inputting the target data feature into the preset recognition model, so as to perform feature processing on the target data feature through the abnormal type prediction module to obtain a feature classification vector and a predicted classification label corresponding to the target data feature; Inputting the target data features and feature classification vectors into the abnormality level prediction module to obtain a prediction level label corresponding to the target data features; The predicted abnormal label is determined according to the predicted classification label and the predicted grade label.
3. A product recommendation method, It is characterized in that include: Receiving a product recommendation instruction sent by a target user; the product recommendation instruction includes required product information; Determine, from a preset commodity information storage platform, preset commodity information matching the required commodity information and record it as commodity information to be screened; The preset commodity information storage platform includes at least one preset commodity information; Inputting the product information to be screened into a product anomaly prediction model to determine the product anomaly type and product anomaly level corresponding to each of the product information to be screened; the product anomaly prediction model is obtained based on the product anomaly prediction model training method according to claim 1 or 2; According to the product abnormality type and the product abnormality level, recommended product information is selected from all the product information to be screened, and the recommended product information is pushed to the target user.
4. A commodity anomaly prediction model training device, It is characterized in that The commodity anomaly prediction model training device is used to execute the commodity anomaly prediction model training method according to claim 1 or 2, comprising: A data feature acquisition module, used to acquire a data feature set; the data feature set includes at least one initial data feature and a commodity abnormality label corresponding to the initial data feature one by one; A cold start processing module, used to perform cold start processing on each of the initial data features to obtain a newly added data feature corresponding to each of the initial data features, and determine a target data feature corresponding to each of the initial data features based on the initial data features and the newly added data features; A data anomaly prediction module, used for inputting the target data features into a preset recognition model including initial parameters to determine a predicted anomaly label corresponding to each of the initial data features; A predicted loss value determination module, used to determine a predicted loss value corresponding to the preset recognition model according to the commodity abnormality label and the predicted abnormality label; The model training module is used to iteratively update the initial parameters in the preset recognition model when the predicted loss value does not reach the preset convergence condition, until the predicted loss value reaches the convergence condition, and record the preset recognition model after convergence as the commodity abnormality prediction model.
5. A product recommendation device, It is characterized in that include: An instruction receiving module, used to receive a product recommendation instruction sent by a target user; the product recommendation instruction includes required product information; A product matching module, used to determine preset product information matching the required product information from a preset product information storage platform and record it as product information to be screened; the preset product information storage platform includes at least one preset product information; A commodity prediction module, used for inputting the commodity information to be screened into a commodity anomaly prediction model to determine the commodity anomaly type and commodity anomaly level corresponding to each commodity information to be screened; the commodity anomaly prediction model is obtained based on the commodity anomaly prediction model training method according to claim 1 or 2; The product recommendation module is used to select recommended product information from all the product information to be screened according to the product anomaly type and the product anomaly level, and push the recommended product information to the target user.
6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the product anomaly prediction model training method as described in claim 1 or 2 is implemented, or when the processor executes the computer program, the product recommendation method as described in claim 3 is implemented.
7. A computer-readable storage medium storing a computer program, It is characterized in that When the computer program is executed by a processor, the product anomaly prediction model training method as described in claim 1 or 2 is implemented, or when the computer program is executed by a processor, the product recommendation method as described in claim 3 is implemented.
