CN109840626A - A kind of Stock Market Forecasting method based on convolutional neural networks model - Google Patents
A kind of Stock Market Forecasting method based on convolutional neural networks model Download PDFInfo
- Publication number
- CN109840626A CN109840626A CN201910041449.5A CN201910041449A CN109840626A CN 109840626 A CN109840626 A CN 109840626A CN 201910041449 A CN201910041449 A CN 201910041449A CN 109840626 A CN109840626 A CN 109840626A
- Authority
- CN
- China
- Prior art keywords
- corpus
- financial
- stock market
- newsletter archive
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 30
- 238000013277 forecasting method Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000013480 data collection Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 6
- 238000003062 neural network model Methods 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 18
- 239000000284 extract Substances 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000007423 decrease Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000002224 dissection Methods 0.000 claims description 2
- 238000002790 cross-validation Methods 0.000 abstract description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000010977 jade Substances 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
Abstract
A kind of Stock Market Forecasting method based on convolutional neural networks model, comprising: S1, data collection, S2, data processing, S3, training neural network model, S4, the prediction big step of stock market's tendency four.The financial and economic news corpus that the present invention passes through acquisition and analysis listed company, and financial and economic news corpus is labeled according to DJIA historical data, using the finance and economics corpus information after mark as convolutional neural networks mode input value, it to which the processing through convolutional neural networks model exports the predicted value of news mark, and then is the ups and downs situation and stock market's tendency of predictable DJIA by the predicted value that news marks.The present invention can achieve 65.5% predictablity rate through cross validation.
Description
Technical field
The present invention relates to financial prediction, natural language processing and deep learning fields, especially a kind of to be based on convolutional Neural
The Stock Market Forecasting method of network model.
Background technique
The Internet media is acknowledged as " the fourth-largest media " after newspaper, broadcast, TV.Internet news report and
When property and the convenience of access have become the main source that the public obtains information.Wherein, internet financial and economic news is not only
Have the advantages that report timeliness and access convenience, and its Reporting also frequently refer to listed company manage shape
Condition, financial report, strategic decision, the trend of stock prices etc., these information can analyze the general trend of market development and throwing for market researcher
Money person grasps investment determination and provides help information.But internet financial and economic news numerous and complicated, and mainly with non-structured
Does is textual form presented, then how to excavate the knowledge useful to market and investor from these financial and economic news data? this
Which type of connection do a little financial and economic news contents and stock market's tendency have again?
In recent years, artificial intelligence technology develops rapidly.Wherein, neural network has good None-linear approximation ability and right
The comprehensive treatment capability of complex information can establish the model with stock price inner link according to information data.Meanwhile mind
There are the characteristics such as self study, adaptive through network model, many limitations in traditional prediction method and complexity can be overcome
Problem.Prior art majority is to predict stock market's tendency based on conventional machines learning method, and by way of manual construction feature
The relationship of financial and economic news and stock market is excavated, manual feature can consume a large amount of costs of labor, and limit method to a certain extent
Generalization Capability.
Summary of the invention
The object of the present invention is to provide a kind of comprehensive treatment capability is strong, predictablity rate is high based on convolutional neural networks mould
The Stock Market Forecasting method of type.
The present invention solves a kind of technical solution used by prior art problem: stock market based on convolutional neural networks model
Prediction technique, comprising the following steps:
S1, data collection: the financial and economic news corpus of listed company is obtained from financial web site using web crawlers, passes through Yahoo
Finance and economics api interface crawls the historical data of Dow-Jones Industrial Average Index；
The following steps are included:
S101, stock code list is obtained from financial web site using crawler technology；
It is S102, corresponding within a preset period of time from the listed company crawled in financial web site in the stock code list
History news web page data, as financial and economic news corpus；
S103, the history that the Dow-Jones Industrial Average Index in preset time span is obtained by Yahoo's finance and economics api interface
Data save as DJIA historical data；The preset time span is greater than the preset time period；
S2, data processing: title, the text of each listed company are extracted from the financial and economic news corpus that step S102 is crawled
Newsletter archive corpus is saved as, and newsletter archive corpus is labeled using the daily ups and downs situation of DJIA historical data；
The following steps are included:
S201, data cleansing: dissection process is carried out to financial and economic news corpus using analytical tool: first to financial and economic news corpus
Html label is removed, headline, text and date are then therefrom extracted, is saved to locally as newsletter archive corpus；
S202, data mark: the DJIA historical data obtained according to step S103, to upper in the stock code list
Company, city carries out closing price comparison one by one: as the t+1 days DJIA closing price P of listed companyt+1, it is higher than the said firm and closes for the t days
Valence ptWhen, then by the t days newsletter archive corpus labelings of the said firm at+1；Conversely, working as Pt+1Lower than ptWhen, mark the said firm t
It newsletter archive corpus is -1, constitutes labeled data collection；Wherein, t >=1；
S3, training neural network model: building convolutional neural networks model, using the newsletter archive corpus that has marked and
DJIA historical data trains Stock Market Forecast Model；
The following steps are included:
S301, it is based on deep learning frame, builds convolutional neural networks model: first will be in each newsletter archive corpus
Vocabulary be shown as the form of term vector；During convolutional neural networks model buildings, with all words in newsletter archive corpus
For the splicing vector of term vector as input, the predicted value to mark financial and economic news corpus passes through supervised learning process as output
Training pattern parameter；
S302, model tuning: integrating temporally width cutting for the labeled data in step S202 and collect as training set and verifying,
It is adjusted using the weight parameter that accuracy rate carries out model as evaluation goal；The method that the weight parameter is adjusted are as follows: in model training
In the process, the weight parameter of neural network is adjusted and modifies using back-propagation algorithm and gradient decline Policy iteration, fitting is simultaneously
The error of the prediction output valve and true tag that reduce model obtains Stock Market Forecast Model until error is no longer reduced.
S4, prediction stock market's tendency: obtaining the financial and economic news corpus on listed company's same day using web crawlers from financial web site,
It inputs in Stock Market Forecast Model and predicts the ups and downs situation and stock market's tendency of same day DJIA；
The following steps are included:
S401, prediction stock market's tendency, the weight parameter after the adjusting obtained by step S302 are based on labeled data collection,
Re -training Stock Market Forecast Model obtains secondary Stock Market Forecast Model；
S402, the financial and economic news corpus for obtaining listed company's same day from financial web site using web crawlers, utilize step
The newsletter archive corpus on the same day is inputted secondary Stock Market Forecasting mould by the newsletter archive corpus that the same day is obtained after the data cleansing of S201
In type, the predicted value of output news mark can be obtained the ups and downs situation of same day DJIA, predict stock market's tendency.
In step S301, the concrete form for the convolutional neural networks model built are as follows:
Define xi∈Rk, xiIndicate that k corresponding to i-th of word ties up term vector, R in newsletter archive corpuskFor k tie up word to
Quantity space；One includes word sequence xi,xi+1,...,xi+jNewsletter archive corpus be expressed as the splicing of cascade to
Amount:
Wherein,Indicate concatenation operator, Xi:i+jIndicate word sequence xi,xi+1,...,xi+jCascade；Utilize cunning
Dynamic filter window w ∈ RhkComplete convolution algorithm, wherein h is the size of sliding window, i.e., the number of word in window；K is term vector
Dimension, RhkIndicate the real number matrix that h is tieed up multiplied by k；And new feature is constructed with this, i.e., interior using window includes word sequence
xi,xi+1,...,xi+h-1Newsletter archive corpus cascade Xi:i+h-1, construct a new feature term vector ci, building process
It is as follows:
ci=f (wXi:i+h-1+b)
Wherein, b ∈ R indicates bias term, and f is nonlinear activation function Sigmoid function；Series connection to newsletter archive corpus
Form X1:h,X2:h+1,...,Xn-h+1:nEach window size be h word sequence be filtered, obtain a new Feature Words
Sequence vector c=[c1,c2,...cn-h+1], wherein c ∈ Rn-h+1, n is the sum of word included in newsletter archive corpus, n-h+
1 is the number for the feature word sequence that length is h；The maximum value in feature vector is taken using the operation of maximum pondization:
It willAs the individual features value obtained after pondization operation, to obtain most important feature in each feature vector, simultaneously
It solves the problems, such as that news length is inconsistent, using variable window sizes, extracts multiple features；What these characteristic uses linked entirely
Mode is connected to output layer, and output is the predicted value of news mark；I, j, h, k are natural number.
In step S302, the cutting ratio that training set integrates with verifying is 8:2.
In step S102, financial and economic news corpus is crawled from financial web site using the urllib2 in python kit；Step
The analytical tool used in rapid S201 is the BeautifulSoup of python kit.
The financial web site is Reuter's financial web site.
The beneficial effects of the present invention are: the present invention passes through the financial and economic news corpus for acquiring and analyzing listed company, and root
The newsletter archive corpus extracted by financial and economic news corpus is labeled according to DJIA historical data, by the newsletter archive after mark
Corpus is as convolutional neural networks mode input value, thus the prediction of the processing output news mark through convolutional neural networks model
Value, and then be that the ups and downs situation and stock market's tendency of DJIA can be predicted by the predicted value that news marks.The present invention is through cross validation
It can achieve 65.5% predictablity rate.
Detailed description of the invention
Fig. 1 is basic flow chart of the invention.
Specific embodiment
Below in conjunction with the drawings and the specific embodiments and cooperate examples illustrate the present invention:
Fig. 1 is a kind of basic flow chart of the Stock Market Forecasting method based on convolutional neural networks model of the present invention.A kind of base
In the Stock Market Forecasting method of convolutional neural networks model, comprising the following steps:
S1, data collection: obtaining listed company from financial web site (such as Reuter's financial web site) using web crawlers
Financial and economic news corpus crawls Dow-Jones Industrial Average Index (Dow Jones Industrial by Yahoo's finance and economics api interface
Average, DJIA) historical data；
The following steps are included:
S101, obtaining stock code list from financial web site using crawler technology, (AAPL, GOOG etc., the present embodiment is climbed
It takes and amounts to 2014 listed company's stock codes)；
S102, it uses using the urllib2 in python kit from financial web site according to above-mentioned stock code list
Listed company in July, 2011 is crawled from Reuter's financial web site (http://www.reuters.com/finance/) extremely
History news web page data corresponding to 2 months 2017, as financial and economic news corpus；
S103, the road fine jade in 2010 to 2017 is obtained by Yahoo's finance and economics api interface (yahoo finance api)
The historical data of this industrial average index saves as DJIA historical data；
S2, data processing: title, the text of each listed company are extracted from the financial and economic news corpus that step S102 is crawled
Newsletter archive corpus is saved as, and newsletter archive corpus is labeled using the daily ups and downs situation of DJIA historical data；
The following steps are included:
S201, data cleansing: financial and economic news corpus is carried out at parsing using the BeautifulSoup of python kit
Reason: first removing html label to financial and economic news corpus, then therefrom extracts headline, text and date, and save to this
Ground is as newsletter archive corpus；
S202, data mark: the DJIA historical data obtained according to step S103, to the collected stock code list of institute
In listed company carry out closing price comparison one by one: as the t+1 days DJIA closing price P of listed companyt+1, it is higher than the said firm t
Its closing price ptWhen, then by the t days newsletter archive corpus labelings of the said firm at+1；Conversely, working as Pt+1Lower than ptWhen, mark the public affairs
The newsletter archive corpus taken charge of the t days is -1, constitutes labeled data collection；Wherein, t >=1；
S3, training neural network model, build convolutional neural networks model, using the newsletter archive corpus that has marked and
DJIA historical data trains Stock Market Forecast Model；
The following steps are included:
S301, it is based on deep learning frame, builds convolutional neural networks model: first will be in each newsletter archive corpus
Vocabulary be shown as the form of term vector, during convolutional neural networks model buildings, with all words in newsletter archive corpus
For the splicing vector of term vector as input, the predicted value to mark newsletter archive corpus passes through supervised learning process as output
Training pattern parameter；
The concrete form for the convolutional neural networks model built are as follows:
Define xi∈Rk, xiIndicate that k corresponding to i-th of word ties up term vector, R in newsletter archive corpuskFor k tie up word to
Quantity space；One includes word sequence xi,xi+1,...,xi+jNewsletter archive corpus be expressed as the splicing of cascade to
Amount:
Wherein,Indicate concatenation operator, Xi:i+jIndicate word sequence xi,xi+1,...,xi+jCascade；Utilize cunning
Dynamic filter window w ∈ RhkComplete convolution algorithm, wherein h is the size of sliding window, i.e., the number of word in window；K is term vector
Dimension, RhkIndicate the real number matrix that h is tieed up multiplied by k；And new feature is constructed with this, i.e., interior using window includes word sequence
xi,xi+1,...,xi+h-1Newsletter archive corpus cascade Xi:i+h-1, construct a new feature term vector ci, building process
It is as follows:
ci=f (wXi:i+h-1+b)
Wherein, b ∈ R indicates bias term, and f is nonlinear activation function Sigmoid function；Series connection to newsletter archive corpus
Form X1:h,X2:h+1,...,Xn-h+1:nEach window size be h word sequence be filtered, obtain a new Feature Words
Sequence vector c=[c1,c2,...cn-h+1], wherein c ∈ Rn-h+1, n is the sum of word included in newsletter archive corpus, n-h+
1 is the number for the feature word sequence that length is h；The maximum value in feature vector is taken using the operation of maximum pondization:
It willAs the individual features value obtained after pondization operation, to obtain most important feature in each feature vector, together
When solve the problems, such as that news length is inconsistent, using variable window sizes, extract multiple features；These characteristic uses link entirely
Mode be connected to output layer, output is the predicted value of news mark；I, j, h, k are natural number.
S302, model tuning: integrating temporally width cutting for the labeled data in step S202 and collect as training set and verifying,
It is preferred that the cutting ratio that training set and verifying integrate is 8:2, specific data set division result is as shown in table 1:
1 data set division proportion of table
It is adjusted using the weight parameter that accuracy rate carries out model as evaluation goal, method particularly includes: during model training,
The weight parameter of neural network is adjusted and modified using back-propagation algorithm and gradient decline Policy iteration, is fitted and reduces model
Prediction output valve and the error of true tag obtain Stock Market Forecast Model until error is no longer reduced.Weight parameter tune
It is as shown in table 2 to save optimal result:
2 convolutional neural networks parameter setting of table
Predictablity rate of the convolutional neural networks on verifying collection is 65.5%.
S4, prediction stock market's tendency: obtaining the financial and economic news corpus on listed company's same day using web crawlers from financial web site,
It inputs in Stock Market Forecast Model and predicts the ups and downs situation and stock market's tendency of DJIA；
The following steps are included:
S401, prediction stock market's tendency, weight parameter (the i.e. weight shown in table 2 after the adjusting obtained by step S302
Parameter value), it is based on labeled data collection, re -training Stock Market Forecast Model obtains secondary Stock Market Forecast Model；
S402, the financial and economic news corpus for obtaining listed company's same day from financial web site using web crawlers, utilize step again
The data cleansing of rapid S201 first removes html label to financial and economic news corpus, then therefrom extract headline, text and
After date, the newsletter archive corpus on the same day is obtained, the newsletter archive corpus on the same day is inputted in secondary Stock Market Forecast Model, output
The predicted value of news mark, can be obtained the ups and downs situation of same day DJIA, predict stock market's tendency.
The above content is combine specific optimal technical scheme further detailed description of the invention, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to of the invention
Protection scope.
Claims (5)
1. a kind of Stock Market Forecasting method based on convolutional neural networks model, which comprises the following steps:
S1, data collection: the financial and economic news corpus of listed company is obtained from financial web site using web crawlers, passes through Yahoo's finance and economics
Api interface crawls the historical data of Dow-Jones Industrial Average Index；
The following steps are included:
S101, stock code list is obtained from financial web site using crawler technology；
S102, it is gone through from the listed company crawled in the stock code list in financial web site is corresponding within a preset period of time
History news web page data, as financial and economic news corpus；
S103, the historical data that the Dow-Jones Industrial Average Index in preset time span is obtained by Yahoo's finance and economics api interface,
Save as DJIA historical data；The preset time span is greater than the preset time period；
S2, data processing: the title of each listed company is extracted from the financial and economic news corpus that step S102 is crawled, text saves
For newsletter archive corpus, and newsletter archive corpus is labeled using the daily ups and downs situation of DJIA historical data；
The following steps are included:
S201, data cleansing: dissection process is carried out to financial and economic news corpus using analytical tool: first financial and economic news corpus being removed
Then html label therefrom extracts headline, text and date, save to locally as newsletter archive corpus；
S202, data mark: the DJIA historical data obtained according to step S103, it is public to the listing in the stock code list
Department carries out closing price comparison one by one: as the t+1 days DJIA closing price P of listed companyt+1, it is higher than the t days closing price p of the said firmt
When, then by the t days newsletter archive corpus labelings of the said firm at+1；Conversely, working as Pt+1Lower than ptWhen, mark the said firm the t days
Newsletter archive corpus is -1, constitutes labeled data collection；Wherein, t >=1；
S3, training neural network model: convolutional neural networks model is built, is gone through using the newsletter archive corpus and DJIA that have marked
History data train Stock Market Forecast Model；
The following steps are included:
S301, it is based on deep learning frame, builds convolutional neural networks model: first by the word in each newsletter archive corpus
It is expressed as the form of term vector；During convolutional neural networks model buildings, with the word of all words in newsletter archive corpus to
For the splicing vector of amount as input, the predicted value to mark financial and economic news corpus passes through the training of supervised learning process as output
Model parameter；
S302, model tuning: the labeled data in step S202 is integrated into temporally width cutting and is collected as training set and verifying, with standard
True rate is the weight parameter adjusting that evaluation goal carries out model；The method that the weight parameter is adjusted are as follows: in model training process
In, the weight parameter of neural network is adjusted and modified using back-propagation algorithm and gradient decline Policy iteration, is fitted and reduces
The prediction output valve of model and the error of true tag obtain Stock Market Forecast Model until error is no longer reduced.
S4, prediction stock market's tendency: the financial and economic news corpus on listed company's same day, input are obtained from financial web site using web crawlers
The ups and downs situation and stock market's tendency of same day DJIA are predicted in Stock Market Forecast Model；
The following steps are included:
S401, prediction stock market's tendency, the weight parameter after the adjusting obtained by step S302 are based on labeled data collection, again
Training Stock Market Forecast Model, obtains secondary Stock Market Forecast Model；
S402, the financial and economic news corpus for obtaining listed company's same day from financial web site using web crawlers, utilize step S201's
The newsletter archive corpus that the same day is obtained after data cleansing inputs the newsletter archive corpus on the same day in secondary Stock Market Forecast Model,
The ups and downs situation of same day DJIA can be obtained in the predicted value for exporting news mark, predicts stock market's tendency.
2. a kind of Stock Market Forecasting method based on convolutional neural networks model according to claim 1, which is characterized in that step
In rapid S301, the concrete form for the convolutional neural networks model built are as follows:
Define xi∈Rk, xiIndicate that k corresponding to i-th of word ties up term vector, R in newsletter archive corpuskIt is empty that term vector is tieed up for k
Between；One includes word sequence xi,xi+1,...,xi+jNewsletter archive corpus be expressed as the splicing vector of cascade:
Wherein,Indicate concatenation operator, Xi:i+jIndicate word sequence xi,xi+1,...,xi+jCascade；Using sliding
Filter window w ∈ RhkComplete convolution algorithm, wherein h is the size of sliding window, i.e., the number of word in window；K is the dimension of term vector
Degree, RhkIndicate the real number matrix that h is tieed up multiplied by k；And new feature is constructed with this, i.e., interior using window includes word sequence xi,
xi+1,...,xi+h-1Newsletter archive corpus cascade Xi:i+h-1, construct a new feature term vector ci, building process is such as
Under:
ci=f (wXi:i+h-1+b)
Wherein, b ∈ R indicates bias term, and f is nonlinear activation function Sigmoid function；To the cascade of newsletter archive corpus
X1:h,X2:h+1,...,Xn-h+1:nEach window size be h word sequence be filtered, obtain a new feature word sequence
Vector c=[c1,c2,...cn-h+1], wherein c ∈ Rn-h+1, n is the sum of word included in newsletter archive corpus, and n-h+1 is
Length is the number of the feature word sequence of h；The maximum value in feature vector is taken using the operation of maximum pondization:I.e.
It willIt is solved simultaneously as the individual features value obtained after pondization operation to obtain most important feature in each feature vector
The certainly inconsistent problem of news length extracts multiple features using variable window sizes；The side that these characteristic uses link entirely
Formula is connected to output layer, and output is the predicted value of news mark；I, j, h, k are natural number.
3. a kind of Stock Market Forecasting method based on convolutional neural networks model according to claim 1, which is characterized in that step
In rapid S302, the cutting ratio that training set integrates with verifying is 8:2.
4. a kind of Stock Market Forecasting method based on convolutional neural networks model according to claim 1, which is characterized in that step
In rapid S102, financial and economic news corpus is crawled from financial web site using the urllib2 in python kit；It is adopted in step S201
Analytical tool is the BeautifulSoup of python kit.
5. a kind of Stock Market Forecasting method based on convolutional neural networks model according to claim 1, which is characterized in that institute
Stating financial web site is Reuter's financial web site.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910041449.5A CN109840626A (en) | 2019-01-16 | 2019-01-16 | A kind of Stock Market Forecasting method based on convolutional neural networks model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910041449.5A CN109840626A (en) | 2019-01-16 | 2019-01-16 | A kind of Stock Market Forecasting method based on convolutional neural networks model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109840626A true CN109840626A (en) | 2019-06-04 |
Family
ID=66883893
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910041449.5A Pending CN109840626A (en) | 2019-01-16 | 2019-01-16 | A kind of Stock Market Forecasting method based on convolutional neural networks model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109840626A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192144A (en) * | 2020-01-03 | 2020-05-22 | 湖南工商大学 | Financial data prediction method, device, equipment and storage medium |
-
2019
- 2019-01-16 CN CN201910041449.5A patent/CN109840626A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192144A (en) * | 2020-01-03 | 2020-05-22 | 湖南工商大学 | Financial data prediction method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hu et al. | Stock trading rule discovery with an evolutionary trend following model | |
| Liu et al. | Evaluation of restrictive competition in PPP projects using real option approach | |
| Kao et al. | A hybrid approach by integrating wavelet-based feature extraction with MARS and SVR for stock index forecasting | |
| Vermillion | Impacts of irrigation management transfer: A review of the evidence | |
| Nagaoka et al. | The incidence of cross-licensing: A theory and new evidence on the firm and contract level determinants | |
| Lazar et al. | Comprehensive bibliometric mapping of publication trends in the development of Building Sustainability Assessment Systems | |
| Chou et al. | Improving classification accuracy of project dispute resolution using hybrid artificial intelligence and support vector machine models | |
| Liu et al. | Real option application in PPP/PFI project negotiation | |
| CN106384166A (en) | Deep learning stock market prediction method combined with financial news | |
| Yang et al. | Coordinated development path of metropolitan logistics and economy in Belt and Road using DEMATEL–Bayesian analysis | |
| Rachlin et al. | Admiral: A data mining based financial trading system | |
| CN107844901A (en) | A kind of method and its system of enterprise operation automated analysis | |
| Brown et al. | Financial statement adequacy and firms' MD&A disclosures | |
| Hung et al. | How can learning-by-doing improve decisions in stormwater management? A Bayesian-based optimization model for planning urban green infrastructure investments | |
| CN109840626A (en) | A kind of Stock Market Forecasting method based on convolutional neural networks model | |
| Ye et al. | A new environmental governance cost prediction method based on indicator synthesis and different risk coefficients | |
| Kiyan et al. | The operation of urban water treatment plants: A Review of smart dashboard frameworks | |
| Liu et al. | Forecasting short-term water demands with an ensemble deep learning model for a water supply system | |
| CN110852496A (en) | Natural gas load prediction method based on LSTM recurrent neural network | |
| Park et al. | Finite-horizon optimal consumption and investment problem with a preference change | |
| Baybordi et al. | Fuzzy delphi algorithm and fuzzy DEMITEL in Designing and explanation of relational capital in banks | |
| Razavi Toosi et al. | Ranking water transfer projects using fuzzy methods | |
| Geressu et al. | Evaluating the sensitivity of robust water resource interventions to climate change scenarios | |
| Van Hoa et al. | Foreign exchange rate forecasting using autoencoder and LSTM networks | |
| Filar et al. | Sustainability screw: role of relative production and abatement time scales |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190604 |
|
| RJ01 | Rejection of invention patent application after publication |