CN115936757B - Method and device for predicting psychological price distribution of bidders - Google Patents
Method and device for predicting psychological price distribution of bidders Download PDFInfo
- Publication number
- CN115936757B CN115936757B CN202211684743.6A CN202211684743A CN115936757B CN 115936757 B CN115936757 B CN 115936757B CN 202211684743 A CN202211684743 A CN 202211684743A CN 115936757 B CN115936757 B CN 115936757B
- Authority
- CN
- China
- Prior art keywords
- auction
- price
- price distribution
- historical
- distribution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention discloses a method and a device for predicting psychological price distribution of a bidder, wherein the method comprises the following steps: acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information; acquiring second commodity characteristic information of the to-be-bidded product; and analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on the non-parametric Bayesian model to obtain price distribution prediction data. According to the invention, the historical auction data and the second commodity characteristic information can be converted into the probability distribution form by using the non-parametric Bayesian model, so that the problem of observation of psychological prices is solved, and the quantized psychological price distribution situation of the auction participants can be predicted by data analysis, thereby reducing the experience requirements of auction policy formulators in the auction process and enabling auction policies to be more easily quantized.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a method and a device for predicting psychological price distribution of an auction player.
Background
The auction is a spot transaction mode in which an auction house specializing in the auction business accepts the entrust of a cargo owner, presents the cargo to be auctioned to a buyer according to a certain rules and regulations at a prescribed time and place, discloses bid bidding, and finally sells the cargo to the buyer with the highest bid by the auctioneer. In an e-commerce shopping environment, web auctions can bring greater convenience to owners and auction operators.
However, for the owner or auction platform side, since the psychological price of the bidder cannot be directly observed, the owner or auction platform can usually only empirically formulate the auction strategy, on the one hand, the experience requirements for formulating the strategicer are high, and on the other hand, the auction strategy is difficult to quantify. Aiming at the problems that the experience requirement on auction policy makers is high and the auction policy is difficult to quantify in the existing auction process, further research on the technology for predicting the psychological price distribution of the bidders is necessary to be carried out.
Disclosure of Invention
The embodiment of the invention provides a method and a device for predicting psychological price distribution of a bidder, which are used for solving the problems that the experience requirement on an auction strategy maker is high and the auction strategy is difficult to quantify in the existing auction process.
In one aspect, the present invention provides a method for predicting mental price distribution of a bidder, comprising:
acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information;
acquiring second commodity characteristic information of the to-be-bidded product;
and analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on the non-parametric Bayesian model to obtain price distribution prediction data.
Further, the non-parametric bayesian model includes a Polya Tree structure model, expressed as:
PT(Π，A)；
wherein a represents a super parameter and pi represents an interval parameter.
Further, the Polya Tree structural model adopts a binary segmentation form to encode the interval, the segmentation point is determined by an interval parameter pi, and the pi is defined as:
Π＝(B 0 ，B 1 ，B 00 ，B 01 ，...)；
the Tree structure of the Polya Tree structural model is:
the super parameter a determines the random probability assigned to each partition set:
A＝(a 0 ，a 1 ，a 00 ，a 01 ，...)；
introducing a set of sets epsilon=epsilon 1 ...ε m ，ε i =0or1, assigning random conditional probabilities to such sets:
i.P(B ε0 |B ε )＝C ε0 each C is ε0 Random variable from a Beta distribution, C ε0 ～Beta(a ε0 |，a ε1 );
iiP(Bζ 1 、Bε)＝Cε 1 ＝1-C ε0 ;
iii.C ε 0 are independent;
get set B ε The probability of ε pi is:
further, the initial value of the super parameter A is as follows:
a ε ＝τ z R(B ε )；
wherein τ z Is a scalar greater than zero, z refers to the depth of epsilon, R is a base measure, and R can be assumed to be a uniform distribution when V is not estimated a priori.
Further, the historical price distribution information includes failure, success, and non-bid bidder price distribution information, the Polya Tree model includes failure, success, and non-bid bidder price intervals, and
the failed auction price interval is:
the successful bidder price interval is:
the bid-free bidder price interval is:
further, the step of analyzing the first commodity feature information, the historical price distribution information and the second commodity feature information based on the non-parametric bayesian model to obtain price distribution prediction data includes:
analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on the PT-RNN cyclic neural network to obtain price distribution prediction data; the PT-RNN recurrent neural network includes: input layer, output layer, hidden layer, training set label and loss function.
Further, the training process of the PT-RNN recurrent neural network comprises the following steps:
dividing the first commodity characteristic information and the historical auction data into a training set and a testing set;
inputting the training set into a PT-RNN (potential to noise) cyclic neural network to be trained, and training the PT-RNN cyclic neural network to be trained to obtain a PT-RNN cyclic neural network to be tested;
and testing the PT-RNN circulating neural network to be tested by using the test set, and if the test passes, obtaining the PT-RNN circulating neural network.
Further, after the step of analyzing the first commodity feature information, the historical price distribution information and the second commodity feature information based on the non-parametric bayesian model to obtain price distribution prediction data, the method further comprises the steps of:
acquiring the output or the expected sales of the to-be-bid products;
obtaining supply chain cost data of the to-be-auctioned products according to the output or the expected sales volume;
optimal pricing data for the item to be auctioned is derived based on the profit function, supply chain cost data, and price distribution forecast data.
Further, the profit function is related as:
where c is the supply chain cost of the commodity, x is the commodity pricing,representing psychological price compliance parameter +.>Is a distribution of (a).
In order to solve the problems that the experience requirement on auction policy makers is high and the auction policy is difficult to quantify in the existing auction process, the invention aims to provide a method for predicting psychological price distribution of a bidder, which comprises the following steps: acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information; acquiring second commodity characteristic information of the to-be-bidded product; and analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on the non-parametric Bayesian model to obtain price distribution prediction data.
Compared with the prior art, the invention has the following advantages:
because the historical auction process can reflect the distribution interval of the psychological prices of the bidders, the non-parametric Bayesian model can convert the information into a probability distribution form, the observation problem of the psychological prices is solved, and the quantized psychological price distribution situation of the bidders can be predicted through data analysis, so that the experience requirements of auction strategy makers in the auction process are reduced, and the auction strategies are easier to quantify.
In another aspect, the present invention provides a device for predicting mental price distribution of a bidder, comprising:
the first acquisition module is used for acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information;
the second acquisition module is used for acquiring second commodity characteristic information of the to-be-auction articles;
and the analysis module is used for analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on the non-parametric Bayesian model to obtain price distribution prediction data.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method of bidder mental price distribution prediction according to one embodiment of the present invention.
Fig. 2 is a schematic diagram of a poly Tree model according to another embodiment of the present invention.
Fig. 3 is another structural schematic diagram of a poly Tree model according to another embodiment of the present invention.
FIG. 4 is a schematic diagram showing the results of PT-RNN circulating neural network according to still another embodiment of the present invention.
FIG. 5 is a schematic diagram of a forward propagation process of a PT-RNN recurrent neural network according to yet another embodiment of the present invention.
FIG. 6 is a schematic diagram illustrating an operation principle of a PT-RNN circulating neural network according to still another embodiment of the present invention.
Fig. 7 is a schematic view of a bidder psychological price distribution prediction apparatus according to still another embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a method for predicting psychological price distribution of a bidder according to an embodiment of the present invention includes the following steps:
step 101, acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information.
In an embodiment of the present invention, the merchandise characteristics include various parameters of the merchandise listed in the auction website, such as characteristics information of jewelry merchandise including (place of origin, designer/design facility, material type, material grade/material attribute, shape, size, weight … …).
Step 102, obtaining second commodity characteristic information of the to-be-bidded product.
And step 103, analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on a non-parametric Bayesian model to obtain price distribution prediction data.
In order to solve the problems that the experience requirement on auction policy makers in the existing auction process is high and the auction policy is difficult to quantify, the invention aims to provide the method for predicting the psychological price distribution of the auction participants.
Each auction player has a judgment value for the auction item, and forms a psychological price based on the judgment value, i.e., the psychological price is the price that the auction player gives to an auction item in a principal and subordinate manner. The psychological prices of all the auctioneers participating in a certain auction can form a psychological price distribution, but the psychological price cannot be directly observed, so the psychological price distribution prediction method of the auctioneer disclosed in the embodiment can predict the psychological price distribution of the auctioneer according to the relevant characteristics of the products to be auctioned.
As another implementation of the above embodiment, the non-parametric bayesian model includes a Polya Tree structure model, expressed as:
PT(Π，A)；
wherein a represents a super parameter and pi represents an interval parameter.
The psychological price of the auction player cannot be directly observed, but the auction process can reflect the distribution interval of the psychological price, and the information can be converted into a probability distribution form by using the PolyaTree structure, so that the problem of observing the psychological price is solved. In the embodiment of the invention, the psychological price distribution of the bidder is described by a PolyaTree model consisting of an interval parameter pi and a super parameter a, denoted PT (II, a). Polya Tree is a process of generating a random probability measure distribution, which is an extension of the Dirichlet process and the Tailfree process. The Dirichlet process is a process of generating a set of discrete random probability measures with a probability of 1, which in experiments easily produces random results with poor fitness, and the Polya Tree can generate a set of continuous or absolute continuous probability measures.
In the embodiment of the present invention, the Polya Tree structure model is a binary Tree structure, and a binary partition (dyadic partition) mode is adopted to code the interval, as shown in fig. 2, the partition point is determined by a parameter n, and n is defined as:
Π＝(B 0 ，B 1 ，B oo ，B o1 ，...)(1)
the tree structure is as follows:
the hyper-parameter a determines the random probability assigned to each partition set, wherein:
A＝(a 0 ，a 1 ，a 00 ，a 01 ，...)(3)
introducing a set of sets epsilon=epsilon 1 …ε m ，ε i =0or1, the Polya Tree distribution assigns random conditional probabilities to such sets:
i.P(B ε0 |B ε )＝C ε0 each C is ε0 The random variable from one Beta distribution,
C ε0 ～Beta(a ε0 ，a ε1 )
ii.P(B ε1 B ε )＝C ε1 ＝1-C ε0
iii.C ε0 are all independent
Thus set B ε The probability of e pi is:
for example:
P(B 001 |A)＝C 0 C 00 C 001 (5)
the iterative process of the model can be obtained according to the property of Beta distribution posterior distribution:
thus, if F θ (v) Is considered as a Polya Tree distribution, and the posterior is also a Polya Tree distribution, can be iterated by updating the super parameter A. It should be noted that the above process needs to be calculated once in each level interval, e.g. when there is an observation v e B 10 Corresponding a 1 And a10 need to be updated to a respectively 1 +1 and a 10 +1, since observations also satisfy v ε B 1 . The learning effect of the Polya Tree model is determined by the super parameter A, and the influence proportion of new observation gradually becomes smaller along with the increase of observation samples (the larger a is).
With respect to the initial value of the super parameter a. Let R be a priori estimate of V, let R be called the base measure (base measure), and when V is not a priori estimated, R can be assumed to be a uniform distribution. The initial value of a can be obtained by the following equation.
a ε ＝τ z R(B ε ) (7)
Wherein τ z Is a scalar greater than zero and z refers to the depth of epsilon, e.g., when epsilon=001, then theta=3. The characteristics of Polya Tree structural model can be known, τ z The larger the value of V, the more closely the shape of R and the smaller the corrective effect of the new observations on V. Thus, if a certain confidence is provided in the shape of R (e.g., expert-based experience-based estimation is employed for a priori), a larger τ can be set z The value should be as small as possible on the contrary.
An important link of the Polya Tree structure is to construct a binary Tree model, and the Polya Tree structure is constructed on a price sequence by utilizing the discrete characteristics of a discrete price model.
Let pi= (B) 0 ，B 1 ，B 00 ，B 01 ,..) divide b) as follows:
if the poly Tree structure is built with the observation points of the auction bid as the partitioning points, the Tree structure needs to be updated again every time new data is entered, and although the complexity of the Tree can be relatively simple in the initial stage, each iteration needs to be divided into intervals again and assigned values.
In the embodiment of the invention, the historical price distribution information comprises failure bidder price distribution information, success bidder price distribution information and non-bidding bidder price distribution information, the Polya Tree model comprises failure bidder price interval, success bidder price interval and non-bidding bidder price interval, and
the failed auction price interval is:
the successful bidder price interval is:
the bid-free bidder price interval is:
specifically, for any bid observed (no bid also calculates an observation)Three types of auction type players can be obtained with their corresponding +.>Is defined in the following ranges:
the first category is failed auctioneers, i.e.The psychological price is as follows:
the second category is successful auctioneers, i.eThe psychological price is as follows:
the third category is the unexpired auctioneer, which if directly deleted would bias the estimation of the final auctioneer's mental price distribution. Because the portion of data contains at least one piece of information, i.e. a portion of the samples are smaller thanIn the above, the true v value may be present +.>Any point in between.
And from practical consideration, it is obvious that the auction player does not consider the value of the auction item to be 0, or else does not participate in the auction, so that the psychological price interval of the crowd can be defined asThe effect of retaining this part of auction data on the population is to increase the predictive value to +.>Is reduced at the same time by falling +.>This is also one of the important advantages of the model, namely that as much information as possible of the data is retained (even if this data is never directly observed).
So the psychological price of the third class of auction players is:
in combination with the above formulas (4), (9) to (11), for any ofThe probability of (2) can be derived from:
in particular, there is a range for any minimum intervalFrom the above formula:
thus (2)Can be written as
When b n At the time of → infinity
Wherein C is ε Is a random variable obeying Beta distribution and C ε0 Are independent and can be calculated according to the following formulaI.e.
The posterior probability distribution of psychological prices of the first and third classes of bidders, i.e., posterior prediction distribution of how to acquire new data in practical use, is described belowWhere Data represents acquired bid Data.
When meeting the requirementsWhen (i.e. psychological prices of the first and third classes of bidders), according to a non-parametric bayesian estimation formula, the expression of E (v|data) can be obtained as follows:
wherein N is ε Indicating that Data falls within interval B ε The number of samples in the sample. It is particularly noted that in this model all observed samples do not fall within interval B ε0 . If aε=a ε1 +a ε。 ＝τR(B ε ) I.e. the model is reduced to a Dirichlet procedure, the above formula can be written as:
when b n The → infinity time is:
for the initial state of the model, the pi in PT (pi, A) is calculated by the formula (II) 1 ) The formula is built and fixed, the only consideration is the value of A, and A is defined i For parameters A, A updated after the ith round of auction 0 Is in an initial state.
The psychological price posterior probability distribution of the second class of auction players is described below.
For each auction activity, there is and only one psychological price of the auction player (auction successful) in the intervalAnd (3) inner part. If N beakers participate in the auction of a commodity, the psychological price of N-1 beakers is in +.>The posterior probability distribution pattern should therefore exhibit a pronounced light tail distribution, with the light tail being more pronounced the greater the N value. However, the posterior probability of the Polya Tree model has the problem of thick tails, and the probability is accumulated in +.>And the interval section is required to correct the probability measure value of the part. To get->Posterior probability at (a) and another Polya Tree model can be constructed. To distinguish the previous partition pi, define a new partition as pi ' = (B ' ' 0 ，B′ 1 ，B′ 00 ，B′ 01 …, the binary tree structure of the model is shown in fig. 3:
in the view of figure 3 of the drawings,the posterior probability of this interval can be obtained as:
wherein R (B ')' 1 ) Is a basic measure, which can be regarded as a priori probability measure, and is therefore a constant, the posterior probability of the interval being inversely related to the N value only.
In the auction of one bidding by combining (17) and (21), the posterior probability distribution of the psychological price of the auction player is as follows:
from the practical point of viewThe psychological price of the bidder falls into the interval [ b ] L Infinity) does not affect the final result, the tail probability distribution is processed as follows:
P(v∈[b L ,b L+1 )|v≥b)＝1 (23)
as still another implementation manner of the above embodiment, the step of analyzing the first commodity feature information, the historical price distribution information, and the second commodity feature information based on the non-parametric bayesian model to obtain price distribution prediction data includes:
analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on the PT-RNN cyclic neural network to obtain price distribution prediction data; the PT-RNN recurrent neural network includes: input layer, output layer, hidden layer, training set label and loss function.
According to the embodiment of the invention, through analysis of historical auction data, psychological price distribution of potential requesters of the commodity with certain characteristics can be obtained, and then, the relation between commodity characteristics and the psychological price distribution of the potential requesters is learned by using a deep learning technology.
Referring to fig. 4, taking jewelry as an example, a ring ornament is selected as an example. According to the auction information and the artificial labels provided by the auction website, the following commodity characteristics are selected: brand awareness, material, place of origin, quality, appearance pictures, design moral, IP information, and the like. Thereby learning the relationship between the commodity characteristics and the psychological price distribution of the potential demander.
(1) Brand awareness: brand awareness depends on the average daily search times (web search popularity) of the search engine.
(2) A material. The materials are divided into main materials and non-main materials, and the main materials of the data are classified as follows according to the collected auction information: hetian jade, amber, jadeite, agate, crystal, tourmaline, pearl, diamond, garnet, green gold stone and other natural jades; the non-main materials are as follows: gold, silver, other metals or materials.
(3) The place of origin. The production place refers to the design production place of the product, and not the production place.
(4) Quality. The quality corresponds to the quality of two types of materials.
(5) And (5) appearance pictures. Auction site provided auction item picture information.
(6) The design implications. The characteristics mainly refer to the moras expressed by ornaments, and the topics of the moras are divided into: love, family, health (peace), romantic, personal and no special meaning.
(7) IP information. IP (Intellectual Property) information refers to other subject information contained in the design of the jewelry, such as stars, cartoons, movies, etc. The part is mainly judged in an artificial subjective mode, scoring is carried out in a scoring mode of 0-9, and scoring (network searching heat) is carried out according to the IP influence from low to high, wherein 0 score refers to that no additional IP information is contained.
The training process of the PT-RNN circulating neural network comprises the following steps:
dividing the first commodity characteristic information and the historical auction data into a training set and a testing set;
inputting the training set into a PT-RNN (potential to noise) cyclic neural network to be trained, and training the PT-RNN cyclic neural network to be trained to obtain a PT-RNN cyclic neural network to be tested;
and testing the PT-RNN circulating neural network to be tested by using the test set, and if the test passes, obtaining the PT-RNN circulating neural network.
And when the test result shows that the similarity between the predicted psychological price distribution and the actual psychological price distribution exceeds the index threshold, the test result can be regarded as passing.
Specifically, the PT-RNN model is a cyclic neural network model of the time-based back propagation algorithm, and the design of the model builds a Polya Tree binary segmentation sequence architecture on the basis of the cyclic neural network model. The structure and learning process of the PT-RNN model are described below.
(1) Sequence structure. As shown in FIG. 5, the sequence structure employed by PT-RNNs is binary segmentation. At the far left end of the sequence, the RNN first calculates the proportion of consumers with psychological prices greater than b1, and then the information is passed on to the next stage, namely elimination of consumers with psychological prices greater than b2, among consumers with psychological prices greater than b1The proportion of the fee is then and so on. When the sequence reaches bL, b is carried out to simplify the model under the premise of not influencing pricing decision L+1 The interval probability measure is set to 0.
( 2 ) Input and output. Input feature x (t) Comprises two parts, including commodity characteristics and price intervals. Commodity features include, but are not limited to, raw material features, design features, and the like. Input feature x (t) Represented by (f) i ，b( t) )，f i I.e. the feature vector of commodity i, b is the price, t represents the current stage.
Output o (t) Expressed as probability, and therefore (f i ，b (t) )→o (t) It can be understood that in the t stage, the price of the commodity is predicted to be larger than b according to the characteristics of the commodity (t) The conditional probability of o is o （t） . The price interval is divided according to a binary division structure, namely pi= (B) 0 ，B 1 ，B 00 ，B 01 ,...). Finally according to% 13 ) And calculating the probability measure value of each unit price interval.
( 3 ) A hidden layer. RNN conveys information through hidden layer h (t) Indicating a price greater than b (t) This information will be passed on to the next layer, formulated as:
o (t) ＝P(v≥b (t) |v≥b (t-1) (24)
(4) Training set labels. y is (t) The training set label is a judging standard of the quality of the RNN output result, namely the true probability measure distributed among different price intervals, and the result is obtained by calculation of a PT model. As shown in the training label module in fig. 5, the probability measure of the price per unit interval is given by equation (17).
(5) A loss function. The loss function L describes the gap between the label and the model training result. As shown in fig. 6, the prediction distribution of RNN is compared with the real distribution, and the parameters are corrected by reverse propagation through BPTT algorithm, so as to realize deep learning.
The purpose of the present disclosure is to make the RNN predicted distribution coincide with the actual distribution, so the loss function needs to embody the difference or distance between the two distributions, and the learning purpose is achieved by continuously reducing the distance. Therefore, the loss function needs to meet L (P, Q) not less than 0, namely, the loss function has a minimum value, and when the value is 0, the two distributions are completely consistent.
Specifically, the loss function uses the Hellinger distance:
where P represents the predicted psychological price distribution and Q represents the auction price distribution.
As a further implementation of the above embodiment, after the step of analyzing the first commodity feature information, the historical price distribution information, and the second commodity feature information based on the non-parametric bayesian model, price distribution prediction data is obtained, the method further includes:
acquiring the output or the expected sales of the to-be-bid products;
obtaining supply chain cost data of the to-be-auctioned products according to the output or the expected sales volume;
optimal pricing data for the item to be auctioned is derived based on the profit function, supply chain cost data, and price distribution forecast data.
The embodiment predicts the psychological price distribution of the potential demander by utilizing the commodity characteristics and provides basis for commodity pricing.
Further, the profit function is related as:
where c is the supply chain cost of the commodity, x is the commodity pricing,representing psychological price compliance parameter +.>Is divided into (1)And (3) cloth.
Referring to fig. 7, a device for predicting mental price distribution of a bidder according to still another embodiment of the present invention may be used to implement the steps of the above method embodiment, including:
the first acquisition module is connected with the analysis module and used for acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information;
the second acquisition module is connected with the analysis module and is used for acquiring second commodity characteristic information of the to-be-auction articles;
and the analysis module is used for analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on the non-parametric Bayesian model to obtain price distribution prediction data.
In yet another aspect, the present invention provides a computer device comprising a memory and a processor, the memory having stored thereon a computer program, the processor implementing a method as described above when executing the computer program.
In yet another aspect, the present invention provides a computer readable storage medium storing a computer program which, when executed by a processor, performs a method as described above.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (8)
1. A method for predicting mental price distribution of a bidder, comprising:
acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information;
acquiring second commodity characteristic information of the to-be-bidded product;
analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on a non-parametric Bayesian model to obtain price distribution prediction data;
the non-parametric bayesian model includes a Polya Tree structure model, which is expressed as:
PT(Π，A)；
wherein A represents a super parameter, and pi represents an interval parameter;
the Polya Tree structural model adopts a binary segmentation form to encode an interval, segmentation points are determined by the interval parameters pi, and the pi is defined as:
Π＝(B 0 ，B 1 ，B 00 ，B 01 ，...)；
the Tree structure of the Polya Tree structural model is as follows:
the super parameter a determines the random probability assigned to each partition set:
A＝(a 0 ，a 1 ，a 00 ，a 01 ，...)；
introducing a set of sets epsilon=epsilon 1 ...ε m ，ε i =0or1, assigning random conditional probabilities to such sets:
ⅰ.P(B ε0 |B ε )＝C ε0 each C is ε0 Random variable from a Beta distribution, C ε0 ～Beta(a ε0 ，a ε1 )；
ii.P(B ε1 |B ε )＝C ε1 ＝1-C ε0 ；
ⅲ.C ε0 Are all independent;
get set B ε The probability of ε pi is:
2. the method for predicting psychological price distribution of an auction according to claim 1, wherein said initial value of said super parameter a is:
a ε ＝τ z R(B ε )；
wherein τ z Is a scalar greater than zero, z refers to the depth of epsilon, R is a base measure, and R can be assumed to be a uniform distribution when V is not estimated a priori.
3. The method of claim 2, wherein the historical price distribution information comprises failure, success, and unbiased price distribution information, the Polya Tree model comprises failure, success, and unbiased price intervals, and
the price interval of the failed auction player is as follows:
the successful bidder price interval is:
the bid-free bidder price interval is:
4. the method according to claim 1, wherein the step of analyzing the first commodity feature information, the historical price distribution information, and the second commodity feature information based on a non-parametric bayesian model to obtain price distribution prediction data comprises:
analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on a PT-RNN (potential Transformer-RNN network) circulating neural network to obtain price distribution prediction data; the PT-RNN recurrent neural network includes: input layer, output layer, hidden layer, training set label and loss function.
5. The method for predicting psychological price distribution of an auction of claim 4, wherein said training procedure of PT-RNN recurrent neural network comprises:
dividing the first commodity characteristic information and the historical auction data into a training set and a testing set;
inputting the training set into a PT-RNN (potential to noise) cyclic neural network to be trained to train the PT-RNN cyclic neural network to be trained, so as to obtain a PT-RNN cyclic neural network to be tested;
and testing the PT-RNN circulating neural network to be tested by using the test set, and obtaining the PT-RNN circulating neural network if the test passes.
6. The method according to claim 1, further comprising, after the step of analyzing the first commodity feature information, the historical price distribution information, and the second commodity feature information based on a non-parametric bayesian model to obtain price distribution prediction data:
acquiring the output or the expected sales of the to-be-bid item;
obtaining supply chain cost data of the to-be-bid item according to the output or the expected sales volume;
optimal pricing data for the item to be auctioned is derived based on a profit function, the supply chain cost data, and the price distribution forecast data.
7. The method for predicting psychological price distribution of an auction of claim 6, wherein said profit function is related by:
where c is the supply chain cost of the commodity, x is the commodity pricing,representing psychological price compliance parameter +.>Is a distribution of (a).
8. A device for predicting psychological price distribution of a bidder, comprising:
the first acquisition module is used for acquiring historical auction data, wherein the historical auction data comprises first commodity characteristic information and historical price distribution information;
the second acquisition module is used for acquiring second commodity characteristic information of the to-be-auction articles;
the analysis module is used for analyzing the first commodity characteristic information, the historical price distribution information and the second commodity characteristic information based on a non-parametric Bayesian model to obtain price distribution prediction data;
the non-parametric bayesian model includes a Polya Tree structure model, which is expressed as:
PT(Π，A)；
wherein A represents a super parameter, and pi represents an interval parameter;
the Polya Tree structural model adopts a binary segmentation form to encode an interval, segmentation points are determined by the interval parameters pi, and the pi is defined as:
Π＝(B 0 ，B 1 ，B 00 ，B 01 ，...)；
the Tree structure of the Polya Tree structural model is as follows:
the super parameter a determines the random probability assigned to each partition set:
A＝(a 0 ，a 1 ，a 00 ，a 01 ，...)；
introducing a set of sets epsilon=epsilon 1 ...ε m ，ε i =0or1, assigning random conditional probabilities to such sets:
i. P(B ε0 |B ε )＝C ε0 each C is ε0 Random variable from a Beta distribution, C ε0 ～Beta(a ε0 ，a ε1 )；
ii. P(B ε1 |B ε )＝C ε1 ＝1-C ε0 ；
iii.C ε0 Are all independent;
get set B ε The probability of ε pi is:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211684743.6A CN115936757B (en) | 2022-12-27 | 2022-12-27 | Method and device for predicting psychological price distribution of bidders |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211684743.6A CN115936757B (en) | 2022-12-27 | 2022-12-27 | Method and device for predicting psychological price distribution of bidders |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115936757A CN115936757A (en) | 2023-04-07 |
| CN115936757B true CN115936757B (en) | 2023-09-22 |
Family
ID=86650604
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211684743.6A Active CN115936757B (en) | 2022-12-27 | 2022-12-27 | Method and device for predicting psychological price distribution of bidders |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115936757B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116934438A (en) * | 2023-04-14 | 2023-10-24 | 济南明泉数字商务有限公司 | AI auction decision method and system based on chatGPT model and calculation force |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109064212A (en) * | 2018-07-04 | 2018-12-21 | 苏州仙度网络科技有限公司 | Price forecasting of commodity method and device |
| CN111275117A (en) * | 2020-01-20 | 2020-06-12 | 金华航大北斗应用技术有限公司 | Bayes-based data fluctuation prediction method and system |
| CN115423576A (en) * | 2022-08-12 | 2022-12-02 | 深圳大学 | Auction parameter dynamic optimization method for digital auction platform and computing terminal |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009137048A1 (en) * | 2008-05-05 | 2009-11-12 | Pristine Infotech, Inc | Consumer goods price prediction and optimization |
-
2022
- 2022-12-27 CN CN202211684743.6A patent/CN115936757B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109064212A (en) * | 2018-07-04 | 2018-12-21 | 苏州仙度网络科技有限公司 | Price forecasting of commodity method and device |
| CN111275117A (en) * | 2020-01-20 | 2020-06-12 | 金华航大北斗应用技术有限公司 | Bayes-based data fluctuation prediction method and system |
| CN115423576A (en) * | 2022-08-12 | 2022-12-02 | 深圳大学 | Auction parameter dynamic optimization method for digital auction platform and computing terminal |
Non-Patent Citations (1)
| Title |
|---|
| 基于博弈机制的网上双向拍卖报价模型研究;罗英;郑灿钊;;现代管理科学(第07期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115936757A (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108665311B (en) | Electric commercial user time-varying feature similarity calculation recommendation method based on deep neural network | |
| CN102902691A (en) | Recommending method and recommending system | |
| CN111967971B (en) | Bank customer data processing method and device | |
| CN115936757B (en) | Method and device for predicting psychological price distribution of bidders | |
| CN113744017A (en) | E-commerce search recommendation method and device, equipment and storage medium | |
| Temür et al. | Comparison of forecasting performance of ARIMA LSTM and HYBRID models for the sales volume budget of a manufacturing enterprise | |
| Wang et al. | Session-based recommendation with time-aware neural attention network | |
| Jap et al. | Bidanalyzer: A method for estimation and selection of dynamic bidding models | |
| Dass et al. | Maximizing bidder surplus in simultaneous online art auctions via dynamic forecasting | |
| CN116757747A (en) | Click rate prediction method based on behavior sequence and feature importance | |
| Kalaycı et al. | Optimal model description of finance and human factor indices | |
| Yadav et al. | Stock price forecasting and news sentiment analysis model using artificial neural network | |
| US20220164836A1 (en) | Methods and systems an obv buyback program | |
| Masoudi et al. | The effect of web interface features on consumer online purchase intentions | |
| CN115222203A (en) | Risk identification method and device | |
| CN114493724A (en) | Multi-task learning-based bidding keyword effect estimation model construction method | |
| Kao et al. | Inter-purchase time prediction based on deep learning. | |
| CN112329917A (en) | Multi-layer neural network integration system and method for price limit order trend prediction | |
| Jiang et al. | Comparison Analysis of Stock Price Prediction Based on Different Machine Learning Methods | |
| Ryzhov | Approximate Bayesian inference for simulation and optimization | |
| Wu et al. | Online reviews can predict long-term returns of individual stocks | |
| Guo et al. | Contextual Collaborative Filtering Recommendation Model Integrated with Drift Characteristics of User Interest | |
| Yong et al. | Adaptive detection of FOREX repetitive chart patterns | |
| Wang et al. | Survey of e-commerce modeling and optimization strategies | |
| Liu et al. | Intelligent modeling of abnormal demand forecasting for medical consumables in smart city |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |