CN112862519A - Sales anomaly identification method for retail data of electric business platform household appliances - Google Patents
Sales anomaly identification method for retail data of electric business platform household appliances Download PDFInfo
- Publication number
- CN112862519A CN112862519A CN202110073348.3A CN202110073348A CN112862519A CN 112862519 A CN112862519 A CN 112862519A CN 202110073348 A CN202110073348 A CN 202110073348A CN 112862519 A CN112862519 A CN 112862519A
- Authority
- CN
- China
- Prior art keywords
- data
- sales
- review
- model
- processing
- 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
Images
Classifications
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a sales quantity abnormity identification method aiming at retail data of an e-commerce platform household appliance, which comprises the following steps: repeatedly training by adopting a multi-layer neural network classification algorithm through long-term accumulated sub-channel data to obtain a highly available training set and a rule base; the system firstly carries out automatic processing, outputs a credible grading result, firstly carries out automatic review according to a review rule base and a flow according to credible grading, if the existing rule base cannot be identified, the review is carried out manually, the review rule base is updated manually according to the result of each review, whether a sample is added into a training set or not is selected, and when the same scene is encountered subsequently, the automatic processing is carried out according to setting. The invention acquires retail data of household appliances and consumer electronics through online acquisition, and utilizes big data analysis mining technology and machine learning algorithm to identify and clean abnormal sales volume and unit price, thereby providing an accurate industry consumption database for household appliance industry.
Description
Technical Field
The invention relates to a sales volume abnormity identification method, in particular to a sales volume abnormity identification method aiming at retail data of household appliances of an e-commerce platform.
Background
At present, the market does not exist basically in a method for quickly and effectively judging abnormal sales volume data. The main method is to manually check the data one by one through market information data, and the method has extremely low efficiency and is difficult to judge the accuracy of the market information data. The other mode is based on the majority theory or normal distribution, and the data which is obviously higher than the conventional data and does not accord with the normal distribution rule is removed. This approach is more efficient than the first method, but less accurate. Human intervention sales data is common for merchants to manufacture hot goods and the like in order to enhance PR influence for each brand. On one hand, each brand company wants to obtain real data of competitors, and meanwhile, artificially releases 'smoke' data, so that the brand company is used as a neutral authoritative data mechanism and provides timely, objective and accurate data for the market, and the brand company is the most core competitiveness of the company.
In the prior art, two implementation schemes exist, namely manual checking is mainly adopted; secondly, the system is automatically processed by setting some simple threshold values; the two modes respectively represent two extreme values, namely the data is relatively accurate, but the efficiency is low, the cost is high, and the timeliness is poor; the latter has fast processing speed, high efficiency, good timeliness but poor accuracy.
In conclusion, the invention designs a sales quantity abnormity identification method aiming at the retail data of the electric business platform household appliances.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a method for identifying abnormal sales volume of electric and commercial platform home appliance retail data, which acquires the retail data (core index data is sales volume and sales price) of home appliances and consumer electronics through online (electric and commercial platform) acquisition, and identifies and cleans abnormal sales volume and unit price by utilizing big data analysis mining technology and machine learning algorithm to provide an accurate industry consumption database for the home appliance industry; the method provides overall market analysis, market segment analysis and analysis of the proportion change of brand owners in different channels and regions for the household appliance and consumer electronics industry.
In order to achieve the purpose, the invention is realized by the following technical scheme: a sales quantity abnormity identification method aiming at retail data of electric business platform household appliances comprises the following steps: through long-term accumulated sub-channel data, a multi-layer neural network classification algorithm is adopted to obtain a highly available training set and a rule base through repeated training. The system firstly carries out automatic processing, outputs credible grading results (namely, various possible categories such as 0: normal, 1: review and 2: abnormal), firstly carries out automatic review according to a review rule base and a flow according to credible grading, if the existing rule base cannot be identified, carries out review manually, and if the existing rule base can not be identified, the review rule base is updated manually, whether a sample is added into a training set or not is selected, and when the same scene is encountered subsequently, the automatic processing can be carried out according to setting.
The sales volume abnormity identification system for the retail data of the electric business platform household appliances comprises a basic classifier, an intelligent selector and a manual recheck, wherein the basic classifier simulates a mode that human brains recognize objects to classify input. In order to improve the accuracy and robustness, a plurality of layers are set for analysis, wherein the first layer is an input layer and is used for carrying out primary classification, trimming and the like on data, the last layer is an output layer, and two hidden layers are arranged in the middle; the final classification result is normal, review and abnormal, so the classification number is set to 3, the maximum iteration number is set to 100, and the basic processing flow comprises the following steps: creating a forward feedback topology; calculating a gradient; and updating the weight value and the like.
The basic classifier comprises an original data structure, preprocessed label feature vector data, input original data and an output processing result, wherein the classification of the original data structure is marked as: 0: normal; 1: reviewing; 2: an anomaly; the pre-processing of the label feature vector data is to complete feature vector conversion through feature hash; the input original data is not classified, and after the input original data is processed by a system, a classification mark is added; the output processing result is processed by a multilayer neural network classification algorithm; description of classification identification: 0: normal; 1: reviewing; 2: and (6) abnormal.
And the intelligent selector distributes the data to a credible database or a manual processing database according to the processing result of the basic classifier and in combination with the rule knowledge base. The core of the data processing is data needing to be reviewed, abnormal data and credible data are filtered from the review data according to knowledge rules, the credible data are written into a credible library, the abnormal data are written into an abnormal library, and the data needing to be verified manually again are written into a manual review library. And synchronously updating the training data set by the credible data filtered from the review library. Each piece of data in the training data set has a time stamp, the updating scale of the training data set is counted regularly, and when the model is updated, the model is trained and updated again.
The rule knowledge base comprises:
1. low-price treatment: deleting data with the price lower than the base price according to the base price dictionary table of each platform, each brand and each category;
2. carrying out abnormal mass processing;
3. and analyzing abnormal change of the sales volume ratio of the marketable model.
The steps of processing the abnormal large amount are as follows:
the first step is as follows: integer processing
Firstly, marking data with sku sales of more than 1000, which can be evenly divided by 100, and unit price of which is more than or equal to average price;
then the processing sales: (taking the average value of sales of the SKU in 4 weeks) and (industry current period/industry previous period) and (random numbers between 0.9 and 1.1), comparing the calculation result with the original sales, if the calculation result is > (original sales 0.5), selecting the original sales and keeping two digits, and if no sales record exists in the comparison period, keeping the calculation result with the two digits and then taking the two digits as the sales.
The second step is that: abnormal change analysis of brand sales volume ratio
(1) First, brand ratios were analyzed as follows. Identifying an abnormal sales volume;
(2) brand (current/ring ratio-1) > 10%;
(3) inquiring 100 models of the industry, (the current period/the previous period-1) > 10%;
(4) the single price of the screened model is more than or equal to the mean price of the model at the current stage and 0.95, and the single price of the selected model is less than or equal to the mean price of the model at the previous stage and 0.5;
(5) if the screened machine type has a plurality of sales records, processing the records with the sales volume more than or equal to 100;
(6) the direct processing of no sale in the last period;
then the processing sales: (the former 4W brand + model mean value) ((the current industry stage/the previous industry stage)) (a random number between 0.9 and 1.1), comparing the calculation result with the original sales volume, if the calculation result is greater than the original sales volume 0.5, selecting the original sales volume to reserve two digits, and if no sales record exists in the comparison period, reserving the calculation result with the two digits to serve as the sales volume.
The third step: analysis of abnormal change of sale quantity ratio of marketable model
Firstly, the proportion of the popular model is analyzed according to the following conditions:
(1) the TOP100 model is screened without being processed, and simultaneously, a single record is more than or equal to 100 records
(2) The model is within TOP10 after being corrected, and the contribution rate of the current period is more than or equal to the average contribution rate of the model in the first four periods. Model contribution rate screening formula: (sales volume of the model in the current period/the current period scale) is not less than (sum of sales volumes of the models in the first four periods/sum of sales volumes of the classes in the first four periods ×. 2), and the contribution rate formula: the sale amount of the machine in the current period/the scale of the current period.
Then the processing sales: (the first 4W brand + model mean value) × (random number between 0.9 and 1.1), comparing the calculation result with the original sales volume, if the calculation result is greater than the original sales volume × 0.7, selecting (random value between 50 and the original sales volume) × 0.7 as the sales volume, and if no sales record exists in the comparison period, keeping two digits of the calculation result as the sales volume.
The manual review: the system adopts a mode of periodical manual review to review the data in the database.
The invention has the beneficial effects that: the method has high data processing efficiency, relatively accurate and objective data and is not influenced by information data with different calibers.
Drawings
The invention is described in detail below with reference to the drawings and the detailed description;
FIG. 1 is a diagram of the basic architecture of the system of the present invention.
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further described with the specific embodiments.
Referring to fig. 1, the following technical solutions are adopted in the present embodiment: a sales quantity abnormity identification method aiming at retail data of electric business platform household appliances comprises the following steps: through long-term accumulated sub-channel data, a multi-layer neural network classification algorithm is adopted to obtain a highly available training set and a rule base through repeated training. The system firstly carries out automatic processing, outputs credible grading results (namely, various possible categories such as 0: normal, 1: review and 2: abnormal), firstly carries out automatic review according to a review rule base and a flow according to credible grading, if the existing rule base cannot be identified, carries out review manually, and if the existing rule base can not be identified, the review rule base is updated manually, whether a sample is added into a training set or not is selected, and when the same scene is encountered subsequently, the automatic processing can be carried out according to setting.
A sales quantity abnormity identification system for retail data of electric business platform household appliances comprises a basic classifier, an intelligent selector and a manual recheck.
Basic classifier
The specific embodiment adopts a Multiple Layer Perceptron Classifier algorithm. And training and modeling the preprocessed vector data. The classifier classifies the input in a way that simulates the human brain in recognizing objects. In order to improve the accuracy and robustness, the scheme sets multiple layers for analysis, wherein the first layer is an input layer, data is subjected to primary classification, trimming and the like, the last layer is an output layer, and two hidden layers are arranged in the middle.
In the present embodiment, the final classification result is normal, review, and abnormal, so the classification number is set to 3, and the maximum iteration number is set to 100. the basic processing flow includes: creating a forward feedback topology; calculating a gradient; and updating the weight value and the like.
(1) Raw data structure
The labeled raw training data (snippets) are as follows, with class labels accounting for: 0: normal; 1: reviewing; 2: and (6) abnormal.
(2) The pre-processing into tag feature vector (Labeled point) data (fragments) is as follows:
the embodiment adopts a universal method facing high-radix class characteristics, and comprises the following steps: feature hashing (FeatureHasher) to complete the feature vector transformation. The number of newly generated features after the commonly used one-hot encoding will vary with the number of classes, and the number of newly generated features by the FeatureHasher method can be given artificially. Thus, the problem of feature redundancy or dimension explosion can be effectively solved.
Description of classification identification: 0: normal; 1: reviewing; 2: abnormality (S)
| Classification | Feature vector |
| 1 | (262144,[10634,114038,117355,134869,165267,167448,193139],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 1 | (262144,[19792,87434,107436,130477,134869,167448,175706],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[10634,93156,107436,122329,139243,167448,244140],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 2 | (262144,[87434,96644,122283,152131,167448,194321,213208],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 1 | (262144,[10634,122329,134869,166080,167448,194321,231787],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[9286,10634,17925,139243,167448,167459,169310],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 1 | (262144,[55636,122329,134869,149248,167448,224859,244140],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 2 | (262144,[10634,93156,122329,152131,167448,187527,244140],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 1 | (262144,[122283,122329,134869,167448,170353,174510,207670],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[3589,10634,88797,139243,162788,167448,193139],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[10634,13692,122329,139243,167448,193147,231379],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[13692,139243,162788,167448,173358,175706,193139],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 2 | (262144,[55636,65698,100505,109514,152131,167448,238475],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 1 | (262144,[77325,80390,122283,134869,135644,167448,167516],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[10634,31824,129235,139243,167448,188302,250936],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 2 | (262144,[10634,63096,80390,107202,152131,167448,192637],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 1 | (262144,[10634,17428,51247,133509,134869,167448,194835],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 1 | (262144,[62408,94159,122283,134869,167448,197375,215802],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[10634,117356,132154,139243,151794,167448,240635],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
| 0 | (262144,[779,35317,121928,139243,167448,183601,203269],[1.0,1.0,1.0,1.0,1.0,1.0,1.0]) |
(3) Inputting raw data
Example fragment data. The input classification original data is not classified, and after the input classification original data is processed by the system, a classification mark is added.
(4) Outputting the processing result
Example fragment data. And (5) processing the result through a multi-layer neural network classification algorithm. Description of classification identification: 0: normal; 1: reviewing; 2: and (6) abnormal.
(II) Intelligent selector
And the intelligent selector distributes the data to a credible database or a manual processing database according to the processing result of the basic classifier and by combining the rule knowledge base. The core of the data processing is data needing to be reviewed, abnormal data and credible data are filtered from the review data according to knowledge rules, the credible data are written into a credible library, the abnormal data are written into an abnormal library, and the data needing to be verified manually again are written into a manual review library. And synchronously updating the training data set by the credible data filtered from the review library. Each piece of data in the training data set has a time stamp, the updating scale of the training data set is counted regularly, and when the model is updated, the model is trained and updated again.
The knowledge rule base is divided into the following parts:
one, low cost treatment
And deleting the data with the price lower than the base price according to the base price dictionary table of each platform, each brand and each category.
Second, exception bulk handling
The first step is as follows: integer processing
Firstly, marking data with sku sales of more than 1000, which can be evenly divided by 100, and unit price of which is more than or equal to average price;
then the processing sales: (taking the average value of sales of the SKU in 4 weeks) and (industry current period/industry previous period) and (random numbers between 0.9 and 1.1), comparing the calculation result with the original sales, if the calculation result is > (original sales 0.5), selecting the original sales and keeping two digits, and if no sales record exists in the comparison period, keeping the calculation result with the two digits and then taking the two digits as the sales.
The second step is that: abnormal change analysis of brand sales volume ratio
First, brand ratios were analyzed as follows. An abnormal sales volume is identified.
1. Brand (current/ring ratio-1) > 10%;
2. inquiring 100 models of the industry, (the current period/the previous period-1) > 10%;
3. the single price of the screened model is more than or equal to the mean price of the model at the current stage and 0.95, and the single price of the selected model is less than or equal to the mean price of the model at the previous stage and 0.5;
4. if the screened machine type has a plurality of sales records, processing the records with the sales volume more than or equal to 100;
5. the direct processing of no sale in the last period;
then the processing sales: (the former 4W brand + model mean value) ((the current industry stage/the previous industry stage)) (a random number between 0.9 and 1.1), comparing the calculation result with the original sales volume, if the calculation result is greater than the original sales volume 0.5, selecting the original sales volume to reserve two digits, and if no sales record exists in the comparison period, reserving the calculation result with the two digits to serve as the sales volume.
The third step: analysis of abnormal change of sale quantity ratio of marketable model
First, the proportion of the mass market model was analyzed under the following conditions.
1) The TOP100 model is screened without being processed, and simultaneously, a single record is more than or equal to 100 records;
2) the model is within TOP10 after being corrected, and the contribution rate of the current period is more than or equal to the average contribution rate of the model in the first four periods. Model contribution rate screening formula: (sales volume of the model in the current period/the current period scale) is not less than (sum of sales volumes of the models in the first four periods/sum of sales volumes of the classes in the first four periods ×. 2), and the contribution rate formula: the sale amount of the machine in the current period/the scale of the current period.
Then the processing sales: (the first 4W brand + model mean value) × (random number between 0.9 and 1.1), comparing the calculation result with the original sales volume, if the calculation result is greater than the original sales volume × 0.7, selecting (random value between 50 and the original sales volume) × 0.7 as the sales volume, and if no sales record exists in the comparison period, keeping two digits of the calculation result as the sales volume.
(III) Manual review
The system adopts a mode of periodical manual review to review the data in the database.
The multi-layer neural network classification algorithm in the present embodiment is as follows:
1) all weights in the network are initialized randomly.
2) For each training example, the following operations are performed:
A) and calculating sequentially from front to back according to the input of the example to obtain the output of each unit of the output layer. The error term for each cell of each layer is then calculated back starting from the output layer.
B) For each cell k of the output layer, its error term is calculated:
δk＝ok(1-ok)(tk-ok)
C) for each hidden unit h in the network, its error term is calculated:
D) updating each weight:
wji＝wji+ηδjxji
△wji＝ηδjxjicalled weight update rule
Description of the symbols:
xji: input from node i to node jIn, wjiRepresenting the corresponding weight.
outputs: representing the set of output level nodes.
The data processing efficiency of the embodiment is high, the data is relatively objective and accurate, and the data processing method is not influenced by information data with different calibers.
The foregoing shows and describes the general principles and broad features of the present invention and advantages thereof. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (7)
1. A sales volume abnormity identification method aiming at retail data of electric business platform household appliances is characterized by comprising the following steps: repeatedly training by adopting a multi-layer neural network classification algorithm through long-term accumulated sub-channel data to obtain a highly available training set and a rule base; the system firstly carries out automatic processing, outputs a credible grading result, firstly carries out automatic review according to a review rule base and a flow according to credible grading, if the existing rule base cannot be identified, the review is carried out manually, the review rule base is updated manually according to the result of each review, whether a sample is added into a training set or not is selected, and when the same scene is encountered subsequently, the automatic processing is carried out according to setting.
2. A sales volume abnormity identification system aiming at retail data of electric business platform household appliances is characterized by comprising a basic classifier, an intelligent selector and an artificial recheck, wherein the basic classifier simulates the mode of recognizing objects by human brains to classify input; in order to improve the accuracy and robustness, a plurality of layers are set for analysis, wherein the first layer is an input layer, data are primarily classified and trimmed, the last layer is an output layer, and two hidden layers are arranged in the middle; the final classification result is normal, review and abnormal, so the classification number is set to 3, the maximum iteration number is set to 100, and the basic processing flow comprises the following steps: creating a forward feedback topology; calculating a gradient; and updating the weight value.
3. The method as claimed in claim 2, wherein the basic classifier comprises a raw data structure, preprocessed into tag feature vector data, input raw data and output processing results, and the raw data structure is classified as: 0: normal; 1: reviewing; 2: an anomaly; the pre-processing of the label feature vector data is to complete feature vector conversion through feature hash; the input original data is not classified, and after the input original data is processed by a system, a classification mark is added; the output processing result is processed by a multilayer neural network classification algorithm; description of classification identification: 0: normal; 1: reviewing; 2: and (6) abnormal.
4. The system for recognizing the sales anomaly of the retail data of the electric business platform household appliances according to the claim 2 is characterized in that the intelligent selector branches the data to a credible database or a manual processing database according to the processing result of the basic classifier and a rule knowledge base; the core of the data processing is data needing to be reviewed, abnormal data and credible data are filtered from the review data according to knowledge rules, the credible data are written into a credible library, the abnormal data are written into an abnormal library, and the data needing to be manually verified again are written into a manual review library; synchronously updating the training data set by the credible data filtered from the review library; each piece of data in the training data set has a time stamp, the updating scale of the training data set is counted regularly, and when the model is updated, the model is trained and updated again.
5. The system of claim 4, wherein the rule repository comprises:
(1) and low-price treatment: deleting data with the price lower than the base price according to the base price dictionary table of each platform, each brand and each category;
(2) and exception handling.
6. The system for identifying abnormal sales volume of retail data of electric business platform home appliances according to claim 5, wherein the abnormal volume processing steps are as follows:
the first step is as follows: integer processing
Firstly, marking data with sku sales of more than 1000, which can be evenly divided by 100, and unit price of which is more than or equal to average price;
then the processing sales: (taking the average value of sales of the SKU in 4 weeks) (industry current stage/industry previous stage) (random numbers between 0.9 and 1.1), comparing the calculation result with the original sales, if the calculation result is > (original sales 0.5), selecting the original sales and keeping two digits, and if no sales record exists in the comparison period, keeping the calculation result with the two digits and then taking the two digits as the sales;
the second step is that: abnormal change analysis of brand sales volume ratio
(1) Firstly, analyzing the ratio of the brand according to the following conditions; identifying an abnormal sales volume;
(2) brand (current/ring ratio-1) > 10%;
(3) inquiring 100 models of the industry, (the current period/the previous period-1) > 10%;
(4) the single price of the screened model is more than or equal to the average price of the model at the current stage by 0.95 and the single price of the selected model is less than or equal to the average price of the model at the previous stage by 0.5;
(5) if the screened machine type has a plurality of sales records, processing the records with the sales volume more than or equal to 100;
(6) the direct processing of no sale in the last period;
then the processing sales: (the first 4W brand + model mean value) ((industry current stage/industry previous stage)) (random number between 0.9 and 1.1), comparing the calculation result with the original sales volume, if the calculation result is greater than the original sales volume 0.5, selecting the original sales volume to reserve two digits, and if no sales record exists in the comparison period, reserving the calculation result with the two digits to serve as the sales volume;
the third step: analysis of abnormal change of sale quantity ratio of marketable model
Firstly, the proportion of the popular model is analyzed according to the following conditions:
(1) the TOP100 model is screened without being processed, and simultaneously, a single record is more than or equal to 100 records
(2) The model is within TOP10 after being corrected, and the contribution rate of the current period is more than or equal to the average contribution rate of the model in the first four periods of the model by 2; model contribution rate screening formula: (sales volume of the model in the current period/the current period scale) is not less than (sum of sales volumes of the models in the first four periods/sum of sales volumes of the classes in the first four periods ×. 2), and the contribution rate formula: the model sales volume/scale of the current period;
then the processing sales: (the first 4W brand + model mean value) × (random numbers between 0.9 and 1.1), comparing the calculation result with the original sales volume, if the calculation result is greater than the original sales volume × 0.7, selecting (random values between 50 and the original sales volume) × 0.7 as the sales volume, and if no sales record exists in the comparison period, keeping two digits of the calculation result as the sales volume.
7. The method for identifying abnormal sales volume of retail data of electric business platform home appliances according to claim 2, wherein the manual review comprises: the system adopts a mode of periodical manual review to review the data in the database.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110073348.3A CN112862519A (en) | 2021-01-20 | 2021-01-20 | Sales anomaly identification method for retail data of electric business platform household appliances |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110073348.3A CN112862519A (en) | 2021-01-20 | 2021-01-20 | Sales anomaly identification method for retail data of electric business platform household appliances |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112862519A true CN112862519A (en) | 2021-05-28 |
Family
ID=76007561
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110073348.3A Pending CN112862519A (en) | 2021-01-20 | 2021-01-20 | Sales anomaly identification method for retail data of electric business platform household appliances |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112862519A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491433A (en) * | 2017-07-24 | 2017-12-19 | 成都知数科技有限公司 | Electric business exception financial products recognition methods based on deep learning |
| CN109086889A (en) * | 2018-09-30 | 2018-12-25 | 广东电网有限责任公司 | Terminal fault diagnostic method neural network based, device and system |
| CN109190608A (en) * | 2018-10-30 | 2019-01-11 | 长威信息科技发展股份有限公司 | A kind of city intelligent identification Method violating the regulations |
| CN110866089A (en) * | 2019-11-14 | 2020-03-06 | 国家电网有限公司 | Robot knowledge base construction system and method based on synonymous multi-language environment analysis |
| CN111313355A (en) * | 2020-03-02 | 2020-06-19 | 国网江苏省电力有限公司南京供电分公司 | Method for updating monitoring signal event rule under manual supervision |
| CN111507320A (en) * | 2020-07-01 | 2020-08-07 | 平安国际智慧城市科技股份有限公司 | Detection method, device, equipment and storage medium for kitchen violation behaviors |
| CN111738011A (en) * | 2020-05-09 | 2020-10-02 | 完美世界(北京)软件科技发展有限公司 | Illegal text recognition method and device, storage medium and electronic device |
| CN112182441A (en) * | 2019-07-02 | 2021-01-05 | 中国移动通信集团贵州有限公司 | Method and device for detecting violation data |
| CN112231431A (en) * | 2020-12-11 | 2021-01-15 | 江苏苏宁银行股份有限公司 | Abnormal address identification method and device and computer readable storage medium |
-
2021
- 2021-01-20 CN CN202110073348.3A patent/CN112862519A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491433A (en) * | 2017-07-24 | 2017-12-19 | 成都知数科技有限公司 | Electric business exception financial products recognition methods based on deep learning |
| CN109086889A (en) * | 2018-09-30 | 2018-12-25 | 广东电网有限责任公司 | Terminal fault diagnostic method neural network based, device and system |
| CN109190608A (en) * | 2018-10-30 | 2019-01-11 | 长威信息科技发展股份有限公司 | A kind of city intelligent identification Method violating the regulations |
| CN112182441A (en) * | 2019-07-02 | 2021-01-05 | 中国移动通信集团贵州有限公司 | Method and device for detecting violation data |
| CN110866089A (en) * | 2019-11-14 | 2020-03-06 | 国家电网有限公司 | Robot knowledge base construction system and method based on synonymous multi-language environment analysis |
| CN111313355A (en) * | 2020-03-02 | 2020-06-19 | 国网江苏省电力有限公司南京供电分公司 | Method for updating monitoring signal event rule under manual supervision |
| CN111738011A (en) * | 2020-05-09 | 2020-10-02 | 完美世界(北京)软件科技发展有限公司 | Illegal text recognition method and device, storage medium and electronic device |
| CN111507320A (en) * | 2020-07-01 | 2020-08-07 | 平安国际智慧城市科技股份有限公司 | Detection method, device, equipment and storage medium for kitchen violation behaviors |
| CN112231431A (en) * | 2020-12-11 | 2021-01-15 | 江苏苏宁银行股份有限公司 | Abnormal address identification method and device and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xian et al. | Zero-shot learning-the good, the bad and the ugly | |
| CN106021364B (en) | Foundation, image searching method and the device of picture searching dependency prediction model | |
| CN112070125A (en) | Prediction method of unbalanced data set based on isolated forest learning | |
| CN107766929B (en) | Model analysis method and device | |
| WO2020008919A1 (en) | Machine learning device and method | |
| CN111259140B (en) | False comment detection method based on LSTM multi-entity feature fusion | |
| CN108921602B (en) | User purchasing behavior prediction method based on integrated neural network | |
| CN113095927B (en) | Method and equipment for identifying suspected transactions of backwashing money | |
| CN105117426A (en) | Intelligent search system for HSCODE | |
| CN112700324A (en) | User loan default prediction method based on combination of Catboost and restricted Boltzmann machine | |
| CN111338950A (en) | Software defect feature selection method based on spectral clustering | |
| CN113240518A (en) | Bank-to-public customer loss prediction method based on machine learning | |
| CN116700172A (en) | Industrial data integrated processing method and system combined with industrial Internet | |
| CN110109902A (en) | A kind of electric business platform recommender system based on integrated learning approach | |
| CN108537342A (en) | A kind of network representation learning method and system based on neighbor information | |
| CN116340726A (en) | Energy economy big data cleaning method, system, equipment and storage medium | |
| JPH10149280A (en) | Device and method for program composition | |
| CN111104614A (en) | Method for generating recall information for tourist destination recommendation system | |
| CN112905906B (en) | Recommendation method and system fusing local collaboration and feature intersection | |
| CN111738870B (en) | Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering | |
| CN112163731A (en) | Special transformer user electric charge recovery risk identification method based on weighted random forest | |
| CN112132589A (en) | Method for constructing fraud recognition model based on multiple times of fusion | |
| CN116503158A (en) | Enterprise bankruptcy risk early warning method, system and device based on data driving | |
| CN112862519A (en) | Sales anomaly identification method for retail data of electric business platform household appliances | |
| CN113259158B (en) | Network flow prediction method and equipment, model construction and training method and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210528 |
|
| RJ01 | Rejection of invention patent application after publication |