CN115660756A - Price monitoring method, device, equipment and medium for E-commerce commodities - Google Patents

Abstract

The invention provides a price monitoring method, a device, equipment and a medium for E-commerce commodities, which are characterized in that an RPA robot flow automation technology is applied to provide intelligent commodity identification monitoring service, an algorithm model is applied to construct commodity aggregation normalization service, the same commodity is accurately positioned from massive commodity data, a manual repetitive operation mode is replaced, special commodities are removed, a reasonable price calculation model is provided to calculate the reasonable price of the same commodity, price change of the same commodity is monitored according to the reasonable price of the commodity, risk reminding is pushed to a user according to the improper change of the price, the purchasing time cost of an enterprise is reduced, and the purchasing efficiency is improved.

Description

Price monitoring method, device, equipment and medium for E-commerce commodities

Technical Field

The invention relates to the technical field of computers, in particular to a price monitoring method, device, equipment and medium for E-commerce commodities.

Background

Under the current situation of the vigorous development of the e-commerce industry, the basic requirements of user purchasing have been met. However, the e-commerce industry needs to perform differentiated and distinctive operation, and the stacking only by conventional functions cannot meet the requirements. In view of the trend of the electric business to be refined, the potential needs and the implicit business scenes of the users need to be continuously mined.

In the case where the product cannot be differentiated in quality in the purchase of the enterprise, the implementation of the buyer focuses on the price of the product. Particularly, under the scene that enterprises frequently carry out centralized batch purchasing, a lot of purchasing expenses can be saved for the enterprises by slight price optimization. How to find the same-style goods with price advantage can be a pain point of enterprise buyers, and the normalizing function of the same-style goods is a direction in which important research is needed in the shopping mall.

In a shopping mall which does not realize the same-style commodity normalization function, a buyer can screen the same-style commodities by searching the key information (such as commodity name, brand and model) of the commodities, and under the condition of knowing the potential user use scene, the shopping mall can easily develop the normalization function capable of directly providing the same-style commodities, the normalization function is strongly matched to a certain extent based on the commodity key information, such as the model, the brand and the item are the same, and the commodities with the names having certain similarity in terms of word frequency can be classified into the same style. The realization mode can achieve better accuracy under the condition of more standard data, but the data source channels in the purchasing mall of the actual enterprise are wide, the data quality of commodity information is low, and the unified mode definition standard is lacked, so that the method frequently makes mistakes in the actual application. Data loss, errors and description similarities and differences of actual commodity data on key information often occur, and data noise aggravates the difficulty in achieving the normalization of the same-style commodities.

The potential meaning of the normalization calculation problem of the commodities of the same type is the calculation of the similarity of the commodities, and the normalization of the commodities with high similarity forms the commodities of the same type (normalization). To realize the normalization of the heterogeneous and disordered commodity data, the scheme adopted at present is mainly to clean the data in advance, and the normalization can be easily realized after unified normalized data is formed. However, the difficulty of data cleaning is not low, the difficulties of data unified mapping, data error correction, default filling and the like are not realized, the deviation of the purpose is easily caused because of operation, and the method has the advantages that byproducts of the normalized data after error correction are easily obtained, other related commodities can be seen from the normalized commodity data, and even the commodity data can be rounded up from more dimensions.

China invention with publication number CN107507075A, published as 20171222, provides a public purchase big data price monitoring method, which comprises the following steps: establishing a unified item library, a product library, a commodity library and an item standard parameter library by a monitoring platform; the monitoring platform collects monitoring data from a plurality of monitoring objects; acquiring price and standard parameters according to the brand model in the monitoring data, acquiring commodity information according to the e-commerce and commodity library in the monitoring data, aggregating the monitoring data acquired from internal existing data and external acquired data to generate reasonable market price, and updating the corresponding item library, product library, commodity library and item standard parameter library; and establishing a data statistical report of the commodity, generating the price trend of the commodity, realizing the price monitoring of the commodity, and displaying the price monitoring result in a graphical interface mode.

The invention discloses a computer system for monitoring commodity price of an e-commerce platform, which is disclosed by China with publication number CN113869987A and publication number 20211231, and comprises: the data acquisition interface is used for acquiring stock unit data corresponding to target commodities with different specifications in the e-commerce platform; the arithmetic unit is used for dividing the stock unit data according to the updating time of the stock unit data to obtain the stock unit data in different time stages and calculating the stage average price corresponding to all the stock unit data in each time stage; the configuration interface is used for distributing corresponding weights for each time stage and determining the target price of the target commodity by using the weights of all the time stages and the stage average price; and the controller is used for monitoring the price of the target commodity in the E-commerce platform based on the target price so as to correspondingly process the target commodity with abnormal price. Through data interaction among the data acquisition interface, the arithmetic unit, the configuration interface and the controller, the commodity price monitoring efficiency and accuracy of the e-commerce platform are improved.

Disclosure of Invention

The invention aims to solve the technical problem of providing a price monitoring method, a device, equipment and a medium for E-commerce commodities, which realize price monitoring on the same commodity in a commodity normalization and reasonable price calculation model so as to reduce the purchasing cost of an enterprise.

In a first aspect, the present invention provides a method for monitoring prices of e-commerce commodities, including:

and (3) commodity normalization: the method comprises the steps that an RPA robot process automation technology is adopted, and the commodities with different sources and different description standards are aggregated and identified according to a uniform commodity attribute standard and a commodity similarity value, so that the same commodity of each category is obtained; the method specifically comprises the following steps:

(1) An environment building process, namely downloading a bert pre-training language model bert-base-chip; pre-training the downloaded bert pre-training language model bert-base-chip through a mass of unmarked corpora to obtain a pre-trained bert model; the mass unmarked corpus is not marked as the title of the commodity;

(2) In the model fine tuning process, a category label set file class.txt is constructed based on the criteria of a category library and is used for reading a pre-trained bert model; performing one-hot coding on the item list to form a label matrix; selecting a run _ classifier. Associating a data set file, a category label set file class.txt and an initial model parameter list in the run _ classifier.py file, so that a labeled data set C and the category label matrix can be loaded as the input of a classifier, the classifier is operated after the initial model parameter list is loaded and an initial learning rate parameter value is configured, and the classifier is used for reasoning the input title information of each commodity to obtain the reasoning result; reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest reasoning capability of the classifier as an optimal model, and storing a model parameter list of the optimal model;

(3) A model evaluation process, wherein the optimal model is operated, and a model parameter list of the optimal model is loaded; inputting a test data set to check the confidence of an output result, checking the accuracy of the model for classifying commodities, if the accuracy is lower than a threshold value, reconfiguring a learning rate value and a model parameter list and then carrying out model fine adjustment until the accuracy is not lower than the threshold value, and then obtaining a bert model which is the category prediction model;

the downstream classification task is used for dividing the commodity title information text into a plurality of commodity categories; the labeling data set C is labeled with a commodity title and a commodity label; the model parameter list comprises related parameters of commodity attributes, names and brands; the test data set is a data set of a certain amount of commodity title information;

and (3) commodity coding: printing a corresponding category code for the same commodity and storing the same in a commodity pool, wherein the category code comprises a major category number, a middle category number, a minor category number and a serial number;

monitoring of special commodities: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology, wherein the commodity information comprises commodity names and brand information; regularly visiting a main-electricity merchant platform and a commodity source website, setting a monitoring range, inputting extracted commodity information for matching search, judging that the commodity is a special supply commodity if the matching search is not successful in the monitoring range and the matching search is still not successful after tracing the commodity source website, and then putting off the shelf for the special supply commodity and generating a suspected special supply commodity list;

and (4) reasonable price monitoring: triggering a regular task, extracting price information of the same commodity of the commodity in the commodity pool according to the category code, transmitting the price information to a reasonable price algorithm model, and calculating the reasonable price of the commodity; storing the reasonable price calculation result of the commodity;

monitoring commodity price change: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology; when the commodity price changes, accessing the commodity link and extracting commodity price information; judging the price change condition of the goods on the shelf, and if the price of the goods is reduced, automatically reducing by default; if the price of the commodity is adjusted upwards, the price is compared with the reasonable price of the commodity and the e-commerce official website price of the commodity, and if the price after the adjustment is higher than the reasonable price, the risk reminding is pushed to the user.

In a second aspect, the present invention provides a price monitoring device for an e-commerce commodity, comprising: the method comprises the following steps:

a commodity normalization module: the method comprises the steps that an RPA robot process automation technology is adopted, and the commodities with different sources and different description standards are aggregated and identified according to a uniform commodity attribute standard and a commodity similarity value, so that the same commodity of each category is obtained; the method specifically comprises the following steps:

(1) An environment building process, namely downloading a bert pre-training language model bert-base-chip; pre-training the downloaded bert pre-training language model bert-base-chip through massive non-labeled corpora to obtain a pre-trained bert model; the mass unmarked corpus is not marked as a commodity title;

(2) In the model fine tuning process, a category label set file class.txt is constructed based on the criteria of a category library and is used for reading a pre-trained bert model; performing one-hot coding on the item list to form a label matrix; selecting a run _ classifier. Associating a data set file, a category label set file class.txt and an initial model parameter list in the run _ classifier.py file, so that a labeled data set C and the category label matrix can be loaded as the input of a classifier, the classifier is operated after the initial model parameter list is loaded and an initial learning rate parameter value is configured, and the classifier is used for reasoning the input title information of each commodity to obtain the reasoning result; reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest reasoning capability of the classifier as an optimal model, and storing a model parameter list of the optimal model;

(3) A model evaluation process, wherein the optimal model is operated, and a model parameter list of the optimal model is loaded; inputting a test data set to check the confidence of an output result, checking the accuracy of the model for classifying commodities, if the accuracy is lower than a threshold value, reconfiguring a learning rate value and a model parameter list and then carrying out model fine tuning until the accuracy is not lower than the threshold value, and then obtaining a bert model which is the category prediction model;

the downstream classification task is used for classifying the commodity title information text into a plurality of commodity categories; the labeling data set C is labeled with a commodity title and a commodity label; the model parameter list comprises related parameters of commodity attributes, names and brands; the test data set is a data set of a certain amount of commodity title information;

the commodity code printing module: printing a corresponding category code for the same commodity and storing the category code in a commodity pool, wherein the category code comprises a major category number, a middle category number, a minor category number and a serial number;

the special supply commodity monitoring module: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology, wherein the commodity information comprises commodity names and brand information; regularly visiting a main-electricity merchant platform and a commodity source website, setting a monitoring range, inputting extracted commodity information for matching search, judging that the commodity is a special supply commodity if the matching search is not successful in the monitoring range and the matching search is still not successful after tracing the commodity source website, and then putting off the shelf for the special supply commodity and generating a suspected special supply commodity list;

reasonable price monitoring module: triggering a regular task, extracting price information of the same commodity of the commodity in the commodity pool according to the category code, transmitting the price information to a reasonable price algorithm model, and calculating the reasonable price of the commodity; storing the reasonable price calculation result of the commodity;

commodity price change monitoring module: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology; when the price of the commodity changes, accessing the commodity link and extracting commodity price information; judging the price change condition of the goods on the shelf, and if the price of the goods is reduced, automatically reducing by default; if the price of the commodity is adjusted upwards, the price is compared with the reasonable price of the commodity and the e-commerce official network price of the commodity, and if the adjusted price is higher than the reasonable price or the e-commerce official network price, the risk reminding is pushed to the user.

In a third aspect, the present invention provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of the first aspect when executing the program.

In a fourth aspect, the invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the method of the first aspect.

One or more technical schemes provided in the embodiments of the present invention have at least the following technical effects or advantages: the invention provides a commodity normalization algorithm-based price monitoring system or terminal, which aligns the application requirements of public purchasing and national enterprise purchasing, provides intelligent commodity identification monitoring service by applying an RPA robot process automation technology, constructs commodity aggregation normalization service by applying an algorithm model, accurately positions the same commodity from mass commodity data, replaces a manual repetitive operation mode, and reduces the manual input cost. The special supply commodities are removed, the reasonable price calculation model is provided for calculating the reasonable price of the same commodity, the price change of the same commodity is monitored according to the reasonable price of the commodity, the risk reminding is pushed to the user according to the improper change of the price, the purchasing time cost of an enterprise is reduced, and the purchasing efficiency is improved. According to the commodity data analysis method, by means of feature extraction of technologies such as big data and artificial intelligence and implementation of a reasonable price algorithm, commodity data are interpreted from two aspects of dimensional modeling and data indexes, digital, automatic and intelligent supervision in the public purchasing field is assisted, and the data generate value.

The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.

Drawings

The invention will be further described with reference to the following examples with reference to the accompanying drawings.

FIG. 1 is a flow chart of a method according to one embodiment of the present invention;

FIG. 2 is a block diagram of a class code according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a process of calculating the reasonable price of a commodity by a reasonable price algorithm model in the embodiment of the present invention;

FIG. 4 is a schematic structural diagram of an apparatus according to a second embodiment of the present invention;

fig. 5 is a schematic structural diagram of an electronic device according to a third embodiment of the invention;

fig. 6 is a schematic structural diagram of a medium according to a fourth embodiment of the present invention.

Detailed Description

The embodiment of the application provides a price monitoring method, a device, equipment and a medium for E-commerce commodities, which realize price monitoring on the same commodity through a commodity normalization and reasonable price calculation model, thereby reducing the purchasing cost of enterprises.

The technical scheme in the embodiment of the application has the following general idea: the intelligent commodity identification monitoring service is provided by the RPA robot process automation technology, the commodity aggregation normalization service is established by the algorithm model, the same commodity is accurately positioned from massive commodity data, the manual repetitive operation mode is replaced, and the manual input cost is reduced. The special supply commodities are removed, the reasonable price calculation model is provided for calculating the reasonable price of the same commodity, the price change of the same commodity is monitored according to the reasonable price of the commodity, the risk reminding is pushed to the user according to the improper change of the price, the purchasing time cost of an enterprise is reduced, and the purchasing efficiency is improved.

Wherein, RPA robot process automation technology is: a series of business processes are automatically executed in a human-simulated mode, and the system is used for intelligently identifying commodities specially supplied in massive commodity data of a purchasing platform and intelligently identifying and monitoring the information of the commodities of the same type, can realize all-weather zero-clearance work, is equivalent to the ultrahigh working efficiency of 15 times of manpower, enables the business to execute a large number of repetitive tasks with zero error rate, and greatly reduces the operation cost of enterprises.

Example one

As shown in fig. 1, the present embodiment provides a price monitoring method for e-commerce commodities, which includes commodity normalization, commodity coding, commodity specific supply monitoring, reasonable price monitoring, and commodity price change monitoring.

And (3) commodity normalization: the method comprises the steps that an RPA robot process automation technology is adopted, and the commodities with different sources and different description standards are aggregated and identified according to a uniform commodity attribute standard and a commodity similarity value, so that the same commodity of each category is obtained; the method specifically comprises the following steps:

extracting the core fields of the commodities by adopting an automatic brand extraction model, an automatic model extraction model and a category prediction model;

the method comprises the following steps of carrying out SPU pre-division on commodities by adopting a commodity semantic recall model, a brand identity relation table and a category identity relation table, wherein the SPU pre-division aims at carrying out primary screening on big data once and reducing the data volume of subsequent flow processing, for example, the data volume is similar to that of the commodity semantic recall model, the brand identity relation table and the category identity relation table;

adopting a commodity similarity judgment model to perform SPU division on commodities;

and a key attribute automatic extraction model is adopted to carry out SKU division on the key attributes of the commodities, so that the same commodity can be classified in the finest mode.

The same-brand relation table is obtained by sorting different expression modes of the same brand, for example, "Huacheng" and "HUAWEI" are actually the same brand;

the same category relation table is obtained by arranging different expression modes of the same category, for example, a Tianmao E-commerce platform has a category called parent-child mother-baby, but a Jingdong platform is a special area called mother-baby, but actually the same category is obtained;

SPU means commodity code, is the minimum unit of the information of the commodity aggregation; the SKU is a categorical attribute under the item, and is the smallest unit of non-reclassification of the item. For example:

SPU: mobile phone- > apple mobile phone- > apple 6, apple 6 is the minimum unit of commodity aggregation information, so apple 6 is the SPU.

SKU: the native gold 16G apple 6, the smallest unit that has not been resolvable.

The specific implementation of the category prediction model comprises the following processes:

(1) An environment building process, namely downloading a bert pre-training language model bert-base-chip; pre-training the downloaded bert pre-training language model bert-base-chip through massive non-labeled corpora to obtain a pre-trained bert model; the mass unmarked corpus is not marked as a commodity title;

(2) A model fine adjustment process, namely constructing a category label set file class.txt based on a category library standard for a pre-trained bert model to read; performing one-hot coding on the item list to form a label matrix; selecting a run _ classifier. Associating a data set file, an item tag set file class txt and an initial model parameter list in the run _ classifier. The code is implemented as follows: bert _ config = modeling, bertconfig.from _ json _ file ("chicken _ L-12_h-768 a-12/bert _ config.json")

model＝modeling.BertModel(

config＝bert_config,

is_training＝is_training,

input_ids＝input_ids,

input_mask＝input_mask,

token_type_ids＝segment_ids,

use_one_hot_embeddings＝use_one_hot_embeddings)。

The classifier is used for reasoning the item of each input commodity title information, and reasoning results are listed; reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest reasoning capability of the classifier as an optimal model, and storing a model parameter list of the optimal model;

the model parameter list comprises related parameters of commodity attributes, names and brands;

the reasoning about the categories of the input title information of each commodity is as follows: the Chinese characters input into some commodity title information are first preprocessed to be converted into integer codes, and each Chinese character corresponds to an integer and then is inferred.

The algorithm adopted in the model fine tuning process is specifically as follows:

the LM language model is trained using the following objective function:

P(w _i |w ₁ ,...,w _i-1 ,w _i+1 ,...,w _n )

csv reads the commodity title X from the labeling data set C as input, splits the commodity title X from a chinese character level to obtain n chinese characters, converts the characteristics of each chinese character into integer codes to obtain n integer codes W1, \ 8230, where Wn is a classified one-hot tag y output as a list of items, and the one-hot tag y is an expression mode of the item tag;

inputting the integer code W1, \8230, wn into a Transformer model to obtain an output set h of the last moment of the uppermost layer _l ⁿ H is to be _l ⁿ Classifying through a softmax layer, wherein the parameter of the softmax layer is Wy, wy is the label code of a certain one-hot label y, finally calculating loss by using Cross EntropyLoss, and adjusting the learning rate of a transform model and the parameter Wy, which are equivalent to maximum likelihood estimation:

use ofMulti-Task Learning, simultaneous maximum likelihood L ₁ And L ₂ 。

L ₃ (C)＝L ₂ (C)+λ×L ₁ (C)

L ₁ Or the loss of the previous language model, the used data is the labeled data set C of the current task, and only X in the labeled data set C is used without a one-hot label y;

L ₂ loss in the fine adjustment process, wherein the used data is a labeled data set C of the current task, and X and a one-hot label y are used simultaneously;

L ₃ is the maximum likelihood value, i.e. is the confidence.

The output of the inference result is an array, the value of each element of the array represents the confidence coefficient of a commodity object, the value of each element is 0.00-1.00, the smaller the value is, the lower the probability of representing the commodity object is, and the larger the value is, the higher the probability of representing the commodity object is.

For example:

the classifier of the run _ classifier.py file loads a train.csv file of the labeling data set C, and traverses and reads each item title information in the train.csv file, for example, a certain item title information is "association (lenova) LJ2605D A4 automatic double-sided black and white laser printer".

Splitting the commodity title X from the Chinese character level to obtain 13 Chinese characters, and performing feature conversion on each Chinese character to obtain integer codes, wherein each Chinese character corresponds to an integer, namely: [135,2102,75,13, \ 8230; \ 8230;, 0,0].

And performing one-hot coding on the item list to form a label matrix. For example, one-hot coding is performed on three categories of 'printer', 'personal guard' and 'cleaning article', and the results are as follows:

[ [ "printer" ], [ "personal protection" ], [ "cleaning article" ] → [ [1, 0], [0,1,0], [0, 1] ]

After reasoning, outputting a result array: [0.91,0.01, \ 8230;, 0.06], the confidence that the model inference article is a printer is 91%. Each element in the array corresponds to a category in the item list one by one, the value of each element represents the confidence coefficient that the model reasoning commodity is the category, and the number of the elements is consistent with the number of the categories in the item list.

Reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest reasoning capability of the classifier as an optimal model, and storing a model parameter list of the optimal model; the model parameter list includes related parameters of the commodity attribute, the name and the brand. The initial model parameter list may be obtained from a parameter file config.

(3) A model evaluation process, wherein the optimal model is operated, and a model parameter list of the optimal model is loaded; inputting a test data set to check the confidence of an output result, checking the accuracy of the model for classifying commodities, if the accuracy is lower than a threshold value, reconfiguring a learning rate value and a model parameter list and then carrying out model fine tuning until the accuracy is not lower than the threshold value, and then obtaining a bert model which is the category prediction model;

the downstream classification task is used for classifying the commodity title information text into a plurality of commodity categories; the labeling data set C is labeled with a commodity title and a commodity label; the model parameter list comprises related parameters of commodity attributes, names and brands; the test data set is a data set of a certain amount of commodity title information.

Code printing of the commodity: and classifying the same commodity to obtain a large class number, a middle class number and a small class number, marking corresponding class codes and storing the class codes in a commodity pool, so that the minimum unit of each commodity corresponds to one class number. As shown in fig. 2: the category code comprises a large category number, a middle category number, a small category number and a serial number, and the serial number is the serial number generated by the category code.

Monitoring of special commodities:

special or special commodities are not allowed to be sold in the general commodity purchasing platform, and the searching of special commodities from mass commodity data in a manual sorting mode is time-consuming and labor-consuming. In order to solve the problem, the invention adopts an RPA robot process automation technology to automatically and intelligently identify the special commodities for the operation. The concrete realization is as follows:

establishing a commodity information extraction flow by adopting an RPA robot flow automation technology, wherein the commodity information comprises commodity names and brand information; regularly visiting a main-electricity commercial platform and a commodity source website, setting a monitoring range, inputting extracted commodity information to perform matching search, judging that the commodity is a special supply commodity if the matching search is not successful in the monitoring range and the matching search is still not successful after tracing the commodity source website, then putting off the shelf of the special supply commodity, and generating a suspected special supply commodity list;

monitoring reasonable price:

under the mode that purchasing platforms of governments, enterprises and colleges are independently built, commodity prices are high and differential price problems of suppliers of manufacturers are not easy to find. In order to solve the problem, the invention adopts a reasonable price algorithm model to automatically calculate the reasonable price of the commodity and provide price comparison basis for the purchase of the user. The concrete implementation is as follows:

triggering a periodic task, extracting price information of the same commodity of the commodity in the commodity pool according to the category code, transmitting the price information to a reasonable price algorithm model, and calculating the reasonable price of the commodity; saving the reasonable price calculation result of the commodity; as shown in fig. 3, the process of calculating the reasonable price of the commodity by the reasonable price algorithm model is as follows:

when the extracted price information of the same commodity is only 1 price, the price information is directly used as a market reasonable price;

when the extracted price information of the same commodity has 2-7 prices, taking the average number as the market reasonable price;

and when the extracted price information of the same commodity has more than 7 prices, effective price screening is carried out, abnormal prices are eliminated, 95% of data is taken as effective confidence interval prices at the rest prices, quartile arrangement is carried out, the quartile at the midpoint position is the median, and the average price is calculated as the market reasonable price by taking the first 50% of prices.

The effective price is screened as follows:

arranging all prices from small to large, and finding out a median price m yuan; the median price m determines a coefficient x, and the current relation between the median price m and the coefficient x is as follows:

m > =5000 yuan, x =0.3;

2000> < m > < 5000 yuan, x =0.5;

when m < 2000-membered, x =1;

prices greater than m (1 + x) or less than m (1-x) are excluded as abnormal prices.

The coefficient x is an empirical value, and the value of the coefficient x is obtained by calculation of existing data and by daily price processing for many years (for example, certain reasonable prices are excluded or unreasonable prices are included in calculation).

After a round of coarse deviation price removal, the remaining market prices can be considered to satisfy normal distribution, and in order to further make price samples reasonable, about 95% of the data is taken as the valid confidence interval price, that is, the data within 2 variances. The price mean value mu and the standard deviation sigma of the price set are calculated. The range x within 2 standard deviations σ is taken as the price confidence interval P (μ -2 σ < x < μ +2 σ) =95.4%.

Monitoring commodity price change:

because enterprises lack efficient multidimensional commodity price analysis capability, in order to solve the problem, the RPA robot process automation technology is adopted to intelligently monitor the price abnormity of commodities on shelves. The concrete implementation is as follows:

establishing a commodity information extraction flow by adopting an RPA robot flow automation technology; when the price of the commodity changes, accessing the commodity link and extracting commodity price information; judging the price change condition of the goods on the shelf, and if the price of the goods is reduced, automatically reducing by default; if the price of the commodity is adjusted upwards, the price is compared with the reasonable price of the commodity and the e-commerce official website price of the commodity, and if the price after the adjustment is higher than the reasonable price, the risk reminding is pushed to the user.

In addition, the invention can also comprise commodity sensitive word monitoring and commodity information abnormity monitoring.

Monitoring commodity sensitive words:

the information of the goods on the shelf is required to meet the national standard requirement, and the data mode of manual verification scale is limited inefficiently. In order to solve the problem, the invention adopts a sensitive word tool realized based on a DFA algorithm to monitor whether the commodity contains sensitive words. The concrete implementation is as follows:

establishing a commodity information extraction flow by adopting an RPA robot flow automation technology;

accessing a commodity link, and extracting information such as a commodity title, a commodity name, commodity description and the like;

transmitting the extracted commodity information to a sensitive word tool, and judging whether the commodity contains sensitive words;

and if the commodity contains sensitive words, pushing risk reminders.

Monitoring commodity information abnormity:

the commodity parameter information has various abnormal conditions, and the problems can not be found in time in a manual checking mode. In order to solve the problem, the invention adopts an RPA robot process automation technology to monitor whether the commodity link is effective, whether the commodity is off-shelf, whether the commodity has stock and whether the commodity price is 0. The concrete implementation steps are as follows:

establishing a commodity information extraction flow by adopting an RPA robot flow automation technology;

accessing the commodity link and judging whether the commodity link is effective or not;

extracting commodity shelf-off information and judging whether the commodity is shelf-off or not;

extracting commodity inventory information and judging whether the commodities are in inventory or not;

extracting commodity price information and judging whether the commodity price is 0 or not;

extracting the evaluation number and the good evaluation rate information of the commodity;

and transmitting the acquisition result of the commodity page information to a background, and pushing a risk prompt.

Regarding risk alert pushing:

when the commodity is abnormal, the user needs to be reminded in time. In order to solve the problem, the invention adopts RabbitMQ message queue service software to push the risk prompt in real time.

Based on the same inventive concept, the application also provides a device corresponding to the method in the first embodiment, which is detailed in the second embodiment.

Example two

As shown in fig. 3, in the present embodiment, there is provided a price monitoring device for an e-commerce commodity, including:

a commodity normalization module:

the method comprises the steps that an RPA robot process automation technology is adopted, and the commodities with different sources and different description standards are aggregated and identified according to uniform commodity attribute standards and commodity similarity values to obtain the same commodity of each category; the commodity normalization specifically comprises the following steps:

extracting commodity core fields by adopting an automatic brand extraction model, an automatic model extraction model and a category prediction model;

adopting a commodity semantic recall model, a brand identity relationship table and a category identity relationship table to pre-divide commodities by SPU;

adopting a commodity similarity judgment model to perform SPU division on commodities;

and adopting a key attribute automatic extraction model to divide SKU (stock keeping unit) of the key attributes of the commodities.

The specific implementation of the category prediction model comprises the following processes:

(1) An environment building process, namely downloading a bert pre-training language model bert-base-chip; pre-training the downloaded bert pre-training language model bert-base-chip through a mass of unmarked corpora to obtain a pre-trained bert model; the mass unmarked corpus is not marked as the title of the commodity;

(2) In the model fine tuning process, a category label set file class.txt is constructed based on the criteria of a category library and is used for reading a pre-trained bert model; performing one-hot coding on the item list to form a label matrix; selecting a run _ classifier. Associating a data set file, a category label set file class.txt and an initial model parameter list in the run _ classifier.py file, so that a labeled data set C and the category label matrix can be loaded as the input of a classifier, the classifier is operated after the initial model parameter list is loaded and an initial learning rate parameter value is configured, and the classifier is used for reasoning the input title information of each commodity to obtain the reasoning result; reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest reasoning capability of the classifier as an optimal model, and storing a model parameter list of the optimal model; the code is implemented as follows:

bert_config＝modeling.BertConfig.from_json_file("chinese_L-12_H-768_A-12/bert_config.json")

model＝modeling.BertModel(

config＝bert_config,

is_training＝is_training,

input_ids＝input_ids,

input_mask＝input_mask,

token_type_ids＝segment_ids,

use_one_hot_embeddings＝use_one_hot_embeddings)。

reasoning the item of each input commodity title information through a classifier, and listing a reasoning result; reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest inference capability of a classifier as an optimal model, and storing a model parameter list of the optimal model;

the model parameter list comprises related parameters of commodity attributes, names and brands;

the reasoning about the categories of the input commodity title information is as follows: the Chinese characters input into some commodity title information are first preprocessed to be converted into integer codes, and each Chinese character corresponds to an integer and then is inferred.

The algorithm adopted in the model fine tuning process is specifically as follows:

the LM language model is trained using the following objective function:

P(w _i |w ₁ ,...,w _i-1 ,w _i+1 ,...,w _n )

csv reads the commodity title X from the labeling data set C as input, splits the commodity title X from a chinese character level to obtain n chinese characters, converts the characteristics of each chinese character into integer codes to obtain n integer codes W1, \ 8230, where Wn is a classified one-hot tag y output as a list of items, and the one-hot tag y is an expression mode of the item tag;

inputting the integer code W1, \8230, wn into a Transformer model to obtain an output set h of the last moment of the uppermost layer _l ⁿ H is to be _l ⁿ Classifying through a softmax layer, wherein parameters of the softmax layer are Wy, the Wy is label coding of a certain one-hot label y, finally calculating loss by using Cross EntropyLoss, adjusting learning rate of a transform model and the parameters Wy, and equating to maximum likelihood estimation:

using Multi-Task Learning, while maximum likelihood L ₁ And L ₂ 。

L ₃ (C)＝L ₂ (C)+λ×L ₁ (C)

L ₁ Or the loss of the previous language model, the used data is the labeled data set C of the current task, and only X in the labeled data set C is used without a one-hot label y;

L ₂ loss in the fine adjustment process, wherein the used data is a labeled data set C of the current task, and an X label and a one-hot label y are used simultaneously;

L ₃ is the maximum likelihood value, i.e. is the confidence.

The output of the inference result is an array, the value of each element of the array represents the confidence coefficient of a commodity object, the value of each element is between 0.00 and 1.00, the smaller the value is, the lower the probability of representing the commodity object is, and the larger the value is, the higher the probability of representing the commodity object is.

For example:

the classifier of the run _ classifier.py file loads a train.csv file of the labeling data set C, and traverses and reads each item title information in the train.csv file, for example, a certain item title information is "association (leonovo) LJ2605D A4 automatic double-sided black and white laser printer".

Splitting the commodity title X from the Chinese character level to obtain 13 Chinese characters, and performing feature conversion on each Chinese character to obtain integer codes, wherein each Chinese character corresponds to an integer, namely: [135,2102,75,13, \8230;, 0].

And performing one-hot coding on the item list to form a label matrix. For example, one-hot coding is performed on three categories of 'printer', 'personal guard' and 'cleaning article', and the results are as follows:

[ [ "printer" ], [ "personal protection" ], [ "cleaning article" ] → [ [1, 0], [0,1,0], [0, 1] ]

After reasoning, outputting a result array: [0.91,0.01, \ 8230;, 0.06], the confidence that the model inference article is a printer is 91%. Each element in the array corresponds to a category in the item list one by one, the value of each element represents the confidence coefficient that the model reasoning commodity is the category, and the number of the elements is consistent with the number of the categories in the item list.

Reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest inference capability of a classifier as an optimal model, and storing a model parameter list of the optimal model; the model parameter list includes relevant parameters of the commodity attribute, the name and the brand. Json may be obtained from a parameter file on the web.

(3) A model evaluation process, wherein the optimal model is operated, and a model parameter list of the optimal model is loaded; inputting a test data set to check the confidence of an output result, checking the accuracy of the model for classifying commodities, if the accuracy is lower than a threshold value, reconfiguring a learning rate value and a model parameter list and then carrying out model fine adjustment until the accuracy is not lower than the threshold value, and then obtaining a bert model which is the category prediction model;

the downstream classification task is used for classifying the commodity title information text into a plurality of commodity categories; the labeling data set C is labeled with a commodity title and a commodity label; the model parameter list comprises related parameters of commodity attributes, names and brands; the test data set is a data set of a certain amount of commodity title information.

The commodity code printing module:

printing a corresponding category code for the same commodity, and storing the same commodity in a commodity pool, as shown in fig. 2: the category code comprises a major class number, a middle class number, a minor class number and a serial number;

the special supply commodity monitoring module:

special or special commodities are not allowed to be sold in the general commodity purchasing platform, and searching special commodities from mass commodity data in a manual sorting mode wastes time and labor. In order to solve the problem, the invention adopts the RPA robot process automation technology, and the automatic operation intelligent identification is specially provided for the commodity.

Establishing a commodity information extraction flow by adopting an RPA robot flow automation technology, wherein the commodity information comprises commodity names and brand information; regularly visiting a main-electricity commercial platform and a commodity source website, setting a monitoring range, inputting extracted commodity information to perform matching search, judging that the commodity is a special supply commodity if the matching search is not successful in the monitoring range and the matching search is still not successful after tracing the commodity source website, then putting off the shelf of the special supply commodity, and generating a suspected special supply commodity list; the suspected special commodity list can be used as a basis for subsequent manual examination and sale of special commodities.

Reasonable price monitoring module:

under the mode that purchasing platforms of governments, enterprises and colleges are independently built, commodity prices are high and differential price problems of suppliers of manufacturers are not easy to find. In order to solve the problem, the invention adopts a reasonable price algorithm model to automatically calculate the reasonable price of the commodity and provide price comparison basis for the purchase of the user. The specific process executed by the reasonable price monitoring module is as follows:

triggering a periodic task, extracting price information of the same commodity of the commodity in the commodity pool according to the category code, transmitting the price information to a reasonable price algorithm model, and calculating the reasonable price of the commodity; saving the reasonable price calculation result of the commodity; the process of calculating the reasonable price of the commodity by the reasonable price algorithm model comprises the following steps:

when the extracted price information of the same commodity is only 1 price, the price information is directly used as a market reasonable price;

when the extracted price information of the same commodity has 2-7 prices, taking the average number as the market reasonable price;

and when the extracted price information of the same commodity has more than 7 prices, effective price screening is carried out, abnormal prices are eliminated, 95% of data is taken as effective confidence interval prices at the rest prices, quartile arrangement is carried out, and the average price is calculated from the first 50% of prices and taken as the market reasonable price.

The effective price is screened as follows:

arranging all prices from small to large, and finding out a median price m yuan; the median price m determines a coefficient x, and the relationship between the current median price m and the coefficient x is as follows:

m > =5000 yuan, x =0.3;

2000> < m >5000 yuan, x =0.5;

when m < 2000-membered, x =1;

prices greater than m (1 + x) or less than m (1-x) are excluded as abnormal prices.

The coefficient x is an empirical value, and the value of the coefficient x is obtained by calculation of existing data and by daily price processing for many years (for example, certain reasonable prices are excluded or unreasonable prices are included in calculation).

After a round of coarse deviation price removal, the remaining market prices can be considered to satisfy normal distribution, and in order to further make price samples reasonable, about 95% of data is taken as effective confidence interval prices, namely data within 2 variances. The price mean value μ and standard deviation σ of the price set are calculated, and the range x within 2 standard deviations σ is taken as the price confidence interval P (μ -2 σ < x < μ +2 σ) =95.4%.

Commodity price change monitoring module:

because enterprises lack efficient multi-dimensional commodity price analysis capability, in order to solve the problem, the RPA robot flow automation technology is adopted to intelligently monitor the price abnormity of the commodities on shelves. The specific flow executed by the commodity price change monitoring module is as follows:

establishing a commodity information extraction flow by adopting an RPA robot flow automation technology; when the commodity price changes, accessing the commodity link and extracting commodity price information; judging the price change condition of the goods on the shelf, and if the price of the goods is reduced, automatically reducing by default; if the price of the commodity is adjusted upwards, the price is compared with the reasonable price of the commodity and the e-commerce official website price of the commodity, and if the price after the adjustment is higher than the reasonable price, the risk reminding is pushed to the user.

In addition, the system can also comprise a commodity sensitive word monitoring module, a commodity information abnormity monitoring module and a risk reminding pushing module.

The commodity sensitive word monitoring module:

the information of the goods on the shelf needs to meet the national standard requirements, and the data mode of manual verification scale is limited in low efficiency. In order to solve the problem, the invention adopts a sensitive word tool realized based on a DFA algorithm to monitor whether the commodity contains sensitive words. The specific flow executed by the commodity sensitive word monitoring module is as follows:

establishing a commodity information extraction flow by adopting an RPA robot flow automation technology;

accessing commodity links, and extracting information such as commodity titles, commodity names, commodity descriptions and the like;

transmitting the extracted commodity information to a sensitive word tool, and judging whether the commodity contains sensitive words;

and if the commodity contains sensitive words, pushing risk reminders.

Commodity information anomaly monitoring module:

the commodity parameter information has various abnormal conditions, and the problems can not be found in time in a manual checking mode. In order to solve the problem, the invention adopts an RPA robot process automation technology to monitor whether the commodity link is effective, whether the commodity is off-shelf, whether the commodity has inventory and whether the commodity price is 0. The specific flow executed by the commodity information abnormity monitoring module is as follows

Establishing a commodity information extraction flow by adopting an RPA robot flow automation technology;

accessing the commodity link and judging whether the commodity link is effective or not;

extracting commodity shelf-off information and judging whether the commodity is shelf-off or not;

extracting commodity inventory information and judging whether the commodities are in inventory or not;

extracting commodity price information and judging whether the commodity price is 0 or not;

extracting the evaluation number and the favorable evaluation rate information of the commodity;

and transmitting the acquisition result of the commodity page information to a background, and pushing a risk prompt.

The risk reminding and pushing module:

when the commodity is abnormal, the user needs to be reminded in time. In order to solve the problem, the risk reminding pushing module adopts RabbitMQ message queue service software to push the risk reminding in real time.

Since the apparatus described in the second embodiment of the present invention is an apparatus used for implementing the method of the first embodiment of the present invention, based on the method described in the first embodiment of the present invention, a person skilled in the art can understand the specific structure and the deformation of the apparatus, and thus the details are not described herein. All the devices adopted in the method of the first embodiment of the present invention belong to the protection scope of the present invention.

Based on the same inventive concept, the application provides an electronic device embodiment corresponding to the first embodiment, and the detailed description is given to the third embodiment.

EXAMPLE III

The embodiment provides an electronic device, as shown in fig. 5, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, any one of the first embodiment modes may be implemented.

Since the electronic device described in this embodiment is a device used for implementing the method in the first embodiment of the present application, based on the method described in the first embodiment of the present application, a specific implementation of the electronic device in this embodiment and various variations thereof can be understood by those skilled in the art, and therefore, how to implement the method in the first embodiment of the present application by the electronic device is not described in detail herein. The equipment used by those skilled in the art to implement the method in the embodiments of the present application is all within the protection scope of the present application.

Based on the same inventive concept, the application provides a storage medium corresponding to the fourth embodiment, which is described in detail in the fourth embodiment.

Example four

The present embodiment provides a computer-readable storage medium, as shown in fig. 6, on which a computer program is stored, and when the computer program is executed by a processor, any one of the embodiments can be implemented.

The method, the device, the system, the equipment and the medium provided by the embodiment of the application have at least the following technical effects or advantages: the invention provides a commodity normalization algorithm-based price monitoring system or terminal, which aligns the application requirements of public purchasing and national enterprise purchasing, provides intelligent commodity identification monitoring service by applying an RPA robot process automation technology, constructs commodity aggregation normalization service by applying an algorithm model, accurately positions the same commodity from mass commodity data, replaces a manual repetitive operation mode, and reduces the manual input cost. The special supply commodities are removed, the reasonable price calculation model is provided for calculating the reasonable price of the same commodity, the price change of the same commodity is monitored according to the reasonable price of the commodity, the risk reminding is pushed to the user according to the improper change of the price, the purchasing time cost of an enterprise is reduced, and the purchasing efficiency is improved. According to the commodity data analysis method, by means of feature extraction of technologies such as big data and artificial intelligence and implementation of a reasonable price algorithm, commodity data are interpreted from two aspects of dimensional modeling and data indexes, digital, automatic and intelligent supervision in the public purchasing field is assisted, and the data generate value.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, apparatus or system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While specific embodiments of the invention have been described, it will be understood by those skilled in the art that the specific embodiments described are illustrative only and are not limiting upon the scope of the invention, as equivalent modifications and variations as will be made by those skilled in the art in light of the spirit of the invention are intended to be included within the scope of the appended claims.

Claims (10)

1. A price monitoring method for E-commerce commodities is characterized by comprising the following steps: the method comprises the following steps:

product normalization: the method comprises the steps that an RPA robot process automation technology is adopted, and the commodities with different sources and different description standards are aggregated and identified according to uniform commodity attribute standards and commodity similarity values to obtain the same commodity of each category;

and (3) commodity coding: printing a corresponding category code for the same commodity and storing the category code in a commodity pool, wherein the category code comprises a major category number, a middle category number, a minor category number and a serial number;

monitoring of special commodities: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology, wherein the commodity information comprises commodity names and brand information; regularly visiting a main-electricity commercial platform and a commodity source website, setting a monitoring range, inputting extracted commodity information to perform matching search, judging that the commodity is a special supply commodity if the matching search is not successful in the monitoring range and the matching search is still not successful after tracing the commodity source website, then putting off the shelf of the special supply commodity, and generating a suspected special supply commodity list;

and (4) reasonable price monitoring: triggering a regular task, extracting price information of the same commodity of the commodity in the commodity pool according to the category code, transmitting the price information to a reasonable price algorithm model, and calculating the reasonable price of the commodity; saving the reasonable price calculation result of the commodity;

monitoring commodity price change: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology; when the price of the commodity changes, accessing the commodity link and extracting commodity price information; judging the price change condition of the goods on the shelf, and if the price of the goods is reduced, automatically reducing by default; if the price of the commodity is adjusted upwards, the price is compared with the reasonable price of the commodity and the e-commerce official network price of the commodity, and if the adjusted price is higher than the reasonable price or the e-commerce official network price, the risk reminding is pushed to the user.

2. The price monitoring method for the electronic commerce goods as claimed in claim 1, wherein: the reasonable price calculation process of the commodity by the reasonable price algorithm model comprises the following steps:

when the extracted price information of the same commodity is only 1 price, the price information is directly used as a market reasonable price;

when the extracted price information of the same commodity has 2-7 prices, taking the average number as the market reasonable price;

and when the extracted price information of the same commodity has more than 7 prices, performing effective price screening to eliminate abnormal prices, taking 95% of data as effective confidence interval prices at the rest prices, performing quartile arrangement, and taking the first 50% of prices to calculate an average price as a market reasonable price.

3. The method of claim 2, wherein: the effective price is screened as follows:

arranging all prices from small to large, and finding out a median price m yuan; the median price m determines a coefficient x, and the relationship between the current median price m and the coefficient x is as follows:

m > =5000 yuan, x =0.3;

2000> < m > < 5000 yuan, x =0.5;

when m < 2000-membered, x =1;

prices greater than m (1 + x) or less than m (1-x) are excluded as abnormal prices.

4. The method of claim 1, wherein: the commodity normalization specifically comprises the following steps:

extracting the core fields of the commodities by adopting an automatic brand extraction model, an automatic model extraction model and a category prediction model;

adopting a commodity semantic recall model, a brand identity relationship table and a category identity relationship table to pre-divide commodities by SPU;

adopting a commodity similarity judgment model to perform SPU division on commodities;

adopting a key attribute automatic extraction model to divide SKU (stock keeping unit) into the key attributes of the commodities;

the implementation of the category prediction model specifically includes:

(1) An environment building process, namely downloading a bert pre-training language model bert-base-chip; pre-training the downloaded bert pre-training language model bert-base-chip through massive non-labeled corpora to obtain a pre-trained bert model; the mass unmarked corpus is not marked as a commodity title;

(2) In the model fine tuning process, a category label set file class.txt is constructed based on the criteria of a category library and is used for reading a pre-trained bert model; performing one-hot coding on the item list to form a label matrix; selecting a run _ classifier. Associating a data set file, a category tag set file class txt and an initial model parameter list in the run _ classifier. Reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest reasoning capability of the classifier as an optimal model, and storing a model parameter list of the optimal model;

(3) A model evaluation process, wherein the optimal model is operated, and a model parameter list of the optimal model is loaded; inputting a test data set to check the confidence of an output result, checking the accuracy of the model for classifying commodities, if the accuracy is lower than a threshold value, reconfiguring a learning rate value and a model parameter list and then carrying out model fine tuning until the accuracy is not lower than the threshold value, and then obtaining a bert model which is the category prediction model;

the downstream classification task is used for dividing the commodity title information text into a plurality of commodity categories; the labeling data set C is labeled with a commodity title and a commodity label; the model parameter list comprises related parameters of commodity attributes, names and brands; the test data set is a data set of a certain amount of commodity title information.

5. A price monitoring device of E-commerce commodity is characterized in that: the method comprises the following steps:

a commodity normalization module: the method comprises the steps that an RPA robot process automation technology is adopted, and the commodities with different sources and different description standards are aggregated and identified according to a uniform commodity attribute standard and a commodity similarity value, so that the same commodity of each category is obtained;

the commodity code printing module: printing a corresponding category code for the same commodity and storing the category code in a commodity pool, wherein the category code comprises a major category number, a middle category number, a minor category number and a serial number;

the special supply commodity monitoring module: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology, wherein the commodity information comprises commodity names and brand information; regularly visiting a main-electricity merchant platform and a commodity source website, setting a monitoring range, inputting extracted commodity information for matching search, judging that the commodity is a special supply commodity if the matching search is not successful in the monitoring range and the matching search is still not successful after tracing the commodity source website, and then putting off the shelf for the special supply commodity and generating a suspected special supply commodity list;

reasonable price monitoring module: triggering a regular task, extracting price information of the same commodity of the commodity in the commodity pool according to the category code, transmitting the price information to a reasonable price algorithm model, and calculating the reasonable price of the commodity; saving the reasonable price calculation result of the commodity;

commodity price change monitoring module: establishing a commodity information extraction flow by adopting an RPA robot flow automation technology; when the price of the commodity changes, accessing the commodity link and extracting commodity price information; judging the price change condition of the goods on the shelf, and if the price of the goods is reduced, automatically reducing by default; if the price of the commodity is adjusted upwards, the price is compared with the reasonable price of the commodity and the e-commerce official website price of the commodity, and if the price after the adjustment is higher than the reasonable price, the risk reminding is pushed to the user.

6. The apparatus of claim 5, wherein: the process of calculating the reasonable price of the commodity by the reasonable price algorithm model comprises the following steps:

when the extracted price information of the same commodity is only 1 price, the price information is directly used as a market reasonable price;

when the extracted price information of the same commodity has 2-7 prices, taking the average number as the market reasonable price;

and when the extracted price information of the same commodity has more than 7 prices, performing effective price screening to eliminate abnormal prices, taking 95% of data as effective confidence interval prices at the rest prices, performing quartile arrangement, and taking the first 50% of prices to calculate an average price as a market reasonable price.

7. The apparatus of claim 6, wherein: the effective price is screened as follows:

arranging all prices from small to large, and finding out a median price m yuan; the median price m determines a coefficient x, and the relationship between the current median price m and the coefficient x is as follows:

m > =5000 yuan, x =0.3;

2000> < m >5000 yuan, x =0.5;

when m < 2000-membered, x =1;

prices greater than m (1 + x) or less than m (1-x) are excluded as abnormal prices.

8. The apparatus of claim 5, wherein: the commodity normalization specifically comprises the following steps:

extracting the core fields of the commodities by adopting an automatic brand extraction model, an automatic model extraction model and a category prediction model;

adopting a commodity semantic recall model, a brand identity relationship table and a category identity relationship table to pre-divide commodities by SPU;

adopting a commodity similarity judgment model to perform SPU division on commodities;

adopting a key attribute automatic extraction model to divide SKU (stock keeping unit) into the key attributes of the commodities;

the implementation of the category prediction model specifically includes:

(1) An environment building process, namely downloading a bert pre-training language model bert-base-chip; pre-training the downloaded bert pre-training language model bert-base-chip through massive non-labeled corpora to obtain a pre-trained bert model; the mass unmarked corpus is not marked as a commodity title;

(2) A model fine adjustment process, namely constructing a category label set file class.txt based on a category library standard for a pre-trained bert model to read; performing one-hot coding on the item list to form a label matrix; selecting a run _ classifier. Associating a data set file, a category tag set file class txt and an initial model parameter list in the run _ classifier. Reconfiguring a learning rate parameter value and a model parameter list, continuously operating the classifier, performing reasoning, and repeating the steps to obtain a plurality of bert models with different reasoning capabilities; selecting a bert model with the highest inference capability of a classifier as an optimal model, and storing a model parameter list of the optimal model;

(3) A model evaluation process, wherein the optimal model is operated, and a model parameter list of the optimal model is loaded; inputting a test data set to check the confidence of an output result, checking the accuracy of the model for classifying commodities, if the accuracy is lower than a threshold value, reconfiguring a learning rate value and a model parameter list and then carrying out model fine tuning until the accuracy is not lower than the threshold value, and then obtaining a bert model which is the category prediction model;

the downstream classification task is used for dividing the commodity title information text into a plurality of commodity categories; the labeling data set C is labeled with a commodity title and a commodity label; the model parameter list comprises related parameters of commodity attributes, names and brands; the test data set is a data set of a certain amount of commodity title information.

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 4 when executing the program.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 4.

Images