Agricultural product transaction data processing method and device based on multi-index linkage analysis

Abstract

The invention discloses an agricultural product transaction data processing method and device based on multi-index linkage analysis, and the method comprises the steps: obtaining a plurality of transaction data of a target agricultural product transaction, and the transaction data comprises the foreign index value, transaction volume and transaction price of the target agricultural product; processing the transaction data based on a pre-constructed price expectation model, and giving an expectation transaction price corresponding to the transaction data; matching a corresponding data processing model by analyzing an expected transaction price; and processing the transaction data by combining the expected transaction price and based on the data processing model. According to the agricultural product transaction data analysis method, the transaction data are analyzed and processed through the analysis processes of price expectation and data processing in sequence, and the rational transaction information and the irrational transaction information in the transaction data are extracted, so that the accuracy of agricultural product transaction data analysis is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural product data processing, and particularly relates to a method and device for processing agricultural product transaction data based on multi-index linkage analysis. Background Art

[0002] In recent years, although agricultural products have gradually become standardized and branded, wholesale and farmers' markets are still important links in the circulation of agricultural products. Generally speaking, the transaction information in agricultural product wholesale and farmers' markets can generate big data, which plays an important role in aspects such as forming prices, transmitting information, providing services, precision marketing, and food safety traceability. To improve the management and service levels of market transactions in agricultural product wholesale and farmers' markets, timely understand the market conditions, and support the real-time business decisions of merchants in the market, many markets have enabled electronic settlement management systems, that is, automatically record tens of thousands of transaction information every day, or timely take pictures of paper transaction vouchers and upload them to the system to form electronic transaction data.

[0003] Electronic transaction data has advantages such as real-time, objectivity, and public transparency. Electronic transaction data will record data generated by irrational behaviors such as "herd effect" and "price discrimination" in product transactions, which can also be called "irrational transaction information". For example, the promotions of a small number of wholesalers at the opening of the wholesale market and the selling off of near-end-of-inventory goods will lead to the formation of "irrational transaction information". Currently, the analysis of electronic transaction data does not fully pay attention to and use this "irrational transaction information".

[0004] The processing method of electronic transaction data that does not distinguish between "rational transaction information" and "irrational transaction information" will cause the analysis of electronic transaction data to be affected by "irrational transaction information", resulting in deviation of data analysis. This deviation of data analysis will have a great impact on the analysis of market conditions, especially for fresh agricultural products that are sensitive to time period prices.

[0005] Therefore, how to extract rational transaction information and irrational transaction information from a large amount of electronic transaction data to improve the accuracy of product electronic transaction data analysis is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] In view of the defects existing in the above-mentioned prior art, the present invention provides a method and device for processing agricultural product transaction data based on multi-index linkage analysis. The method includes: obtaining a plurality of transaction data of target agricultural product transactions, where the transaction data includes exogenous index values, trading volumes, and transaction prices of the target agricultural products; processing the transaction data based on a pre-constructed price expectation model to give the expected transaction price corresponding to the transaction data; analyzing the expected transaction price to match the corresponding data processing model; and completing the processing of the transaction data by combining the expected transaction price and based on the data processing model. The present invention analyzes the transaction data through the processes of price expectation and data processing in sequence, extracts rational transaction information and irrational transaction information in the transaction data, so as to improve the accuracy of analyzing agricultural product transaction data.

[0007] In a first aspect, the present invention provides a method for processing agricultural product transaction data based on multi-index linkage analysis, specifically including the following steps:

[0008] Obtain a plurality of transaction data of target agricultural product transactions, where the transaction data includes exogenous index values, trading volumes, and transaction prices of the target agricultural products;

[0009] Process the transaction data based on a pre-constructed price expectation model to give the expected transaction price corresponding to the transaction data;

[0010] Analyze the expected transaction price to match the corresponding data processing model;

[0011] Complete the processing of the transaction data by combining the expected transaction price and based on the data processing model.

[0012] Further, processing the transaction data based on a pre-constructed price expectation model to give the expected transaction price corresponding to the transaction price specifically includes the following steps:

[0013] Divide the multiple exogenous indexes of the target agricultural product to determine continuous indexes and classification indexes;

[0014] Based on the classification index values, divide the transaction data into grades and calibrate the classification grades of the transaction data;

[0015] Through the first hook function, combine the classification grades of the transaction data, match the price expectation models of each classification grade, and give the expected transaction price corresponding to the transaction data, where the price expectation models of each classification grade are iteratively trained from the initial price expectation model through the transaction data of each classification grade;

[0016] Among them, the construction of the price expectation models of each classification grade specifically includes:

[0017] Obtain the initial price expectation model and the transaction data of each classification grade;

[0018] Based on the processing of transaction data at each classification level using the initial price expectation model, the expected transaction prices corresponding to the transaction data at each classification level are formed.

[0019] Combining the transaction data at each classification level and the expected transaction prices corresponding to the transaction data at each classification level, iterative processing is performed on the initial price expectation model at each classification level to form the initial price expectation model at each classification level.

[0020] Continue to iterate the initial price expectation model at each classification level through the transaction data and expected transaction prices at the previous classification level, and repeat this iterative step to form the price expectation model at each classification level.

[0021] Furthermore, the construction of the initial price expectation model specifically includes the following steps:

[0022] Obtain transaction sample data, group the transaction sample data to form a transaction sample data module.

[0023] Perform factor analysis on each continuous index value of each transaction sample data module, and screen and form multiple sentiment factors for the target agricultural product transaction.

[0024] Based on the correlation analysis of the classification indicators of the target agricultural product and multiple sentiment factors, construct the initial price expectation model for each transaction sample data module.

[0025] Based on the mean processing of the initial price expectation models of each transaction sample data module, give the initial price expectation model of the target agricultural product.

[0026] Furthermore, performing factor analysis on each continuous index value of each transaction sample data module, and screening and forming multiple sentiment factors for the target agricultural product transaction specifically includes the following steps:

[0027] Through each continuous index value of each transaction sample data module, screen the continuous indexes to obtain multiple common factors and form a common factor set.

[0028] Based on a preset parameter matrix and the common factor set, construct a sentiment factor model, where the sentiment factor model is specifically expressed as:

[0029] X p =μ + A·F q + e

[0030] Among them, X p is the continuous index set of p continuous indexes, μ is the preset parameter matrix, A is the loading matrix, F q is the common factor set of q common factors, and e is a q-dimensional random variable;

[0031] Through the sentiment factor model, load analysis is performed on each common factor to obtain the explanatory power of each common factor;

[0032] Based on the explanatory power of each common factor, each common factor is sorted;

[0033] With the strategy that the cumulative explanatory power reaches the threshold and the number of common factors is the least, the target common factor is determined as the sentiment factor for the target agricultural product transaction.

[0034] Furthermore, based on the correlation analysis of the classification indicators of the target agricultural product and multiple sentiment factors, an initial price expectation model for each transaction sample data module is constructed, specifically including:

[0035] For each transaction sample data module, a relationship model between the transaction price of the target agricultural product, the classification indicator, and the sentiment factor is constructed respectively, where the parameters in the relationship model are obtained through parameter estimation;

[0036] According to the parameters in the relationship model corresponding to each transaction sample data module, the parameters of the initial price expectation model are determined, and the initial price expectation model for each transaction sample data module is constructed;

[0037] The initial price expectation model for each transaction sample data module is specifically expressed as:

[0038]

[0039] Among them, g(u) is the transformation function of the expected transaction price of the target agricultural product, u is the expected transaction price of the target agricultural product, FL s is the classification indicator of the s-th transaction sample data module, f(·) is the parameter function, K i (·) is the i-th non-parametric function, CF si is the i-th sentiment factor of the s-th transaction sample data module, and m is the number of sentiment factors in the s-th transaction sample data module.

[0040] Furthermore, based on the mean processing of the initial price expectation models of each transaction sample data module, the initial price expectation model of the target agricultural product is given, specifically including:

[0041] Perform parameter estimation on the initial price expectation model of each transaction sample data module to obtain the parameter set β s (s = 1, 2... n), where each parameter contains d parameter estimation values;

[0042] Give the parameter estimation value set [(β 11 , β 12 ... β 1d )(β 21 , β22 ...β 2d )...(β n1 、β n2 ...β nd )], and perform an averaging process on the parameter estimation values of each parameter to obtain a set of estimated average values of the parameters

[0043] Replace the parameter estimation values of each parameter in the initial price expectation model of each transaction sample data module with the set of estimated average values of the parameters to give the initial price expectation model of the target agricultural product.

[0044] Furthermore, the data processing model includes a first data processing model and a second data processing model;

[0045] Through the analysis of the expected transaction price, match the corresponding data processing model, which specifically includes the following steps:

[0046] Based on the comparative analysis of the expected transaction price and the transaction price, determine the expected transaction price as a first-class expected transaction price and a second-class expected transaction price;

[0047] Through the second hook function, match the first data processing model and the second data processing model with the first-class expected transaction price and the second-class expected transaction price respectively.

[0048] Furthermore, the first data processing model includes multiple classification-level first data processing models;

[0049] Through the second hook function, match the first data processing model and the second data processing model with the first-class expected transaction price and the second-class expected transaction price respectively, specifically including:

[0050] Through the second hook function, judge the classification level of the transaction data corresponding to the first-class expected transaction price and each classification-level first data processing model to complete the matching of each classification-level first data processing model with the first-class expected transaction price;

[0051] Through the second hook function, match the second-class expected transaction price with the second data processing model.

[0052] Furthermore, combine the expected transaction price and, based on the data processing model, complete the processing of the transaction data, specifically including:

[0053] Based on the first data processing model, perform a weighted average process on the transaction data corresponding to the first-class expected transaction price to obtain the weighted average price of the transaction data;

[0054] Based on the second data processing model, analyze the second-class expected transaction price to obtain the second-class cumulative transaction amount and the second-class cumulative transaction volume corresponding to the transaction data.

[0055] In a second aspect, the present invention further provides an agricultural product trading data processing device based on multi-index linkage analysis, which adopts the agricultural product trading data processing method based on multi-index linkage analysis as described in any one of the above, and includes:

[0056] A data acquisition unit, configured to acquire a plurality of trading data of target agricultural product trading, wherein the trading data includes exogenous index values, trading volumes, and trading prices of the target agricultural product;

[0057] A price determination unit, configured to process the trading data based on a pre-constructed price expectation model and give an expected trading price corresponding to the trading data;

[0058] A model matching unit, configured to match a corresponding data processing model by analyzing the expected trading price;

[0059] A data processing unit, configured to combine the expected trading price and complete the processing of the trading data based on the data processing model.

[0060] The agricultural product trading data processing method and device based on multi-index linkage analysis provided by the present invention have at least the following beneficial effects:

[0061] (1) The present invention analyzes and processes trading data through the analysis processes of price expectation and data processing in sequence, extracts rational trading information and irrational trading information in the trading data, so as to improve the accuracy of analyzing agricultural product trading data.

[0062] (2) The price expectation model constructed by combining the sentiment factor and the classification index eliminates the continuous index values with less influence on the trading price, can accurately reflect the trading price of the target agricultural product, and can also improve the processing efficiency of the price expectation model for trading data.

[0063] (3) The transformation function g(u) is jointly composed of the parameter function f() that correlates the classification index and the sentiment factor, and the non-parametric function K i () for each sentiment factor. Among them, the parameter function has high interpretability, the non-parametric function is relatively flexible, and the integration of the parameter function and the non-parametric function can improve the generalization ability, prediction accuracy, adaptability of the price expectation model, reduce the dependence on the data volume, and at the same time improve the interpretability of the model, especially applicable to the scenario of processing agricultural product trading data with complex relationships. Description of the Drawings

[0064] Figure 1 It is a schematic flowchart of an agricultural product trading data processing method based on multi-index linkage analysis provided by the present invention;

[0065] Figure 2Schematic diagram of the process for giving the expected transaction price provided by the embodiment of the present invention;

[0066] Figure 3 Schematic diagram of the process for constructing the initial price expectation model provided by the embodiment of the present invention;

[0067] Figure 4 Schematic diagram of the process for screening and forming multiple sentiment factors for target agricultural product transactions provided by the embodiment of the present invention;

[0068] Figure 5 Schematic diagram of the process for constructing the initial price expectation model of each transaction sample data module provided by the embodiment of the present invention;

[0069] Figure 6 Example diagram of classification level calibration and data processing provided by the embodiment of the present invention;

[0070] Figure 7 Structural block diagram of a product transaction data processing device based on multi-index linkage analysis provided by the present invention. Detailed implementation manners

[0071] In order to better understand the above technical solutions, the following will describe the above technical solutions in detail in conjunction with the accompanying drawings of the specification and specific implementation manners. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0072] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms of "a", "the" and "said" used in the embodiments of the present invention and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. "Multiple" generally includes at least two.

[0073] It should also be noted that the term "comprises", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a commodity or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such commodity or device. Without further limitation, the element defined by the statement "comprising a..." does not exclude the existence of additional identical elements in the commodity or device including the said element.

[0074] The transaction information in agricultural product wholesale and trading markets can generate big data, which plays an important role in aspects such as price formation, information transmission, service provision, precision marketing, and food safety traceability. However, randomly collected information such as agricultural product wholesale and trading prices will record the "irrational transaction information" generated by irrational behaviors such as "herd effect" and "price discrimination" in product transactions, resulting in deviations in data analysis. For fresh agricultural products with relatively frequent price changes, its accuracy is difficult to guarantee.

[0075] Therefore, the present invention provides a method and device for processing agricultural product transaction data based on multi-index linkage analysis. By analyzing the transaction data, an emotion factor is given, and based on the emotion factor, the real-time obtained transaction data is judged to determine whether the current transaction data is an "irrational transaction". Based on the judgment result, the transaction data is further classified, studied, and analyzed to improve the accuracy of the analysis and research of agricultural product transaction data, provide more accurate information support for market optimization service capabilities, service levels, and pricing mechanisms, and for the macro-control of the market.

[0076] As Figure 1 shown, an embodiment of the present invention provides a method for processing agricultural product transaction data based on multi-index linkage analysis, which specifically includes the following steps:

[0077] S101: Obtain multiple transaction data of the target agricultural product transaction.

[0078] Specifically, the transaction data includes the transaction time period, exogenous index value, transaction volume, and transaction price of the target agricultural product. The target agricultural product can be fresh agricultural products such as pork, vegetables, and fruits.

[0079] Exogenous indicators refer to indicators determined by non-economic factors outside the economic system, including agricultural product self-characteristic indicators, transaction environment indicators, and market information indicators that have an impact on prices. Agricultural product self-characteristic indicators include quality specifications (QS), source of goods (SOG), etc.; transaction environment indicators include suppliers (SU), dealers (AP), market types (MC), etc.; market information indicators include net change in yesterday's inventory (NICY), yesterday's transaction price (NICP), net change in the day before yesterday's inventory (NICYPD), day before yesterday's transaction price (NICPPD), yesterday's entry volume (YEV), today's entry volume (TEV), today's already-transacted price (TCP), today's already-transacted quantity (TCV), etc. Of course, exogenous indicators also include indicators related to the confidence index of both buyers and sellers. For example, the consumer confidence index (CCI), online public opinion sentiment index (NPOSI).

[0080] In a specific example, taking the target agricultural product as fresh agricultural product - white-striped pork, the multiple transaction data of the target agricultural product transaction can be the transaction data for a one-week period, with a total of 35,345 transaction data. Each transaction data covers transaction time, corresponding exogenous index values, transaction volume, transaction price, etc. It can also be roughly understood from the one-week transaction data of white-striped pork that the transactions are mainly concentrated from 0:00 to 6:00 every day, and the transaction density is relatively high.

[0081] S102: Process the transaction data based on a pre-constructed price expectation model, and give the expected transaction price corresponding to the transaction data.

[0082] By judging the expected transaction price of the transaction data, the "rational transaction information" and "irrational transaction information" in the transaction data can be extracted, and they are respectively analyzed and processed to eliminate the deviation in the analysis of the transaction data.

[0083] Specifically, as Figure 2 shown, processing the transaction data based on a pre-constructed price expectation model and giving the expected transaction price corresponding to the transaction price specifically includes the following steps:

[0084] Classify the multiple exogenous indicators of the target agricultural product to determine continuous indicators and classification indicators;

[0085] Based on the classification indicator values, classify the transaction data and calibrate the classification levels of the transaction data;

[0086] Through the first hook function, combined with the classification levels of the transaction data, match the price expectation models for each classification level, and give the expected transaction price corresponding to the transaction data.

[0087] For a certain target agricultural product, the corresponding exogenous indicators are usually numerous, diverse in type, and relatively strongly correlated. However, some exogenous indicators are mainly descriptions and characterizations of the agricultural product category, with a relatively stable impact on price and usually no immediate changes in the impact effect. Some other exogenous indicators, on the other hand, are quite sensitive to the impact on price.

[0088] These exogenous indicators are then classified. One type is the classification indicator (FL), which refers to the indicator that describes or characterizes the categories of agricultural products, that is, the name indicator that explains the categories of agricultural products, and the value taken is the classification indicator value, including quality grading (QS), source of goods (SOG), supplier (SU), distributor (AP), market type (MC), etc.; the other type is the continuous indicator, which refers to the indicator that has an immediate impact on the price of agricultural products, and the value taken is the continuous indicator value, which is approximately normally distributed under the condition of big data, such as unit output weight (DZ), proportion of defective products in random inspection (CC), net change in inventory yesterday (NICY), transaction price yesterday (NICP), net change in inventory the day before yesterday (NICYPD), transaction price the day before yesterday (NICPPD), quantity of goods entering the market yesterday (YEV), quantity of goods entering the market today (TEV), transaction price already completed today (TCP), quantity already transacted today (TCV), consumer confidence index (CCI), online public opinion sentiment index (NPOSI), etc.

[0089] According to the different classification indicator values, different transaction data are classified and graded, and the classification grade calibration is given. According to the classification grade of the transaction data, it is matched with each pre-constructed price expectation model for each classification grade, and finally the expected transaction price corresponding to the transaction data is given. Among them, the first hook function used for matching each classification grade of transaction data with each pre-constructed price expectation model for each classification grade mainly plays a matching role. The type of the first hook function is not limited here. It can be a common type function or an anonymous function. As long as the programming mechanism or design pattern based on the first hook function can realize the matching of transaction data and the price pre-fetching model.

[0090] Among them, each classification grade price expectation model is obtained by iteratively training the initial price expectation model with each classification grade of transaction data.

[0091] For the construction of each classification grade price expectation model, it specifically includes:

[0092] Obtain the initial price expectation model and the transaction data for each classification grade;

[0093] Based on the processing of each classification grade of transaction data by the initial price expectation model, the expected transaction price corresponding to each classification grade of transaction data is formed;

[0094] Combining each classification grade of transaction data and the expected transaction price corresponding to each classification grade of transaction data, the initial price expectation model is iteratively processed for each classification grade respectively to form the initial price expectation model for each classification grade;

[0095] Continue to iterate the initial price expectation model for each classification grade through the transaction data and expected transaction price of the previous time, and repeat this iterative step to form the price expectation model for each classification grade.

[0096] Trading data with different classification index values has different market cycles and volatility rules. When obtaining the expected trading price, different classification level price expectation models need to be adopted. The acquisition of each different classification level price expectation model is iteratively trained from the initial price expectation model.

[0097] The data set used for iterative training is the trading data of each classification level and the expected trading price formed for the current trading data of each classification level. Based on the initial price expectation models of each classification level initially formed, the model coefficients are updated, and the above update and iteration process is repeated repeatedly until the price expectation models of each classification level are finally formed. The specific convergence condition for iterative training can be the accuracy rate or the number of iterations, which is not specifically limited here.

[0098] As Figure 3 shown, for the construction of the initial price expectation model, the following steps are specifically included:

[0099] Obtain trading sample data, group the trading sample data, and form trading sample data modules;

[0100] Perform factor analysis on each continuous index value of each trading sample data module, and screen and form multiple sentiment factors for the target agricultural product trading;

[0101] Based on the correlation analysis of the classification index of the target agricultural product and multiple sentiment factors, construct the initial price expectation model of each trading sample data module;

[0102] Based on the verification of the initial price expectation models of each trading sample data module, give the initial price expectation model of the target agricultural product.

[0103] Randomly split a trading sample data of the target agricultural product into n groups, each group is a trading sample data module, and each trading sample data module contains exogenous index values (including classification index values and continuous index values) that affect the price of the target agricultural product and the corresponding trading price and trading volume of the target agricultural product.

[0104] In view of the fact that the classification index is mainly a description and characterization of the agricultural product category, and its influence on the price is relatively stable and usually does not show an immediate change in the price, when considering the change of sentiment, only the continuous index needs to be factor analyzed.

[0105] In a specific example, the continuous indexes considering the change of sentiment include the entry information and trading information based on the previous day and the previous yesterday, the inbound and outbound volume of dealers on the previous day, the inbound and outbound volume of dealers on the previous yesterday, etc.

[0106] Using a parallel algorithm, perform factor analysis on the continuous indicator values in each trading sample data module to obtain multiple sentiment factors, and then combine the classification indicators for the trading sample data, establish classification indicators, multiple sentiment factors, and an initial price expectation model for the expected trading price of the target agricultural product.

[0107] As Figure 4 shown, perform factor analysis on the respective continuous indicator values of each trading sample data module, screen and form multiple sentiment factors for the trading of the target agricultural product, specifically including:

[0108] Through the respective continuous indicator values of each trading sample data module, screen the continuous indicators to obtain multiple common factors and form a common factor set;

[0109] Based on a preset parameter matrix and the common factor set, construct a sentiment factor model, where the sentiment factor model is specifically expressed as:

[0110] X p = μ + A·F q + e

[0111] where, X p is the continuous indicator set of p continuous indicators, μ is the preset parameter matrix, A is the loading matrix, F q is the common factor set of q common factors, and e is a q-dimensional random variable;

[0112] Through the sentiment factor model, perform loading analysis on each common factor to obtain the explanatory power of each common factor;

[0113] Based on the explanatory power of each common factor, sort each common factor;

[0114] With the strategy that the cumulative explanatory power reaches the threshold and the number of common factors is the least, determine the target common factor as the sentiment factor for the trading of the target agricultural product.

[0115] In a certain embodiment, a certain trading sample data module includes p continuous indicators, and the formed continuous indicator set is X p . The p continuous indicators are the external manifestations of q common factors in this trading sample data module. One or more of the q common factors have an impact on the trading data, and these common factors that have an impact on the trading data are the target common factors, that is, the sentiment factors.

[0116] Therefore, mining the target common factors that affect the trading data from the q common factors, which are the sentiment factors, and constructing a price expectation model by combining the sentiment factors with the classification indicators can eliminate the continuous indicator values that have a relatively small impact on the trading price, accurately reflect the trading price of the target agricultural product, and improve the processing efficiency of the price expectation model for trading data.

[0117] To mine the common factors that affect transaction data from q common factors, a sentiment factor model is required. The establishment of the sentiment factor model is based on a parameter matrix, a loading matrix, and a set of common factors. Among them, the parameter matrix μ is preset in advance according to different scenarios, and the loading matrix A is composed of various loading elements. Each loading element is the loading of a certain common factor relative to a certain continuous indicator. The loading of a certain common factor relative to all continuous indicators is the explanatory power of this common factor. The explanatory power refers to the ability to explain the data variability, that is, the higher the explanatory power of a certain common factor, the greater the impact of this common factor on the transaction sample data module.

[0118] Through the sentiment factor model, load analysis is performed on each common factor to obtain the explanatory power of each common factor, which is specifically expressed as:

[0119]

[0120] Among them, is the explanatory power of the xth common factor, a xy is the loading of the xth common factor relative to the yth continuous indicator, p is the number of continuous indicators, and q is the number of common factors.

[0121] The common factors are sorted from largest to smallest according to the explanatory power, and the first m common factors with the cumulative explanatory power reaching the threshold and the smallest number of common factors are selected as the sentiment factors. The threshold is preset in advance, and the value of the threshold is related to the specific application scenario and no specific numerical limit is made.

[0122] Before constructing the initial price expectation model for each transaction sample data module, it also includes the verification of multiple sentiment factors, specifically including:

[0123] Set the influence parameters and establish a verification model for each continuous indicator on each sentiment factor. Among them, the verification model is specifically expressed as:

[0124]

[0125] Among them, CF x is the xth sentiment factor, p is the number of sentiment factors, b xy is the influence parameter of the yth continuous indicator on the xth sentiment factor, X y is the yth continuous indicator, C x is the adjustment parameter of the xth sentiment factor, p is the number of continuous indicators, and m is the number of sentiment factors;

[0126] Perform a significance test on the influence parameters. If it is non-zero, it is determined that the verification of this sentiment factor fails.

[0127] Cx is a regulatory parameter related to the parameter matrix μ and the loading matrix A in the sentiment factor model. The regulatory parameters of each sentiment factor are different. If all influence parameters are zero, it indicates that the sentiment factor can reflect each continuous indicator. If the influence parameter is not zero, it means that the sentiment factor does not strongly reflect each continuous indicator, cannot support the external manifestation of the continuous indicator, and cannot pass the significance verification.

[0128] As Figure 5 shown, based on the correlation analysis of the classification indicators of the target agricultural product and multiple sentiment factors, an initial price expectation model for each transaction sample data module is constructed, specifically including:

[0129] For each transaction sample data module, a relationship model between the transaction price of the target agricultural product, the classification indicator, and the sentiment factor is constructed respectively, where the parameters in the relationship model are obtained through parameter estimation;

[0130] According to the parameters in the relationship model corresponding to each transaction sample data module, the parameters of the initial price expectation model are determined, and the initial price expectation model for each transaction sample data module is constructed;

[0131] The initial price expectation model for each transaction sample data module is specifically expressed as:

[0132]

[0133] where g(u) is the transformation function of the expected transaction price of the target agricultural product, u is the expected transaction price of the target agricultural product, FL s is the classification indicator of the s-th transaction sample data module, f() is the parameter function, K i () is the i-th non-parametric function, CF si is the i-th sentiment factor of the s-th transaction sample data module, and m is the number of sentiment factors in the s-th transaction sample data module.

[0134] In a specific example, after determining the sentiment factor, an initial price expectation model of the target agricultural product is constructed by combining the classification indicator and the sentiment factor. That is, it should be understood that when constructing the initial price expectation model, the situation where the target agricultural product is affected by continuous indicators and sentiment factors is mainly considered.

[0135] In the initial price expectation model, the transformation function g(u) presents the expected transaction price u using the g() change function according to the value density distribution of the transaction data (mainly the transaction price P). The type of the transformation function g(u) will be different according to different transaction data, and no specific limitation is made here. Similarly, the parameter function f() can be a function following the normal distribution law or a function following the Poisson distribution, and the non-parametric function K i() is an estimation model based on the data itself, which can be kernel density estimation or K-nearest neighbor algorithm, etc. The parametric function f() and the non-parametric function K i () types will vary according to different transaction data and are not specifically limited here.

[0136] For example, when the transaction price P follows a normal distribution, g(u) = u; when the transaction price P follows a 0-1 distribution, g() can adopt the logit function, and at this time g(u) = logit(u) = P(u = 1|X).

[0137] The transformation function g(u) is composed of the parametric function f() that correlates classification indicators and sentiment factors, and the non-parametric function K i () together. Among them, the parametric function has high interpretability, and the non-parametric function is relatively flexible. Combining the parametric function and the non-parametric function can improve the generalization ability, prediction accuracy, adaptability of the price expectation model, reduce the dependence on the amount of data, and at the same time improve the interpretability of the model, especially suitable for the scenario of processing agricultural product transaction data with complex relationships.

[0138] Analyze each transaction sample data module using the above initial price expectation model. When the model forms given by each transaction sample data module are stable and unified, it is determined as the final model form, that is, the initial price expectation model is obtained.

[0139] Based on the mean processing of the initial price expectation models of each transaction sample data module, the initial price expectation model of the target agricultural product is given, specifically including:

[0140] Perform parameter estimation on the initial price expectation model of each transaction sample data module to obtain the parameter set β s (s = 1, 2...n), where each parameter contains d parameter estimation values;

[0141] Give the parameter estimation value set [(β 11 , β 12 ...β 1d )(β 21 , β 22 ...β 2d )...(β n1 , β n2 ...β nd )], and perform an averaging process on the parameter estimation values of each parameter to obtain the estimated average value set of the parameters

[0142] Replace the parameter estimation values of each parameter in the initial price expectation model of each transaction sample data module with the estimated average value set of the parameters to give the initial price expectation model of the target agricultural product.

[0143] S103: Match the corresponding data processing model by analyzing the expected transaction price.

[0144] Bring the transaction data generated by the target agricultural product into the price expectation model, and the expected transaction price will be obtained. By comparing the expected transaction price with the transaction price in the transaction data, it can be determined whether this transaction data is "emotionally abnormal transaction".

[0145] By comparing the expected transaction price with the transaction price in the transaction data, it is mainly to judge the gap between the expected transaction price and the actual transaction price in the transaction data, and determine whether the expected transaction price is a type-I expected transaction price or a type-II expected transaction price according to the gap. Different analysis and processing are performed on the transaction data corresponding to the type-I expected transaction price and the type-II expected transaction price respectively.

[0146] The processing of the transaction data in combination with the expected transaction price adopts a data processing model, and the data processing model includes a first data processing model and a second data processing model;

[0147] Match the corresponding data processing model by analyzing the expected transaction price, which specifically includes the following steps:

[0148] Based on the comparative analysis of the expected transaction price and the transaction price, determine the expected transaction price as a type-I expected transaction price and a type-II expected transaction price;

[0149] Through the second hook function, match the type-I expected transaction price and the type-II expected transaction price with the first data processing model and the second data processing model respectively.

[0150] The second hook function mainly plays a role in matching. The type of the second hook function is not limited here. It can be a common type function or an anonymous function. As long as the programming mechanism or design pattern based on the second hook function can realize the matching of the transaction data and the data processing model.

[0151] Based on the comparative analysis of the expected transaction price and the transaction price, determine the expected transaction price as a type-I expected transaction price and a type-II expected transaction price, and the comparison can be carried out by setting a gap threshold.

[0152] For example, in a specific example, if then it is considered that this transaction data is generated by an "emotionally abnormal transaction", that is, the expected transaction price is a type-II expected transaction price. Otherwise, it is determined as an "emotionally normal transaction", that is, the expected transaction price is a type-I expected transaction price.

[0153] Through the above steps, the transaction data can be divided into "normal-emotion transactions" and "abnormal-emotion transactions", that is, the transaction data corresponding to the first type of expected transaction price is normal-emotion transactions, and the transaction data corresponding to the second type of expected transaction price is abnormal-emotion transactions.

[0154] When analyzing and processing the transaction data corresponding to the first type of expected transaction price, it is also necessary to consider the classification levels determined by the classification indicators of the transaction data. Based on the differences in the classification levels, the first data processing model includes multiple classification-level first data processing models;

[0155] Through the second hook function, the first data processing model and the second data processing model are respectively matched with the first type of expected transaction price and the second type of expected transaction price, specifically including:

[0156] Through the second hook function, judge the classification level of the transaction data corresponding to the first type of expected transaction price and each classification-level first data processing model, and complete the matching of each classification-level first data processing model with the first type of expected transaction price;

[0157] Through the second hook function, match the second type of expected transaction price with the second data processing model.

[0158] Specifically, through the second hook function, judge the classification level corresponding to the first type of expected transaction price and the classification level corresponding to the first data processing model, and complete the matching of the first data processing model with the first type of expected transaction price. Through the second hook function, judge the second type of expected transaction price and the second data processing model, and complete the matching of the second data processing model with the second type of expected transaction price.

[0159] S104: Combine the expected transaction price and, based on the data processing model, complete the processing of the transaction data.

[0160] Specifically, it includes:

[0161] Based on the first data processing model, perform weighted average processing on the transaction data corresponding to the first type of expected transaction price to obtain the weighted average price of the transaction data;

[0162] Based on the second data processing model, analyze the second type of expected transaction price to obtain the second cumulative transaction amount and the second cumulative transaction volume corresponding to the transaction data.

[0163] Among them, the weighted average price of the transaction data = the first cumulative transaction amount / the first cumulative transaction volume.

[0164] The first data processing model is used to count the first cumulative transaction amount and the first cumulative transaction volume corresponding to the first type of expected transaction price, and perform weighted average processing on the transaction data corresponding to the first type of expected transaction price to obtain the weighted average price of the transaction data.

[0165] The second data processing model is used to count the cumulative trading volume of the second category and the cumulative trading volume of the second category.

[0166] The first data processing model and the second data processing model can be conventional data analysis and data statistical functions, and no specific model type is limited.

[0167] For example, in a specific implementation, through the analysis of the expected transaction price, the corresponding data processing model is matched, and then combined with the expected transaction price, and based on the data processing model, the processing of the transaction data is completed. The specific steps are as follows:

[0168] First, judge the expected transaction price. If the transaction data corresponds to "emotionally abnormal transactions", the expected transaction price is defined as the second-category expected transaction price and matched to the second data processing model for processing. Conversely, if the transaction data corresponds to "emotionally normal transactions", the expected transaction price is defined as the first-category expected transaction price and matched to the first data processing model for processing.

[0169] There are multiple first data processing models according to different classification levels of the target agricultural products. Therefore, after determining that it is an emotionally normal transaction, it is necessary to judge the classification level corresponding to the transaction data and bring the transaction data into the first data processing models with different classification levels respectively. The first data processing model calculates the cumulative trading volume and cumulative trading volume of different suppliers of the target agricultural products respectively according to the obtained transaction data, and obtains the weighted average price through weighted processing. The second data processing model calculates the cumulative trading volume and cumulative trading volume of the target agricultural products in emotionally abnormal transactions, that is, the cumulative trading volume of the second category and the cumulative trading volume of the second category. Finally, the data obtained by processing the above transaction data is presented and displayed.

[0170] Taking the white-striped pork of fresh agricultural products as an example, the transaction data of white-striped pork is processed, specifically referring to Figure 6, in this example, there is only one classification index, i.e., the quality grade, which has a significant impact on the price of white-striped pork. However, the specific slaughtering enterprise has no significant impact on the pork price. Therefore, the quality grade is used as the classification level to process the data of white-striped pork. Among them, the Reduce2 process is to sort out the transaction information, cumulative trading volume, and trading amount of the "emotionally abnormal transactions" of each slaughtering enterprise. Given that the trading time is an important and special index affecting the electronic settlement price and trading volume of white-striped pork in the wholesale market. For example, the concentrated trading period of white-striped pork is from 0 to 6 o'clock every morning. At the beginning of the trading, the sellers are reluctant to sell, the number of buyers is small, and it is easy for both buyers and sellers to deadlock, resulting in a small trading volume and the trading price being greatly affected by the previous day's trading information. As the trading time continues, the number of buyers increases, the sellers' emotions gradually stabilize, the trading volume increases, and the fluctuation range of the trading price gradually narrows. At the end of the trading, because the value of white-striped pork decreases greatly after entering the cold storage, most sellers will choose to promote sales at a reduced price.

[0171] The processing of the transaction data is as follows:

[0172] The transaction data of white-striped pork are respectively input into the stream processing nodes 1-1...1-n. In each stream processing node, the Map1 process is executed - classifying each piece of data according to the quality grade, and then obtaining the intermediate data. In this example, white-striped pork can be divided into four grades.

[0173] Based on the first hook function hook1, the obtained intermediate data are divided into the stream processing nodes 2-1, 2-2, 2-3, 2-4, where hook1 is the link relationship between the transaction data and the quality grade.

[0174] The stream processing nodes 2-1, 2-2, 2-3, 2-4 execute the Map2 process - performing discriminant analysis using the price expectation model:

[0175] First, each piece of transaction data is input into the price expectation model to obtain the expected transaction price; then, judge the gap between the expected transaction price and the actual transaction price in the transaction data. If it is considered that the transaction is an "emotionally abnormal transaction", otherwise it is judged as an "emotionally normal transaction". The transaction data corresponding to the "emotionally normal transaction" are used as the training data set to update the parameters in the expected transaction model. The Map2 process divides the transaction data into "emotionally normal transactions" and "emotionally abnormal transactions".

[0176] Based on the second hook function hook2, the intermediate data obtained from Map2 is divided into 5 stream processing nodes 3-1, 3-2, 3-3, 3-4, and 3-5 in the third link. Among them, the second hook function hook2 is a link relationship constructed according to whether the transaction data is "normal emotion transaction" and the quality grade of white-striped pork. The input data of nodes 3-1, 3-2, 3-3, and 3-4 respectively correspond to the "normal emotion transactions" in nodes 2-1, 2-2, 2-3, and 2-4, and node 3-5 aggregates all the "abnormal emotion transactions".

[0177] Execute the Reduce1 process within nodes 3-1, 3-2, 3-3, and 3-4 - calculate the weighted settlement price and cumulative trading volume of white-striped pork on the same day within the node; execute the Reduce2 process within node 3-5 - sort out the "abnormal emotion transactions" of each slaughter enterprise, calculate the cumulative trading volume and trading amount of the "abnormal emotion transactions". Finally, output the weighted trading price and weighted trading volume of "normal emotion" white-striped pork of each quality grade, the trading information of "abnormal emotion" white-striped pork, and the cumulative trading volume and trading amount.

[0178] The product transaction data processing method based on multi-index linkage analysis provided by the present invention processes the real-time generated transaction data, screens out the transaction data of "abnormal emotion transactions", and at the same time gives the transaction average price and cumulative trading volume corresponding to the "normal emotion transactions", the transaction average price and trading volume corresponding to the "normal emotion transactions" of each dealer, and finally comprehensively gives the transaction average price and total transaction volume corresponding to the "normal emotion transactions" of the target agricultural product. Through this processing method, the total transaction amount of the daily wholesale market service can be analyzed more scientifically and objectively, thereby providing data support for scientifically and reasonably formulating the transaction service fee charging mechanism.

[0179] Refer to Figure 7 , the embodiment of the present invention provides an agricultural product transaction data processing device based on multi-index linkage analysis, which adopts the agricultural product transaction data processing method based on multi-index linkage analysis as described in any one of the above, including:

[0180] A data acquisition unit 201 for acquiring multiple transaction data of the target agricultural product transaction, where the transaction data includes the transaction period, exogenous index value, trading volume, and transaction price of the target agricultural product;

[0181] A price determination unit 202 for processing the transaction data based on a pre-constructed price expectation model and giving the expected transaction price corresponding to the transaction data;

[0182] A model matching unit 203 for matching the corresponding data processing model by analyzing the expected transaction price;

[0183] A data processing unit 204, configured to combine the expected transaction price and complete the processing of transaction data based on a data processing model.

[0184] Through the method and device for processing agricultural product transaction data based on multi-index linkage analysis provided by the present invention, the affected irrational transactions can be stripped, and then data analysis can be performed to obtain data results such as the average transaction price and transaction volume of agricultural products with different quality specifications. On the one hand, irrational transactions can be screened out from a large amount of transaction data, and then the reasons for irrationality can be found, and some irrational transactions can be managed or avoided; on the other hand, the value of transaction data can be maximized, more accurate and objective market conditions information can be provided for market managers, the market transaction volume can be calculated more accurately, reasonable management charging standards can be formulated, and the market is encouraged to improve the service level and service quality, thereby enhancing the market competitiveness.

[0185] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the described modules can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0186] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A method for processing agricultural product transaction data based on multi-index linkage analysis, characterized in that: The specific steps include: Acquire multiple transaction data of target agricultural products, wherein the transaction data includes exogenous indicator values, transaction volumes and transaction prices of the target agricultural products; Process the transaction data based on the pre-built price expectation model and give the expected transaction price of the corresponding transaction data; By analyzing the expected transaction price, the corresponding data processing model is matched; Combined with the expected transaction price and based on the data processing model, the transaction data is processed.

2. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 1 is characterized in that: The transaction data is processed based on the pre-built price expectation model to give the expected transaction price of the corresponding transaction price, which specifically includes the following steps: Divide multiple exogenous indicators of target agricultural products into indicators and determine continuous indicators and categorical indicators; Based on the classification index value, the transaction data is classified and the classification level of the transaction data is calibrated; Through the first hook function, combined with the classification level of the transaction data, the price expectation model of each classification level is matched, and the expected transaction price of the corresponding transaction data is given, wherein the price expectation model of each classification level is iteratively trained on the initial price expectation model through the transaction data of each classification level; The construction of the price expectation model for each classification level specifically includes: Obtain the initial price expectation model and transaction data for each classification level; Processing the transaction data of each classification level based on the initial price expectation model to form the expected transaction price corresponding to the transaction data of each classification level; Combine the transaction data of each classification level and the expected transaction price corresponding to the transaction data of each classification level, iterate the initial price expectation model for each classification level respectively, and form the initial price expectation model for each classification level; Continue to iterate the initial price expectation model of each classification level through the previous transaction data and expected transaction price of each classification level, repeat the iterative step, and form the price expectation model of each classification level.

3. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 2 is characterized in that: The construction of the initial price expectation model includes the following steps: Acquire transaction sample data, group the transaction sample data, and form a transaction sample data module; Perform factor analysis on each continuous indicator value of each transaction sample data module to screen and form multiple sentiment factors of target agricultural product transactions; Based on the classification indicators of target agricultural products and the correlation analysis of multiple sentiment factors, the initial price expectation model of each transaction sample data module is constructed; Based on the mean processing of the initial price expectation model of each transaction sample data module, the initial price expectation model of the target agricultural product is given.

4. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 3 is characterized in that: Factor analysis is performed on each continuous indicator value of each transaction sample data module to screen and form multiple sentiment factors of target agricultural product transactions, which specifically includes the following steps: Through each continuous indicator value of each transaction sample data module, the continuous indicators are screened to obtain multiple common factors to form a common factor set; Based on the preset parameter matrix and common factor set, an emotional factor model is constructed, wherein the emotional factor model is specifically expressed as: X p =μ+A·F q +e Among them, X p is a continuous index set of p continuous indicators, μ is the preset parameter matrix, A is the load matrix, F q is a common factor set of q common factors, and e is a q-dimensional random variable; Through the emotional factor model, load analysis is performed on each common factor to obtain the explanatory power of each common factor; Rank each common factor based on its explanatory power; Using the strategy of cumulative explanatory power reaching the threshold and the minimum number of common factors, the target common factor is determined as the sentiment factor of the target agricultural product transaction.

5. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 3 is characterized in that: Based on the classification indicators of target agricultural products and the correlation analysis of multiple sentiment factors, the initial price expectation model of each transaction sample data module is constructed, including: For each transaction sample data module, the relationship model between the transaction price of the target agricultural product and the classification index and sentiment factor is constructed respectively, wherein the parameters in the relationship model are obtained through parameter estimation; Determine the parameters of the initial price expectation model according to the parameters in the relational model corresponding to each transaction sample data module, and construct the initial price expectation model for each transaction sample data module; The initial price expectation model for each transaction sample data module is specifically expressed as: Among them, g(u) is the transformation function of the expected transaction price of the target agricultural product, u is the expected transaction price of the target agricultural product, FL s is the classification index of the sth transaction sample data module, f() is the parameter function, K i () is the i-th non-parametric function, CF si is the i-th sentiment factor of the s-th transaction sample data module, and m is the number of sentiment factors in the s-th transaction sample data module.

6. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 3 is characterized in that: Based on the mean processing of the initial price expectation model of each transaction sample data module, the initial price expectation model of the target agricultural product is given, which specifically includes: The parameters of the initial price expectation model of each transaction sample data module are estimated to obtain the parameter set β s (s=1,2...n), where each parameter contains d parameter estimates; Given a set of parameter estimates [(β 11 , β 12 ...β 1d )(β 21 , β 22 ...β 2d )...(β n1 , β n2 ...β nd )], and the parameter estimation value of each parameter is averaged to obtain the estimated average value set of the parameter The parameter estimation value of each parameter in the initial price expectation model of each transaction sample data module is replaced by a set of estimated average values ​​of the parameters to give an initial price expectation model for the target agricultural product.

7. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 1 is characterized in that: The data processing model includes a first data processing model and a second data processing model; By analyzing the expected transaction price, the corresponding data processing model is matched, which specifically includes the following steps: Based on the comparative analysis of the expected transaction price and the transaction price, the expected transaction price is determined as a first-category expected transaction price and a second-category expected transaction price; Through the second hook function, the first data processing model and the second data processing model are used to match the first and second expected transaction prices respectively.

8. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 7 is characterized in that: The first data processing model includes a plurality of classification level first data processing models; Through the second hook function, the first data processing model and the second data processing model are used to match the first type of expected transaction price and the second type of expected transaction price respectively, specifically including: Through the second hook function, the transaction data classification level corresponding to a type of expected transaction price and the first data processing model of each classification level are judged, and the matching of the first data processing model of each classification level with a type of expected transaction price is completed; Through the second hook function, the second type of expected transaction price and the second data processing model are matched.

9. The agricultural product transaction data processing method based on multi-index linkage analysis as claimed in claim 8, characterized in that: Combined with the expected transaction price and based on the data processing model, the transaction data is processed, including: Based on the first data processing model, weighted average processing is performed on the transaction data corresponding to a type of expected transaction price to obtain a weighted average price of the transaction data; Based on the second data processing model, the second type of expected transaction prices are analyzed to obtain the second type of cumulative transaction amounts and the second type of cumulative transaction volumes corresponding to the transaction data.

10. A device for processing agricultural product transaction data based on multi-index linkage analysis, characterized in that: The method for processing agricultural product transaction data based on multi-index linkage analysis as claimed in any one of claims 1 to 9 comprises: The data acquisition unit is used to acquire multiple transaction data of target agricultural product transactions, wherein: Transaction data include exogenous index values, transaction volumes and transaction prices of target agricultural products; A price determination unit, used to process the transaction data based on a pre-built price expectation model and provide an expected transaction price for the corresponding transaction data; A model matching unit, used to match the corresponding data processing model by analyzing the expected transaction price; The data processing unit is used to complete the processing of transaction data in combination with the expected transaction price and based on the data processing model.

Images