CN114610992A - Depolarization recommendation method and device for intelligent workshop product information - Google Patents
Depolarization recommendation method and device for intelligent workshop product information Download PDFInfo
- Publication number
- CN114610992A CN114610992A CN202210203220.9A CN202210203220A CN114610992A CN 114610992 A CN114610992 A CN 114610992A CN 202210203220 A CN202210203220 A CN 202210203220A CN 114610992 A CN114610992 A CN 114610992A
- Authority
- CN
- China
- Prior art keywords
- recommendation
- training
- product information
- attribute
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000028161 membrane depolarization Effects 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 72
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000005457 optimization Methods 0.000 claims abstract description 12
- 230000000694 effects Effects 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 3
- 230000002441 reversible effect Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 2
- 230000008092 positive effect Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 description 9
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 230000001755 vocal effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Manufacturing & Machinery (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a method and a device for depolarization recommendation facing intelligent workshop product information, wherein the method comprises the following steps: (1) acquiring an Adult data set as a training data set of a product information recommendation model, extracting sensitive samples from the training data set and determining sensitive attributes; (2) carrying out depolarization processing on the sensitive attribute influence on the training sample by adopting a SHAP interpreter and orthogonalization operation to obtain a depolarized sample vector; (3) training the constructed product recommendation model by using the de-biased sample vector to obtain a product recommendation model without the influence of the sensitive attribute; (4) carrying out optimization training on the product recommendation model trained in the step (3) by utilizing the collected various workshop product information to obtain a personalized product recommendation model adapted to the various workshop product information; (5) the individual product recommendation model is used for carrying out depolarization recommendation on the product information, and the recommendation and depolarization effect on the workshop product information can be achieved.
Description
Technical Field
The invention belongs to the field of product information recommendation, and particularly relates to a depolarization recommendation method for intelligent workshop product information.
Background
The intelligent workshop management system is a set of combination of various software and hardware, the whole improvement of the production level of products is taken as a core, the numerical control automation equipment is interconnected and intercommunicated through a network and a software management system, and the intelligent workshop can improve the capacity of production equipment, the logistics capacity of the workshop and the energy management capacity of the workshop.
The intelligent workshop product information recommendation is a customized service realized for a client, and deviation and unfairness in a recommendation system are inherent with the birth of a deep learning algorithm and are not artificially and deliberately generated. For example, in terms of aesthetics, the mindset of cities and the view of rural areas are difficult to coexist. Therefore, sensitive attributes causing unfairness should be removed in the intelligent workshop product information recommendation, and fairness of recommendation results is guaranteed.
Patent document CN103268344A discloses a method for matching advertisement and page without position bias effect, which comprises the following steps: a. determining the display positions, the display times and the click times of all advertisements in each search page; b. by utilizing a statistical method, the influence of the position of the page where the advertisement is positioned on the advertisement clicking can be eliminated, and the inherent clicking rate of each advertisement published on each page is calculated; c. calculating the similarity relation between the pages related to the advertisement according to the inherent click rate of the pages and the advertisement; d. and matching a target page to be recommended with a proper advertisement according to similar other pages. The patent application adjusts the click rate of a specific advertisement under a specific page according to the advertisement display position so as to realize recommendation of page matching advertisements.
Patent document CN114037321A discloses a fairness-oriented crowdsourcing tester recommendation method, which includes expressing requirement description of a crowdsourcing task as descriptive term vector; predicting the defect detection probability of each person to be recommended in the crowdsourcing task according to the descriptive term vector and the attribute information of the person to be recommended; and setting a plurality of fairness targets including the maximum defect discovery probability and the minimum recommendation frequency difference, performing multi-objective optimization, and obtaining the recommendation result of crowdsourcing testers. The invention can reduce the unfairness problem and generate fairer recommendation.
The above two patent documents can realize unfairness problem in the recommendation process of the traditional algorithm, but cannot solve the unfairness problem in the recommendation process based on the deep learning model.
Disclosure of Invention
In view of the above, the present invention provides a method for depolarization recommendation of intelligent workshop product information, which achieves the effect of recommendation and depolarization of the workshop product information by using a SHAP interpreter and orthogonalization operation.
In order to achieve the above object, the embodiments provide the following technical solutions:
a depolarization method for recommending intelligent workshop product information comprises the following steps:
(1) acquiring an Adult data set as a training data set of a product information recommendation model, extracting a sensitive sample from the training data set and determining a sensitive attribute;
(2) carrying out depolarization processing on the sensitive attribute influence on the training sample by adopting a SHAP interpreter and orthogonalization operation to obtain a depolarized sample vector;
(3) training the constructed product recommendation model by using the de-biased sample vector to obtain a product recommendation model without the influence of the sensitive attribute;
(4) carrying out optimization training on the product recommendation model trained in the step (3) by utilizing the collected various workshop product information to obtain a personalized product recommendation model adapted to the various workshop product information;
(5) and performing depolarization recommendation on the product information by using the personalized product recommendation model.
In step (1) of one embodiment, extracting sensitive samples from the training dataset and determining sensitive attributes includes:
judging when x is satisfiedA≠x′A,xNA=x′NAWhen θ (x) ≠ θ (x'), i.e. when the attribute value x of the attribute A of the data sample xAAttribute value x 'of Attribute A with data sample x'ANot equal, attribute value x of other attribute NA than attribute A of data sample xNANOT-ORDER TO DATA SAMPLE XAttribute value x 'of other Attribute NA of Property A'NAAnd if the recommendation result of the recommendation model theta to the data sample x is not equal to the recommendation result to the data sample x ', the attribute A is considered as the sensitive attribute, (x, x'A) Is a sensitive sample.
In step (2) of an embodiment, a SHAP interpreter and an orthogonalization operation are used to perform a deskewing process on training samples for sensitive attribute influence, which includes:
firstly, detecting and obtaining a characteristic vector x which greatly contributes to the recommendation result of the model in a training sample by adopting a SHAP interpreterimpSimultaneously obtaining the feature vector x of the sensitive attributeA(ii) a Then, the feature vector x is divided intoimpAnd the feature vector xAAnd performing orthogonal calculation to eliminate the influence of the sensitive attribute on the recommendation result to obtain a sample vector after depolarization.
In one embodiment, the detection of the feature vector x which greatly contributes to the model recommendation result in the training sample by using the SHAP interpreter is obtainedimpThe method comprises the following steps:
suppose that the ith training sample xiIs xikModel pair training sample xiPredicted value of yiBase line y of the model as the mean of the target variables of all training samplesbaseThen the SHAP value obeys the following equation:
yi=ybase+f(xi1)+f(xi2)+…+f(xik)
wherein, f (x)ik) Is xikThe SHAP value of (1), i.e. to the final predicted value yiWhen f (x) is the contribution value ofik)>0, illustrating feature xikThe predicted value is improved, and a forward effect is achieved; on the contrary, the feature x is explainedikThe reduction of the predicted value is reduced, and the reverse effect is achieved.
In step (3) of one embodiment, the product recommendation model is constructed as a fully-connected neural network composed of fully-connected layers, each layer employing an activation function ReLU or an activation function SoftMax.
In step (3) and step) (4) of one embodiment, the loss function used in training the product recommendation model is a cross-entropy loss function, and the optimizer used uses an Adam optimizer.
In one embodiment, the method further includes the step (3') of extracting data from the result data set as a test sample, testing the product recommendation model trained in the step (3) by using the test sample, and when the test result of the test sample meets the fairness assessment index chance equality, considering that the trained product recommendation model reaches fairness, wherein the chance equality formula is as follows:
wherein,and (3) representing a prediction result, wherein A is a sensitive attribute, and when the results on the left side and the right side of the formula are equal or the difference is within a threshold range, the product recommendation model is considered to achieve the depolarization effect.
In order to achieve the above object, an embodiment of the present invention further provides a depolarization recommendation apparatus for intelligent workshop product information recommendation, including:
the sensitive attribute determining module is used for acquiring an result data set as a training data set of the product information recommendation model, extracting a sensitive sample from the training data set and determining a sensitive attribute;
the depolarization processing module is used for carrying out depolarization processing on the sensitive attribute influence on the training samples by adopting a SHAP interpreter and orthogonalization operation to obtain a depolarization sample vector;
the model training module is used for training the constructed product recommendation model by utilizing the depolarized sample vector to obtain a product recommendation model without the influence of the sensitive attribute;
the model retraining module is used for performing re-optimization training on the product recommendation model trained by the model training module by utilizing the collected various workshop product information to obtain an individualized product recommendation model adaptive to various workshop product information;
and the depolarization recommending module is used for carrying out depolarization recommendation on the product information by utilizing the personalized product recommending model.
In order to achieve the above object, an embodiment of the present invention further provides a device for removing bias recommendation for intelligent workshop product information recommendation, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the method for removing bias recommendation for intelligent workshop product information when executing the computer program.
Compared with the prior art, the method has the beneficial effects that at least:
extracting sensitive attributes influencing a recommendation result in the recommendation task based on an Adult data set, then utilizing the feature vector of the sensitive attributes to be orthogonal with the feature vector which is obtained through a SHAP interpreter and has an influence on the recommendation result so as to remove the influence of the sensitive attributes on the recommendation result, and then utilizing the de-biased sample vector to train a product recommendation model so as to achieve the de-bias effect of the product recommendation model on product recommendation. Meanwhile, the recommendation effect of various workshop product information is improved through the re-optimization process of the product recommendation model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart of a method for unbiased recommendation for intelligent workshop product information according to an embodiment;
fig. 2 is a schematic structural diagram of a depolarization recommendation device for intelligent workshop product information according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
FIG. 1 is a flowchart of a method for depolarization recommendation for intelligent workshop product information according to an embodiment. As shown in fig. 1, the method for removing bias recommendation for intelligent workshop product information includes the following steps:
step 1, acquiring an Adult data set as a training data set of a product information recommendation model, extracting a sensitive sample from the training data set and determining a sensitive attribute.
In the embodiment, an Adult data set is used as a training data set for intelligent workshop product information recommendation. The result of the method is that the result of the Adult data set is that the Adult data set comprises 48,842 continuous or discrete sample samples with abundant information, and the Adult data set comprises 14 attribute categories of age, work, academic calendar, occupation, gender and the like, wherein 6 continuous variables and 8 noun attribute variables are used as sensitive attributes, and one of the categories is used as a sensitive attribute to cause unfairness in the product recommendation process.
In the embodiment, the sensitive attribute is considered to be essentially irrelevant to the recommendation problem, and should not influence the attribute of the recommendation result. Based on this, the way to extract sensitive samples from the training dataset and determine the sensitive attributes includes: judging when x is satisfiedA≠x′A,xNA=x′NAWhen θ (x) ≠ θ (x'), i.e., when the attribute value x of the attribute A of the data sample x isAAttribute value x 'of Attribute A with data sample x'ANot equal, attribute value x of other attribute NA than attribute A of data sample xNAAttribute value x 'of other Attribute NA other than Attribute A for data sample x'NAAnd if the result of recommending the data sample x by the recommending model theta is not equal to the result of recommending the data sample x ', the attribute A is considered as a sensitive attribute, (x, x'A) Is a sensitive sample. The determined sensitivity attribute may be work, academic, vocational, etc., and a particular sensitivity attribute is to be associated with a particular recommended task.
And 2, carrying out depolarization processing on the sensitive attribute influence on the training sample by adopting a SHAP interpreter and orthogonalization operation to obtain a depolarized sample vector.
In an embodiment, performing a depolarization process on a training sample by using a SHAP interpreter and an orthogonalization operation includes:
firstly, detecting and obtaining a characteristic vector x which greatly contributes to the recommendation result of the model in a training sample by adopting a SHAP interpreterimp. Namely, the importance of each feature vector to the model prediction result is measured by using SHAP, which comprises the following steps:
suppose that the ith training sample xiIs xikModel pair training sample xiPredicted value of yiBase line y of the model as the mean of the target variables of all training samplesbaseThen the SHAP value obeys the following equation:
yi=ybase+f(xi1)+f(xi2)+…+f(xik)
wherein, f (x)ik) Is xikThe SHAP value of (1), i.e. to the final predicted value yiWhen f (x) is the contribution value ofik)>0, illustrating feature xikThe predicted value is improved, and a positive effect is achieved; on the contrary, the feature x is explainedikThe method reduces the reduction of the predicted value, plays a reverse role, and can find the SHAP value composition feature vector x contributing to the model recommendation result greatly based on the modeimp。
It should be noted that the model used in the process of obtaining the recommendation result for the model in the training sample may be a model different from other models except the product recommendation model, and the model needs to implement a product recommendation task to obtain the recommendation result expressed by the predicted value.
Then, a feature vector x of the sensitive attribute is obtainedA. In the embodiment, the sensitive attribute is subjected to binarization processing and is encoded to obtain a feature vector x corresponding to the sensitive attributeA。
Finally, the feature vector x is divided intoimpAnd the feature vector xAAnd performing orthogonal calculation to eliminate the influence of the sensitive attribute on the recommendation result to obtain a sample vector after depolarization.
And 3, training the constructed product recommendation model by using the depolarized sample vector to obtain the product recommendation model without the influence of the sensitive attribute.
In the embodiment, the constructed product recommendation model is a fully-connected neural network composed of fully-connected layers, and each fully-connected layer adopts an activation function ReLU or an activation function SoftMax. In one embodiment, a five-layer fully-connected neural network is constructed to train the model. The number of the neurons of each layer of the fully-connected neural network is respectively 64, 32, 16, 8 and 2, and the activation function of each layer of the first four layers is ReLU. And the full connection layer uses SoftMax to carry out classification identification on the data for the activation function.
And (5) de-biasing the sample vector. The product recommendation model is input into the product recommendation model for training, and the purpose is to improve the fairness of the product recommendation model. The size of the training batch is 100, a preheating learning rate strategy is adopted in the training stage, and Adam is adopted for optimization by the optimizer. The loss functions of the training process all use a cross entropy loss function (binary _ cross). And after training, obtaining a product recommendation model without the influence of the sensitive attribute.
And 4, extracting data from the result data set as a test sample, and testing the product recommendation model trained in the step 3 by using the test sample.
In the embodiment, a part of data is extracted from the result data set to serve as a test sample, the product recommendation model trained in the step (3) is tested by using the test sample, and when the test result of the test sample meets the fairness assessment index chance equality, the trained product recommendation model is considered to reach fairness, wherein the chance equality formula is as follows:
wherein,and representing a prediction result, wherein A is a sensitive attribute, A-0 represents one type of coded value of the sensitive attribute, A-1 represents another type of coded value of the sensitive attribute, Y represents a truth value variable, Y also represents a truth label value, and Pr () represents a fairness evaluation index function.
When the results of the left side and the right side of the above formula are equal or the difference is within the threshold range, the product recommendation model is considered to achieve the depolarization effect,
and 5, performing optimization training on the product recommendation model trained in the step 3 by using the collected various workshop product information to obtain a personalized product recommendation model adaptive to the various workshop product information.
The product classification model constructed in the steps 1-4 is trained based on the example sample of the Adult data set, so that the influence of universality on sensitive attributes is eliminated, but the model is required to be subjected to re-optimization training by each type of workshop product information, so that the recommendation accuracy of the product classification model on the specific type of workshop product information is improved. In the embodiment, when the product recommendation model trained in the step 3 is subjected to re-optimization training by using the collected various workshop product information, a cross entropy loss function and an Adam optimizer are still adopted, so that the personalized product recommendation model adaptive to various product information is obtained.
And 6, performing depolarization recommendation on the product information by using the personalized product recommendation model.
In the embodiment, the information of the workshop products to be recommended is input into the personalized product recommendation model, and a predicted value is obtained through calculation and is used as a recommendation result.
Fig. 2 is a schematic structural diagram of a depolarization recommendation device for intelligent workshop product information according to an embodiment. As shown in fig. 2, the depolarization recommendation apparatus provided in the embodiment includes:
the sensitive attribute determining module is used for acquiring an Adult data set as a training data set of the product information recommendation model, extracting a sensitive sample from the training data set and determining a sensitive attribute;
the depolarization processing module is used for carrying out depolarization processing on the sensitive attribute influence on the training samples by adopting a SHAP interpreter and orthogonalization operation to obtain a depolarization sample vector;
the model training module is used for training the constructed product recommendation model by utilizing the depolarized sample vector to obtain a product recommendation model without the influence of the sensitive attribute;
the model retraining module is used for performing re-optimization training on the product recommendation model trained by the model training module by utilizing the collected various workshop product information to obtain an individualized product recommendation model adaptive to various workshop product information;
and the depolarization recommending module is used for carrying out depolarization recommendation on the product information by utilizing the personalized product recommending model.
It should be noted that, when performing a depolarization recommendation, the depolarization recommendation apparatus provided in the above embodiment should be exemplified by the division of the above functional modules, and the above function distribution may be completed by different functional modules as needed, that is, the internal structure of the terminal or the server is divided into different functional modules to complete all or part of the above described functions. In addition, the depolarization recommendation device and the depolarization recommendation method provided by the above embodiments belong to the same concept, and specific implementation processes thereof are described in detail in the embodiments of the depolarization recommendation method, and are not described herein again.
The embodiment also provides a depolarization device for intelligent workshop product information recommendation, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor realizes the steps of the depolarization recommendation method for intelligent workshop product information when executing the computer program.
In practical applications, the memory may be a volatile memory at the near end, such as RAM, a non-volatile memory, such as ROM, FLASH, a floppy disk, a mechanical hard disk, etc., or a remote storage cloud. The processor can be a Central Processing Unit (CPU), a microprocessor unit (MPU), a Digital Signal Processor (DSP), or a Field Programmable Gate Array (FPGA), i.e., the depolarization recommendation of the intelligent workshop product information can be realized through the processors.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.
Claims (9)
1. A depolarization recommendation method for intelligent workshop product information is characterized by comprising the following steps:
(1) acquiring an Adult data set as a training data set of a product information recommendation model, extracting sensitive samples from the training data set and determining sensitive attributes;
(2) carrying out depolarization processing on the sensitive attribute influence on the training sample by adopting a SHAP interpreter and orthogonalization operation to obtain a depolarized sample vector;
(3) training the constructed product recommendation model by using the de-biased sample vector to obtain a product recommendation model without the influence of the sensitive attribute;
(4) carrying out optimization training on the product recommendation model trained in the step (3) by utilizing the collected various workshop product information to obtain a personalized product recommendation model adapted to the various workshop product information;
(5) and performing depolarization recommendation on the product information by using the personalized product recommendation model.
2. The method for unbinding the intelligent workshop product information recommendation according to claim 1, wherein in the step (1), the extracting of the sensitive samples from the training data set and the determining of the sensitive attributes comprise:
judging when x is satisfiedA≠x′A,xNA=x′NAWhen θ (x) ≠ θ (x'), i.e., when the attribute value x of the attribute A of the data sample x isAAttribute value x 'of Attribute A with data sample x'ANot equal, attribute value x of other attribute NA than attribute A of data sample xNAAttribute value x 'of other Attribute NA other than Attribute A for data sample x'NAAnd if the recommendation result of the recommendation model theta to the data sample x is not equal to the recommendation result to the data sample x ', the attribute A is considered as the sensitive attribute, (x, x'A) Is a sensitive sample.
3. The method for unbiasing intelligent workshop product information recommendation according to claim 1, wherein in the step (2), the training samples are subjected to unbiasing processing of sensitive attribute influence by using a SHAP interpreter and an orthogonalization operation, and the method comprises the following steps:
firstly, detecting and obtaining a characteristic vector x which greatly contributes to the recommendation result of the model in a training sample by adopting a SHAP interpreterimpSimultaneously obtaining the feature vector x of the sensitive attributeA(ii) a Then, the feature vector x is divided intoimpAnd the feature vector xAAnd performing orthogonal calculation to eliminate the influence of the sensitive attribute on the recommendation result to obtain a sample vector after depolarization.
4. The method for unbiased recommendation for intelligent plant product information as claimed in claim 3, wherein the feature vector x that contributes significantly to the recommendation result of the model in the training sample is detected by using the SHAP interpreterimpThe method comprises the following steps:
suppose that the ith training sample xiIs xikModel pair training sample xiPredicted value of yiBase line y of the model as the mean of the target variables of all training samplesbaseThen the SHAP value obeys the following equation:
yi=ybase+f(xi1)+f(xi2)+…+f(xik)
wherein, f (x)ik) Is xikThe SHAP value of (1), i.e. to the final predicted value yiWhen f (x) is the contribution value ofik)>0, illustrating feature xikThe predicted value is improved, and a positive effect is achieved; on the contrary, the feature x is explainedikThe reduction of the predicted value is reduced, and the reverse effect is achieved.
5. The method for unbiasing intelligent workshop product information recommendation according to claim 1, wherein in the step (3), the constructed product recommendation model is a fully-connected neural network composed of fully-connected layers, and each fully-connected layer adopts an activation function ReLU or an activation function SoftMax.
6. The method for unbiasing intelligent workshop product information recommendation according to claim 1, wherein in the step (3) and the step (4), the loss function adopted in training the product recommendation model is a cross entropy loss function, and an Adam optimizer is used as the adopted optimizer.
7. The method for unbiasing intelligent workshop product information recommendation according to claim 1, further comprising a step (3') of extracting data from the Adult data set as a test sample, testing the product recommendation model trained in the step (3) by using the test sample, and when the test result of the test sample meets the fairness evaluation index chance equality, considering that the trained product recommendation model achieves fairness, wherein the chance equality formula is as follows:
8. A remove inclined to one side recommendation device towards intelligent workshop product information recommendation includes:
the sensitive attribute determining module is used for acquiring an Adult data set as a training data set of the product information recommendation model, extracting a sensitive sample from the training data set and determining a sensitive attribute;
the depolarization processing module is used for carrying out depolarization processing on the sensitive attribute influence on the training samples by adopting a SHAP interpreter and orthogonalization operation to obtain a depolarization sample vector;
the model training module is used for training the constructed product recommendation model by utilizing the depolarized sample vector to obtain a product recommendation model without the influence of the sensitive attribute;
the model retraining module is used for performing re-optimization training on the product recommendation model trained by the model training module by utilizing the collected various workshop product information to obtain an individualized product recommendation model adaptive to various workshop product information;
and the depolarization recommending module is used for carrying out depolarization recommendation on the product information by utilizing the personalized product recommending model.
9. An intelligent workshop product information recommendation-oriented depolarization recommendation device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the intelligent workshop product information recommendation-oriented depolarization recommendation method steps of any one of claims 1 to 7 when executing the computer program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210203220.9A CN114610992A (en) | 2022-03-03 | 2022-03-03 | Depolarization recommendation method and device for intelligent workshop product information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210203220.9A CN114610992A (en) | 2022-03-03 | 2022-03-03 | Depolarization recommendation method and device for intelligent workshop product information |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114610992A true CN114610992A (en) | 2022-06-10 |
Family
ID=81861452
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210203220.9A Pending CN114610992A (en) | 2022-03-03 | 2022-03-03 | Depolarization recommendation method and device for intelligent workshop product information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114610992A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112184391A (en) * | 2020-10-16 | 2021-01-05 | 中国科学院计算技术研究所 | Recommendation model training method, medium, electronic device and recommendation model |
CN113298254A (en) * | 2021-06-10 | 2021-08-24 | 浙江工业大学 | Deskewing method and device for deep migration learning |
CN113361652A (en) * | 2021-07-09 | 2021-09-07 | 浙江工业大学 | Individual income prediction oriented depolarization method and device |
US20210287119A1 (en) * | 2020-03-12 | 2021-09-16 | Atb Financial | Systems and methods for mitigation bias in machine learning model output |
CN113434761A (en) * | 2021-06-25 | 2021-09-24 | 平安科技(深圳)有限公司 | Recommendation model training method and device, computer equipment and storage medium |
-
2022
- 2022-03-03 CN CN202210203220.9A patent/CN114610992A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210287119A1 (en) * | 2020-03-12 | 2021-09-16 | Atb Financial | Systems and methods for mitigation bias in machine learning model output |
CN112184391A (en) * | 2020-10-16 | 2021-01-05 | 中国科学院计算技术研究所 | Recommendation model training method, medium, electronic device and recommendation model |
CN113298254A (en) * | 2021-06-10 | 2021-08-24 | 浙江工业大学 | Deskewing method and device for deep migration learning |
CN113434761A (en) * | 2021-06-25 | 2021-09-24 | 平安科技(深圳)有限公司 | Recommendation model training method and device, computer equipment and storage medium |
CN113361652A (en) * | 2021-07-09 | 2021-09-07 | 浙江工业大学 | Individual income prediction oriented depolarization method and device |
Non-Patent Citations (3)
Title |
---|
HICKEY J M, DI STEFANO P G, VASILEIOU V: "Fairness by explicability and adversarial SHAP learning", LECTURE NOTES IN COMPUTER SCIENCE BOOK SERIES(LNAI), vol. 12459, 25 February 2021 (2021-02-25), pages 174 - 190 * |
蔡主希: "智能风控与反欺诈 体系算法与实践", 31 March 2021, 机械工业出版社, pages: 109 - 110 * |
陈晋音, 等: "面向深度学***性研究综述", 计算机研究与发展, vol. 58, no. 2, 28 February 2021 (2021-02-28), pages 264 - 280 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11334795B2 (en) | Automated and adaptive design and training of neural networks | |
CN108171209B (en) | Face age estimation method for metric learning based on convolutional neural network | |
CN111091175A (en) | Neural network model training method, neural network model classification method, neural network model training device and electronic equipment | |
CN111667022A (en) | User data processing method and device, computer equipment and storage medium | |
CN111028100A (en) | Refined short-term load prediction method, device and medium considering meteorological factors | |
CN111161814A (en) | DRGs automatic grouping method based on convolutional neural network | |
CN111881722B (en) | Cross-age face recognition method, system, device and storage medium | |
CN101893704A (en) | Rough set-based radar radiation source signal identification method | |
CN112529638B (en) | Service demand dynamic prediction method and system based on user classification and deep learning | |
CN110705600A (en) | Cross-correlation entropy based multi-depth learning model fusion method, terminal device and readable storage medium | |
CN111178196B (en) | Cell classification method, device and equipment | |
CN114139624A (en) | Method for mining time series data similarity information based on integrated model | |
CN116089708A (en) | Agricultural knowledge recommendation method and device | |
CN116805051A (en) | Double convolution dynamic domain adaptive equipment fault diagnosis method based on attention mechanism | |
CN115034315A (en) | Business processing method and device based on artificial intelligence, computer equipment and medium | |
CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
CN113298254A (en) | Deskewing method and device for deep migration learning | |
Zhao | The water potability prediction based on active support vector machine and artificial neural network | |
CN116340643A (en) | Object recommendation adjustment method and device, storage medium and electronic equipment | |
CN116451081A (en) | Data drift detection method, device, terminal and storage medium | |
CN114610992A (en) | Depolarization recommendation method and device for intelligent workshop product information | |
CN115345248A (en) | Deep learning-oriented data depolarization method and device | |
CN112529637B (en) | Service demand dynamic prediction method and system based on context awareness | |
CN115423091A (en) | Conditional antagonistic neural network training method, scene generation method and system | |
Li et al. | Time series classification with deep neural networks based on Hurst exponent analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |