CN113869990A - Shop lease pricing method based on space-time representation learning model - Google Patents

Shop lease pricing method based on space-time representation learning model Download PDF

Info

Publication number
CN113869990A
CN113869990A CN202111455937.4A CN202111455937A CN113869990A CN 113869990 A CN113869990 A CN 113869990A CN 202111455937 A CN202111455937 A CN 202111455937A CN 113869990 A CN113869990 A CN 113869990A
Authority
CN
China
Prior art keywords
shop
pricing
shops
model
encoder
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111455937.4A
Other languages
Chinese (zh)
Inventor
高文飞
王辉
王瑞雪
王磊
张玉欣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Rongling Technology Group Co ltd
Original Assignee
Shandong Rongling Technology Group Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shandong Rongling Technology Group Co ltd filed Critical Shandong Rongling Technology Group Co ltd
Priority to CN202111455937.4A priority Critical patent/CN113869990A/en
Publication of CN113869990A publication Critical patent/CN113869990A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0645—Rental transactions; Leasing transactions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0283—Price estimation or determination
    • G06Q30/0284—Time or distance, e.g. usage of parking meters or taximeters

Abstract

The invention belongs to the technical field of urban data planning and mining, and relates to a shop lease pricing method based on a space-time representation learning model. The method comprises the following steps of data set preparation, data set preprocessing, region shop feature acquisition, GRU model training, encoder and decoder training, test set testing and shop pricing result calculation. The method utilizes the GCN method, the GRU and the seq2seq model to learn the potential change rule of the shop pricing, can intelligently adjust the price, does not need personnel intervention, helps enterprises to make reasonable shop pricing, and improves the operation efficiency of the enterprises.

Description

Shop lease pricing method based on space-time representation learning model
Technical Field
The invention belongs to the technical field of urban data planning and mining, and relates to a shop lease pricing method based on a space-time representation learning model.
Background
The shop leasing is an important component in the economic system of the real estate enterprise, and has important significance for enabling the enterprise to obtain higher profits and properly pricing the shops. Traditionally, the property estimates the pricing of the shops roughly according to the section, area and pricing of the shops around the property. Early traditional prediction methods are costly and inefficient, consuming both human time and energy, and limiting the enterprise to obtain higher profits. If the estimated shop pricing is higher, the business inviting effect is poor, and if the estimated pricing is lower, the enterprise cannot obtain high profit, and the return period is prolonged.
Shop pricing is affected by a variety of factors, such as location, area, traffic, grade, building design, etc., which makes evaluating shop pricing solely by human judgment a very difficult and laborious task. The spatio-temporal feature learning model is used as an important method for deep learning, and is applied to various fields such as image generation, traffic prediction, anomaly detection and the like. The spatio-temporal representation learning model is applied to the field of shop pricing, and the prediction of the shop pricing by combining multiple factors is a very worthy research direction.
Disclosure of Invention
The invention provides a novel shop leasing pricing method based on a space-time representation learning model, aiming at the problems of difficulty and energy consumption in the traditional shop pricing.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
a shop lease pricing method based on a space-time representation learning model comprises the following steps:
s1: the preparation of the data set is carried out,
collecting historical pricing data sets of all shops in the previous T time periods, and positions and areas of all shops in the T time periods;
s2: the pre-processing of the data set is performed,
carrying out normalization processing on a historical shop pricing data set, and dividing the data set into a training set and a testing set;
s3: the characteristics of the regional shop are obtained,
acquiring the features of n surrounding areas of the shops, extracting the geographic position features and pricing similarity features of the shops in the surrounding areas, and acquiring a final embedded feature vector;
s4: the GRU model is trained to be,
and embedding the embedded feature vector obtained in the step S3 into a Gated current Unit (GRU), and obtaining a potential representation feature h of the shop.
S5-training the encoder and decoder,
and inputting the historical shop unit price data sets XP of each region in the previous T time period into an encoder by using a seq2seq framework to obtain potential spatial mapping, inputting the final output of the encoder into a decoder for training, predicting pricing in the T +1 time period of shops, calculating a loss function, and training to obtain an optimized encoder and decoder.
S6: the test set is tested by a test set testing system,
the test set is tested using the trained model.
S7: the results of the shop pricing are calculated,
and obtaining the average unit price per square meter of the shop through the seq2seq model, and multiplying the average unit price by the area of the shop to obtain the final pricing of the shop.
Preferably, in the data set preparation step, the areas where shops are located are gridded, corresponding to 400 areas D (D)1,d2,……d400) And collecting historical pricing data sets X and attribute data sets S of various shops in the previous T time periods of the various shops in each area, wherein the attribute data sets of the shops comprise positions and areas of the shops. Wherein the unit of T is month and the unit of area is square meter.
Preferably, in the step of preprocessing the data set, pricing of each shop in the data set X in the previous T time periods is divided by the area corresponding to the shop in the data set S to obtain a historical unit price data set of each shop in the previous T time periods per 1 square meter, unit prices of each shop belonging to the same region i (1, 2, …, 400) are averaged, abnormally priced shops are screened out according to the average value, a historical unit price data set XP of each region in the previous T time periods is finally obtained, and finally the data set X is divided according to a 7:3 ratio to obtain a training set and a test set.
Preferably, in the step of obtaining the regional shop features, inspiration is obtained from a Graph Convolutional neural Network (GCN), and the distance between the regions is calculated by using a euclidean distance formula in consideration of the geographic position, so as to obtain a shop geographic neighborhood matrix. And (4) calculating the difference value of the shop unit price of each area by using an Euclidean distance method in consideration of unit price similarity to obtain a shop unit price similarity neighborhood matrix. The geographic neighborhood and the unit price similarity neighborhood matrix of the shop are normalized, and the regional shop d is obtained by training a pre-weighting aggregator functioniGeographic feature vector of
Figure 302851DEST_PATH_IMAGE001
And monovalent similarity feature vector
Figure 827635DEST_PATH_IMAGE002
Combining the two to finally obtain the embedded feature vector
Figure 538102DEST_PATH_IMAGE003
。
Preferably, in the step of training the GRU model, the GRU model is used, the embedded feature vector and the historical hidden state of the shop are used as input, the reset gate and the update gate information are calculated, and the potential representation h of the shop area is obtained by learning the spatio-temporal features of the shop areat。
Figure 141122DEST_PATH_IMAGE004
，
Preferably, in the step of training the encoder and the decoder, the framework of the present invention is implemented by using a seq2seq architecture, wherein the encoder and the decoder both use historical shop unit price data set XP of each region in the previous T period as input based on the GRU model, output predicted shop unit price data through the encoder and the decoder, calculate loss values of the predicted value and the real unit price, and propagate backward to calculate a gradient, the gradient is transmitted from a loss function to the decoder and then from the decoder to the encoder, and update model parameters of the encoder and the decoder by using a gradient descent method to obtain the trained encoder and decoder.
Preferably, in the step of predicting the test set, the test set is tested using the model trained in S5, an error between the predicted shop pricing and the true pricing is calculated, and an evaluation index RMSE is calculated.
Preferably, in the step of calculating the pricing result of the shops, the final average unit price of the area of the shops is obtained through the steps of S1-S6, and the final pricing of each shop is the product of the average unit price per square meter of the area and the area of the shop.
The invention is based on the seq2seq model, wherein the encoder and the decoder are based on the GCN method and the GRU model, and the invention considers the influences of shop history pricing, shop area, surrounding shop area pricing and the like. Firstly, based on GCN, carrying out grid division on the region where the shops are located. In the spatial dimension, the invention considers the influence of the area around the shop on the shop, analyzes from two angles of geographic distance and pricing similarity to obtain the final embedded feature vector of the prediction area, and uses the GRU model to obtain the potential representation of the shop by inputting the embedded feature vector and the historical hidden state. In the time dimension, a seq2seq model is utilized, and the average unit price of the regions is trained and predicted by inputting a historical shop unit price data set XP and a commodity attribute data set S of each region in the previous T time period.
Compared with the prior art, the invention has the advantages and positive effects that:
the method utilizes the GCN method and the GRU and seq2seq models to learn the potential change rule of the shop pricing, can intelligently adjust the price, does not need personnel intervention, helps enterprises to make reasonable shop pricing, and improves the operation efficiency of the enterprises.
Drawings
FIG. 1 is a full flow chart of the present invention.
Detailed Description
In order that the above objects, features and advantages of the present invention may be more clearly understood, the present invention will be further described with reference to specific embodiments. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described herein, and thus the present invention is not limited to the specific embodiments of the present disclosure.
Example 1
The embodiment provides a shop lease pricing method based on a spatio-temporal representation learning model.
S1: the preparation of the data set is carried out,
collecting pricing and attribute data sets of shops, determining the city district of the shop, gridding the district into 20X 20 areas D, and collecting historical pricing data sets X (X) of the shops in all the areas of the district in the former T time periods1，X2，…，X400) And a shop attribute data set S (S)1，S2…，S400)。XiAnd SiRepresenting historical pricing and store attribute data sets for the first T time periods of the ith region, wherein
Figure 50172DEST_PATH_IMAGE005
M represents an area diThe number of shops in the system, 3, respectively corresponds to the three attributes of the positions, areas and the levels of the belonged business circles of the shops.
In the above step, region D refers to the entire region, and the pricing of the shops is analyzed, where all shops of a region are analyzed, and the region is denoted by D. D is then meshed into 20 x 20 small regions, each denoted by D. R is a representation illustrating the shape of the data set.
S2: the pre-processing of the data set is performed,
data set X of each regioniThe pricing of each store is divided by the data set SiObtaining the historical unit price data set of each shop per 1 square meter in the previous T time period according to the area corresponding to the middle shop and the formula
Figure 854180DEST_PATH_IMAGE006
For region diThe unit price of the shop of (a) is averaged,
Figure 910998DEST_PATH_IMAGE007
and
Figure 458654DEST_PATH_IMAGE008
respectively, the region d in the time period tiAverage shop unit price and area diThe lease unit price of shop m according to the formula
Figure 222210DEST_PATH_IMAGE009
Screening abnormal definite values with larger difference with the average value, calculating unit price average value of the screened data, and finally obtaining historical shop unit price data set XP (XP) of each region in the previous T time period1,XP2，…，XPT) XP of the samet
Figure 587333DEST_PATH_IMAGE010
R400And finally, dividing the data set XP according to the ratio of 7:3 to obtain a training set and a testing set.
In this step, since the data set may exist for some reason, there may be some shops that have high or low pricing and such abnormal pricing needs to be rejected. Then, the pricing of all the rest shops is averaged to obtain the average value of the pricing of the shops
Figure 741233DEST_PATH_IMAGE007
，
Figure 718679DEST_PATH_IMAGE008
Is the region diThe rental unit price of shop m, L is a threshold value, if
Figure 336742DEST_PATH_IMAGE011
And
Figure 748132DEST_PATH_IMAGE007
is less than L, then the difference is retained
Figure 513963DEST_PATH_IMAGE008
Otherwise, reject
Figure 403421DEST_PATH_IMAGE008
。
S3: the characteristics of the regional shop are obtained,
region diEnhancing self characteristics by obtaining shop characteristics of surrounding areas, respectively obtaining geographical positions and pricing similarity characteristics of the surrounding areas based on a graph convolution neural network, calculating geographical distances between the areas by using an Euclidean distance method in consideration of geographical distances to obtain a geographical domain matrix between the areas, normalizing the geographical domain matrix, selectively putting emphasis on the areas with similar geography by using a pre-weighted aggregator function, wherein the expression of the geographical pre-weighted aggregator function is
Figure 141570DEST_PATH_IMAGE012
Wherein the training is obtained
Figure 848495DEST_PATH_IMAGE013
Denotes diGeographic distance feature embedding vector at time t, WdAnd
Figure 976988DEST_PATH_IMAGE010
are the trainable weights and training parameters,
Figure 466875DEST_PATH_IMAGE014
and
Figure 856268DEST_PATH_IMAGE015
respectively represent the regions diAnd djThe method is characterized in that the difference value of the shop unit price of each area is calculated by utilizing an Euclidean distance formula in consideration of unit price similarity, a unit price similar distance matrix between the areas is obtained, and normalization is carried out. diAnd djRespectively, the i-th and j-th regions, dis (d)i,dj) Denotes diAnd djDirect distance, denominator dis (d)i,dj) The summation of (a) represents: region diThe sum of the distances from other regions j, (j belonging to (1, 400), i.e. all other regions than i), is here equivalent to normalization. While selectively placing emphasis on regions of similar unit price using a unit price pre-weighted aggregator function expressed as
Figure 609460DEST_PATH_IMAGE016
Wherein the training is obtained
Figure 287566DEST_PATH_IMAGE017
Denotes giPricing similarity feature embedding vector at time t, WsAnd
Figure 404646DEST_PATH_IMAGE018
are trainable weights and training parameters. Finally, the geographic distance features are embedded into the vectors
Figure 789491DEST_PATH_IMAGE019
And pricing similarity feature embedding vector
Figure 244743DEST_PATH_IMAGE020
By the formula
Figure 738041DEST_PATH_IMAGE021
The fusion is carried out by fusing the components,
Figure 507414DEST_PATH_IMAGE022
for training parameters, the final embedded feature vector is obtained
Figure 402558DEST_PATH_IMAGE023
For the whole large area, a series of embedded feature vectors of time series T are obtained
Figure 28711DEST_PATH_IMAGE024
。
S4: the GRU model is trained to be,
embedding feature vectors of the previous T time period
Figure 619093DEST_PATH_IMAGE025
As input, the GRU model is passed through
Figure 582370DEST_PATH_IMAGE026
Will be provided with
Figure 738544DEST_PATH_IMAGE027
Encoding to hidden state
Figure 4441DEST_PATH_IMAGE028
The GRU captures long-term dependence using the following equation:
Figure 442638DEST_PATH_IMAGE029
，
Figure 881709DEST_PATH_IMAGE030
，
Figure 361232DEST_PATH_IMAGE031
=tanh(
Figure 657084DEST_PATH_IMAGE032
)，
Figure 222058DEST_PATH_IMAGE033
+
Figure 527137DEST_PATH_IMAGE034
，
wherein
Figure 657904DEST_PATH_IMAGE035
Is the Simoid activation function, ztAnd
Figure 24DEST_PATH_IMAGE036
respectively, update gate and reset gate, W, Wz andwr is a trainable weight. h ist-1Is a hidden state at time t-1.
Figure 849031DEST_PATH_IMAGE037
Including the last moment ht-1And currently entered useful information
Figure 692222DEST_PATH_IMAGE038
Data, potential representation of each region obtained by GRU model
Figure 880758DEST_PATH_IMAGE039
。
S5-training the encoder and decoder,
using seq2seq model as the overall architecture of the present invention, the GCN and GRU modules in the above-mentioned steps S3 and S4 are connected in tandem, and the encoder and decoder are based on the GCN and GRU modules, in the encoder, the data set XP and data set S are input into the GCN and GRU models to get the potential representation characteristics of the region, the initial hidden state of the GRU is set to 0, and the last hidden state output is used as the input of the decoder. In the decoder, the input data for the first iteration is set to 0, and the hidden state of the output of the GCN + GRU module is input to the fully-connected layer to predict the shop pricing for time period t +1
Figure 456096DEST_PATH_IMAGE040
Then pricing the forecasted shops
Figure 621760DEST_PATH_IMAGE040
And as input, continuously predicting the shop pricing of the next time period, calculating a loss function in the training process, reversely calculating a gradient, and updating the model parameters of the encoder and the decoder by a gradient descent method. Obtaining regional shop pricing for the future t +1 to t + m time period after a plurality of iterations
Figure 878429DEST_PATH_IMAGE041
，
Figure 311685DEST_PATH_IMAGE042
，……，
Figure 57924DEST_PATH_IMAGE043
}。
S6: the test set is tested by a test set testing system,
training the test set by using a model, and calculating an evaluation index RMSE, wherein the formula of the RMSE is as follows:
RMSE=
Figure 553627DEST_PATH_IMAGE044
，
wherein
Figure 4200DEST_PATH_IMAGE045
Is a region d at time tiReal shop rental unit price;
Figure 964066DEST_PATH_IMAGE046
is a region d at time tiThe predicted unit price of (1). In general, the RMSE value obtained by comparing with other methods is smaller than that obtained by other methods, so that the method is superior to other methods, and the RMSE value obtained by the embodiment is 28.63.
S7: the results of the shop pricing are calculated,
inputting historical shop unit price data sets XP and attribute data sets of shops of each region in the previous T time period, and predicting to obtain average unit price Y of each region in the previous T time period through the stepstRegion diHas a shop average unit price of
Figure 84468DEST_PATH_IMAGE047
By the formula
Figure 457681DEST_PATH_IMAGE048
=
Figure 321732DEST_PATH_IMAGE049
，
Figure 136104DEST_PATH_IMAGE050
Is a region diThe area corresponding to the mth shop. Average unit price
Figure 319086DEST_PATH_IMAGE051
Multiplying the area corresponding to each shop to obtain the pricing of each shop,
Figure 789382DEST_PATH_IMAGE052
m is the region diThe number of shops, and the final pricing of each shop.
Regarding the notation of some nouns in this embodiment:
1. position: in step 1, the invention converts a large area D into a 20 × 20 small area D, where the position is the position coordinate where the shop is located in the area D, and determines which small area D the shop belongs to according to the position.
2. Area: which refers to how many square meters a store occupies.
3. Potential representation characteristics H of the shop: the store pricing is translated into another potential representation method.
4. Final output of the encoder: hidden state (containing information of the entire input sequence XP).
5. Encoder and decoder: the encoder is used to analyze the input sequence and the decoder is used to generate the output sequence. The encoder encodes input sequence information by taking historical shop unit price data set XP of each region in previous T time period as input to obtain hidden state representation (information of the whole input sequence XP is encoded), and then inputs the final output result (hidden state representation) of the encoder into a decoder to obtain shop pricing of each region in future time period from T +1 to T + m
Figure 253861DEST_PATH_IMAGE053
，
Figure 250636DEST_PATH_IMAGE054
，……，
Figure 712841DEST_PATH_IMAGE055
}. Both the encoder and decoder of the present aspect are GCN + GRU structures.
6. Acquiring the shop characteristics of the surrounding area to strengthen the characteristics of the shop: considering the influence of surrounding shops on the shops, the distance between the two shops is close, and the similarity of pricing shows that the two shops are very similar.
7. Graph convolution neural network: i.e., GCN, whose chinese name is the graph convolution neural network.
8. The Euclidean distance method comprises the following steps: a similarity measurement method.
9. Normalization: the normalization method of the invention is mainly used for normalizing the area diAnd region djThe distance and the pricing similarity are scaled to be between 0 and 1.
Seq2seq model: the most basic Seq2Seq model consists of three parts, namely an encoder, a decoder and an intermediate hidden state representation connecting the two, where our model has an encoder and a decoder, and can be said to be based on the Seq2Seq model.
11. GCN and GRU modules: corresponding to the methods of step three and step four, respectively.
12. Gradient descent method: in a common optimization method, when the loss function is minimized, iterative solution can be performed step by a gradient descent method to obtain the minimized loss function and the model parameter value.
The above description is only a preferred embodiment of the present invention, and not intended to limit the present invention in other forms, and any person skilled in the art may apply the above modifications or changes to the equivalent embodiments with equivalent changes, without departing from the technical spirit of the present invention, and any simple modification, equivalent change and change made to the above embodiments according to the technical spirit of the present invention still belong to the protection scope of the technical spirit of the present invention.

Claims (3)

1. A shop lease pricing method based on a space-time representation learning model is characterized by comprising the following steps:
s1: the preparation of the data set is carried out,
collecting historical pricing data sets of all shops in the previous T time periods and the positions and the areas of all shops;
s2: the pre-processing of the data set is performed,
normalizing historical pricing data sets of all shops in the previous T time periods, and dividing the data sets into a training set XP and a testing set;
s3: the characteristics of the regional shop are obtained,
acquiring the features of n surrounding areas of the shops, extracting the geographic position features and pricing similarity features of the shops in the surrounding areas, and acquiring a final embedded feature vector;
s4: the GRU model is trained to be,
embedding the embedded feature vector obtained in the step S3 into a gating circulation unit to obtain potential representation features of the shop;
s5: the encoder and the decoder are trained and,
inputting the training set XP in the step S2 into an encoder by using a seq2seq architecture to obtain potential spatial mapping, inputting the final output result of the encoder into a decoder for training, predicting pricing in the T +1 time period of a shop, calculating a loss function, and training to obtain an optimized encoder and decoder;
s6: the test set is tested by a test set testing system,
training and testing the test set XP in the step S2 by using the GRU model trained in the step S4 and the encoder and the decoder trained in the step S5, and calculating an evaluation index RMSE to finally obtain a qualified seq2seq model;
s7: the results of the shop pricing are calculated,
and obtaining the average unit price per square meter of the shop through a seq2seq model, and multiplying the average unit price by the area of the shop to obtain the final pricing of the shop.
2. The shop lease pricing method based on the spatio-temporal representation learning model as claimed in claim 1, wherein in step S3, the region where each shop is located is characterized by obtaining the shops in the surrounding region to enhance their own features, and based on the graph convolution neural network, the geographic location and pricing similarity features of the surrounding region are respectively obtained, and from the consideration of the geographic location, the geographic location between the regions is calculated by using the euclidean distance method to obtain the geographic domain matrix between the regions, the geographic domain matrix is normalized, and the pre-weighted aggregator function is used to select to put emphasis on the regions with close geographic proximity.
3. The shop lease pricing method based on spatio-temporal representation learning model of claim 1, wherein said step S4 specifically operates as follows:
embedding feature vectors of the previous T time period
Figure 787114DEST_PATH_IMAGE001
As input, the GRU model is passed through
Figure 736616DEST_PATH_IMAGE002
Will be provided with
Figure 123997DEST_PATH_IMAGE001
Encoding to hidden state
Figure 184357DEST_PATH_IMAGE003
The GRU model captures long-term dependence using the following formula:
Figure 144222DEST_PATH_IMAGE004
，
Figure 654838DEST_PATH_IMAGE005
，
Figure 637838DEST_PATH_IMAGE006
=tanh(
Figure 298626DEST_PATH_IMAGE007
)，
Figure 706474DEST_PATH_IMAGE008
+
Figure 732199DEST_PATH_IMAGE009
，
wherein
Figure 264811DEST_PATH_IMAGE010
Is a function of the activation of the sialoid,
Figure 57187DEST_PATH_IMAGE011
and
Figure 929328DEST_PATH_IMAGE012
update and reset gates, W, Wz and Wr, respectively, are trainable weights; h ist-1Is a hidden state at the moment of t-1;
Figure 188271DEST_PATH_IMAGE013
including the last moment ht-1And currently entered useful information
Figure 771961DEST_PATH_IMAGE014
Data, potential representation of each region obtained by GRU model
Figure 977814DEST_PATH_IMAGE015
。
CN202111455937.4A 2021-12-02 2021-12-02 Shop lease pricing method based on space-time representation learning model Pending CN113869990A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111455937.4A CN113869990A (en) 2021-12-02 2021-12-02 Shop lease pricing method based on space-time representation learning model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111455937.4A CN113869990A (en) 2021-12-02 2021-12-02 Shop lease pricing method based on space-time representation learning model

Publications (1)

Publication Number Publication Date
CN113869990A true CN113869990A (en) 2021-12-31

Family

ID=78985459

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111455937.4A Pending CN113869990A (en) 2021-12-02 2021-12-02 Shop lease pricing method based on space-time representation learning model

Country Status (1)

Country Link
CN (1) CN113869990A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114238428A (en) * 2022-02-24 2022-03-25 亿景智联(北京)科技有限公司 Urban dynamic geographic portrait generation method and device

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106846671A (en) * 2017-01-23 2017-06-13 唐劲松 Shop-within-a-shop's sale management system
US20180336368A1 (en) * 2017-05-22 2018-11-22 International Business Machines Corporation Anonymity assessment system
CN109658162A (en) * 2019-01-10 2019-04-19 广东工业大学 A kind of agricultural product price prediction technique based on seq2seq model and CNN model
CN110619543A (en) * 2019-08-23 2019-12-27 深圳市新系区块链技术有限公司 House lease price prediction method and related device
CN112232483A (en) * 2020-10-15 2021-01-15 昆明理工大学 Flight average fare prediction method combining CNN and LSTM
CN112288488A (en) * 2020-11-02 2021-01-29 上海字符飞舞科技有限公司 Short tenant intelligent pricing system and method
CN113034242A (en) * 2021-04-23 2021-06-25 中国建设银行股份有限公司 Rental assistance method, device, equipment and storage medium
CN113034252A (en) * 2021-04-23 2021-06-25 杭州邻汇网络科技有限公司 Offline field lease auction pricing system and method
CN113204719A (en) * 2021-04-30 2021-08-03 武汉大学 Urban house rent assessment method based on position information superposition and deep neural network

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106846671A (en) * 2017-01-23 2017-06-13 唐劲松 Shop-within-a-shop's sale management system
US20180336368A1 (en) * 2017-05-22 2018-11-22 International Business Machines Corporation Anonymity assessment system
CN109658162A (en) * 2019-01-10 2019-04-19 广东工业大学 A kind of agricultural product price prediction technique based on seq2seq model and CNN model
CN110619543A (en) * 2019-08-23 2019-12-27 深圳市新系区块链技术有限公司 House lease price prediction method and related device
CN112232483A (en) * 2020-10-15 2021-01-15 昆明理工大学 Flight average fare prediction method combining CNN and LSTM
CN112288488A (en) * 2020-11-02 2021-01-29 上海字符飞舞科技有限公司 Short tenant intelligent pricing system and method
CN113034242A (en) * 2021-04-23 2021-06-25 中国建设银行股份有限公司 Rental assistance method, device, equipment and storage medium
CN113034252A (en) * 2021-04-23 2021-06-25 杭州邻汇网络科技有限公司 Offline field lease auction pricing system and method
CN113204719A (en) * 2021-04-30 2021-08-03 武汉大学 Urban house rent assessment method based on position information superposition and deep neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
孙德山等: "基于主成分分析和递归神经网络的短期股票指数预测", 《辽宁师范大学学报(自然科学版)》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114238428A (en) * 2022-02-24 2022-03-25 亿景智联(北京)科技有限公司 Urban dynamic geographic portrait generation method and device
CN114238428B (en) * 2022-02-24 2022-05-13 亿景智联(北京)科技有限公司 Urban dynamic geographic portrait generation method and device

Similar Documents

Publication Publication Date Title
CN109919358A (en) A kind of real-time site traffic prediction technique based on neural network space-time attention mechanism
CN113537600B (en) Medium-long-term precipitation prediction modeling method for whole-process coupling machine learning
CN109657156A (en) A kind of personalized recommendation method generating confrontation network based on circulation
CN113283588B (en) Near-shore single-point wave height forecasting method based on deep learning
CN110599234A (en) Product sales prediction method
Hong et al. Locational marginal price forecasting using deep learning network optimized by mapping-based genetic algorithm
CN111160622A (en) Scenic spot passenger flow prediction method and device based on hybrid neural network model
CN113255900A (en) Impulse load prediction method considering improved spectral clustering and Bi-LSTM neural network
Chen et al. House price prediction based on machine learning and deep learning methods
CN106970986A (en) Urban waterlogging influence degree method for digging and system based on deep learning
Ye et al. Evaluating performance of different generative adversarial networks for large-scale building power demand prediction
Lv et al. An improved long short-term memory neural network for stock forecast
Cheng et al. Analysis and forecasting of the day-to-day travel demand variations for large-scale transportation networks: a deep learning approach
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
CN113869990A (en) Shop lease pricing method based on space-time representation learning model
CN111292121A (en) Garden load prediction method and system based on garden image
CN117217779A (en) Training method and device of prediction model and information prediction method and device
Wu et al. Hybrid genetic-based support vector regression with feng shui theory for appraising real estate price
Sa’at et al. Enhancing the accuracy of malaysian house price forecasting: a comparative analysis on the forecasting performance between the hedonic price model and artificial neural network model
Fang Forecast of foreclosure property market trends during the epidemic based on GA-BP neural network
Zhang Sales Prediction Based on ARIMA Time Series and Multifactorial Linear Model
CN115063972A (en) Traffic speed prediction method and system based on graph convolution and gate control cyclic unit
Baltagi et al. Modelling housing using multi-dimensional panel data
Islam et al. Addressing structural instability in housing market segmentation of the used houses of Tokyo, Japan
Borhani et al. Soft computing modelling of urban evolution: Tehran metropolis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20211231