WO2019214142A1 - 电子装置、基于研报数据的预测方法、程序和计算机存储介质 - Google Patents

电子装置、基于研报数据的预测方法、程序和计算机存储介质 Download PDF

Info

Publication number
WO2019214142A1
WO2019214142A1 PCT/CN2018/107508 CN2018107508W WO2019214142A1 WO 2019214142 A1 WO2019214142 A1 WO 2019214142A1 CN 2018107508 W CN2018107508 W CN 2018107508W WO 2019214142 A1 WO2019214142 A1 WO 2019214142A1
Authority
WO
WIPO (PCT)
Prior art keywords
research report
data
prediction
hidden layer
time interval
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.)
Ceased
Application number
PCT/CN2018/107508
Other languages
English (en)
French (fr)
Inventor
李正洋
李海疆
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.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen 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 Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2019214142A1 publication Critical patent/WO2019214142A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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

Definitions

  • the present application relates to the field of computer technologies, and in particular, to an electronic device, a prediction method based on research report data, a program, and a computer storage medium.
  • the research report is a research report made by the researcher of the brokerage firm on the value of securities and related products, or the factors affecting their market prices.
  • the main purpose of the present application is to provide an electronic device, a prediction method based on research report data, a program, and a computer storage medium, aiming at solving the problem that the current prediction method based on the research report data is inefficient and difficult to make real-time prediction.
  • an electronic device includes a memory and a processor, and the memory stores a prediction program based on the research report data, and the prediction program based on the research report data is described.
  • the processor implements the following steps when it executes:
  • Obtaining step obtaining multiple research report data of the prediction object in the first preset time interval
  • the analyzing step analyzing and processing the obtained plurality of research report data according to a predetermined analysis rule to obtain the research report prediction data of each research report;
  • the prediction step input all the obtained prediction data to the pre-established prediction model, and use the result data output by the prediction model as the predicted value of the prediction object.
  • the present application proposes a prediction method based on research report data, the method comprising the steps of:
  • Obtaining step obtaining multiple research report data of the prediction object in the first preset time interval
  • the analyzing step analyzing and processing the obtained plurality of research report data according to a predetermined analysis rule to obtain the research report prediction data of each research report;
  • the prediction step input all the obtained prediction data to the pre-established prediction model, and use the result data output by the prediction model as the predicted value of the prediction object.
  • the present application provides a prediction program based on research report data, and the prediction program based on the research report data includes:
  • An obtaining module configured to acquire multiple research report data of the predicted object in the first preset time interval
  • An analysis module configured to analyze and process the obtained plurality of research report data according to a predetermined analysis rule to obtain the research report prediction data of each research report;
  • a prediction module configured to input all the obtained prediction data to the pre-established prediction model, and use the result data output by the prediction model as the predicted value of the prediction object.
  • the present application provides a computer readable storage medium storing a prediction program based on research data, the prediction program based on the research data may be at least one processor Executing to cause the at least one processor to perform the following steps:
  • Obtaining step obtaining multiple research report data of the prediction object in the first preset time interval
  • the analyzing step analyzing and processing the obtained plurality of research report data according to a predetermined analysis rule to obtain the research report prediction data of each research report;
  • the prediction step input all the obtained prediction data to the pre-established prediction model, and use the result data output by the prediction model as the predicted value of the prediction object.
  • the application obtains multiple research report data of the prediction object in the first preset time interval; analyzes and processes the plurality of research report data to obtain the research report prediction data of each research report; and inputs the obtained research report prediction data
  • the pre-established prediction model is used, and the result data outputted by the prediction model is used as a prediction value of the prediction object.
  • the prediction model of the present application is trained based on multiple historical research reports data, and can not only obtain more objective and accurate prediction results based on the research report data in real time, but also has the advantages of high efficiency and labor saving. .
  • FIG. 1 is a schematic diagram of an operating environment of a first embodiment of a prediction program based on research report data according to the present application;
  • FIG. 2 is a program module diagram of a first embodiment of a prediction program based on research report data according to the present application
  • FIG. 3 is a schematic diagram of a refinement program module of an analysis module of the present application.
  • FIG. 4 is a schematic flow chart of a second embodiment of a method for predicting research report data according to the present application
  • FIG. 5 is a schematic diagram of a refinement process of a predetermined analysis rule of the present application.
  • the present application proposes a prediction program based on the research report data.
  • FIG. 1 is a schematic diagram of the operating environment of the first embodiment of the prediction program 10 based on the research report data.
  • the prediction program 10 based on the report data is installed and operated in the electronic device 1.
  • the electronic device 1 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a server.
  • the electronic device 1 may include, but is not limited to, a memory 11, a processor 12, and a display 13.
  • Figure 1 shows only the electronic device 1 with components 11-13, but it should be understood that not all illustrated components may be implemented, and more or fewer components may be implemented instead.
  • the memory 11 may be an internal storage unit of the electronic device 1, such as a hard disk or memory of the electronic device 1, in some embodiments.
  • the memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in hard disk equipped on the electronic device 1, a smart memory card (SMC), and a secure digital (SD). Card, flash card, etc.
  • the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device.
  • the memory 11 is used to store application software and various types of data installed in the electronic device 1, for example, program code of the prediction program 10 based on the report data.
  • the memory 11 can also be used to temporarily store data that has been output or is about to be output.
  • the processor 12 in some embodiments, may be a Central Processing Unit (CPU), a microprocessor or other data processing chip for running program code or processing data stored in the memory 11, such as performing a research report. Data prediction program 10 and so on.
  • CPU Central Processing Unit
  • microprocessor or other data processing chip for running program code or processing data stored in the memory 11, such as performing a research report.
  • Data prediction program 10 and so on.
  • the display 13 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch sensor, or the like in some embodiments.
  • the display 13 is for displaying information processed in the electronic device 1 and a user interface for displaying visualization.
  • the components 11-13 of the electronic device 1 communicate with each other through a program bus.
  • FIG. 2 is a program module diagram of the first embodiment of the prediction program 10 based on the research report data.
  • the prediction program 10 based on the report data can be divided into one or more modules, one or more modules are stored in the memory 11, and processed by one or more processors (this embodiment is processed The device 12) is executed to complete the application.
  • the prediction program 10 based on the research report data may be divided into an acquisition module 101, an analysis module 102, and a prediction module 103.
  • the module referred to in the present application refers to a series of computer program instruction segments capable of performing a specific function, and is more suitable than the program to describe the execution process of the prediction program 10 based on the research report data in the electronic device 1, wherein:
  • the obtaining module 101 is configured to obtain multiple pieces of research data of the prediction object in the first preset time interval.
  • the above forecast targets include securities trends, industry trends, etc., which usually refer to the forecast targets of the research report.
  • the forecast target is the amount of change in the securities of Automotive Electronics Corporation A.
  • the web crawler framework includes a downloader and a parser.
  • the downloader is responsible for crawling the web page, and the parser is responsible for parsing the web page into the library.
  • the information sources include, but are not limited to, websites (including financial websites, news websites, etc.), Weibo, WeChat, Post Bar, forums and the like.
  • the analyzing module 102 is configured to analyze and process the obtained plurality of research report data according to a predetermined analysis rule to obtain the research report prediction data of each research report.
  • the analysis module 102 includes a first extraction unit 1021, a word segmentation unit 1022, a second extraction unit 1023, and a parsing unit 1024, where:
  • the first extracting unit 1021 is configured to extract the research report text data corresponding to each research report based on the research report data.
  • the first extracting unit 1021 is specifically configured to:
  • the research report text data corresponding to each research report is directly extracted based on the research report data.
  • the research report data is subjected to character recognition processing, and the research report text data corresponding to the research report data is extracted.
  • the research report data for example, research report data in PDF format, etc.
  • character recognition processing of the research report data is required, for example, by OCR (Optical Character Recognition)
  • OCR Optical Character Recognition
  • the word segmentation unit 1022 is configured to perform word segmentation processing on the research report text data corresponding to each research report, and obtain the word segmentation data corresponding to each research report.
  • the word segmentation unit 1022 is specifically configured to:
  • the research report text data corresponding to each research report is processed in a sentence (for example, according to the character code, the punctuation in the research report text data is obtained, and according to the obtained punctuation, the research report text data is subjected to clause processing to segment At least one clause is obtained), and the clause data corresponding to each research report is obtained. Then, the clause data is subjected to word segmentation processing to obtain word segmentation data corresponding to each research report.
  • each word can be labeled according to their position in the word.
  • the commonly used labels are as follows: B, Begin, which means that the word is the first word of a word; M, Middle, indicates this Is the word in the middle of a word; E, End, which means that this is the tail of a word; S, Single, which means that this is a word.
  • the process of word segmentation is to input a character into the sequence labeling model, obtain the corresponding tag sequence, and then segment the word according to the tag sequence.
  • the commonly used Chinese word segmentation includes Jieba, Yaha, Finalseg, etc., mostly based on CRF (Conditional Random Field) algorithm, and some algorithms based on HMM (HiddenMarkov Model). .
  • the second extracting unit 1023 is configured to extract keyword data in the word segmentation data corresponding to each research report.
  • the second extraction unit 1023 is specifically configured to:
  • the keyword library is set in advance.
  • the keyword library may also be replaced by a reference table including keywords.
  • the participle is marked as a keyword.
  • Keyword data in the word segmentation data is obtained.
  • the conclusive words in the research report can be set as keywords according to the needs. For example, set “Buy”, “Overweight”, “Neutral”, “Reduce”, etc. as keywords.
  • the analyzing unit 1024 analyzes the keyword data corresponding to each research report, and outputs the analysis result corresponding to each research report as the research report prediction data corresponding to the research report.
  • a keyword analysis table is set in advance, and the keyword analysis table is queried to obtain an analysis result corresponding to each keyword.
  • the keyword analysis table can be referred to the following example.
  • the above keyword analysis representation example only shows partial keywords and their corresponding analysis results.
  • the keywords are far more than the above listed contents.
  • the keywords may also include “strongly recommended”, “recommended”, “Watching,” “selling,” “selling strongly,” etc., for words with similar meanings, the analytical results can be the same.
  • the analysis results of the above keywords reflect the predicted value of the research report on the forecasted target for the next six months, with positive values rising and negative values falling.
  • the prediction period and the variation range can be set according to specific application scenarios.
  • the prediction module 103 is configured to input all the obtained prediction data into a pre-established prediction model, and use the result data output by the prediction model as a predicted value of the prediction object.
  • the above prediction model is a model based on historical data of the research report prediction data, which combines the prediction capabilities of a plurality of research institutions and can output more accurate prediction values.
  • a prediction conclusion is generated based on the predicted value of the predicted object. For example, when the predicted value is greater than 15%, the predicted conclusion "buy” is generated. When the predicted value is within the interval [5%, 15%], the predicted conclusion “overweight” is generated, when the predicted value is in the interval [-5% When 5%], the prediction result is "neutral”, and when the predicted value is less than -5%, the prediction conclusion "underweight” is generated.
  • the prediction data program based on the research report data further includes a model building module (not shown), and the model building module is configured to:
  • the training model is constructed based on the neural network, and the historical research report prediction data corresponding to each sampling time interval is used as the sample of the sampling time interval, and the real value corresponding to the samples of each sampling time interval is determined, and each sampling is input.
  • the samples of the time interval and their corresponding real values are trained to the training model.
  • the method for determining the true value corresponding to the sample of the sampling time interval includes:
  • the actual variation range of the prediction object is taken as the real value of the sampling time interval sample prediction time (for example, the sixth month of the time point corresponding to the midpoint value in the sampling time interval).
  • the average value of the actual fluctuation range of the prediction object in the sampling time interval sample prediction time interval (for example, within the next six months of the time point corresponding to the midpoint value in the sampling time interval) is taken as the true value.
  • the above training model based on the neural network includes an input layer, a hidden layer and an output layer, and the specific structure thereof can be referred to the following example:
  • the input layer is a layer, including N nodes.
  • N may be set to be the total number of research teams that have made research reports on the industry in which the predicted object is located in the second preset time interval.
  • Each input should be an N-dimensional vector whose value is the historical report prediction data corresponding to the research report published by each research team within a sampling time interval. If a research team does not target the prediction object within the sampling time interval. Or if the industry in which it is published publishes a research report, the node input value corresponding to the predicted object is 0.
  • the above N may also be set to the total number of researchers who have made research reports on the industry in which the predicted object is located within the second preset time interval.
  • the hidden layer includes a first hidden layer, a second hidden layer, and a third hidden layer, wherein:
  • the third hidden layer (also referred to as a Softmax layer) includes N nodes, and the connection manner with the second hidden layer is a full connection, and the values of the nodes in the third hidden layer are normalized and processed by the Softmax function, and the output is performed.
  • the expression of the Softmax function is as follows:
  • S i is the output value of the value of the i-th node of the third hidden layer after normalization by Softmax
  • the output layer is a layer comprising one node.
  • the structure of the above training model based on neural network construction is for reference only, and the structure of the training model can be adjusted according to specific application scenarios to achieve better training results.
  • some hyper parameters can be adjusted.
  • the value of the training round can be set to be greater than 2000 epoch (the number of iterations).
  • learning rate can be set to be greater than 1. .
  • the training method of the model can adopt Stochastic gradient descent (SGD), which has a fast execution speed and can improve training efficiency.
  • SGD Stochastic gradient descent
  • Other model training methods may be used for training according to different application scenarios, which is not limited in this application.
  • the model training is ended, and the generated training model is used as the prediction model.
  • the hyperparameter is adjusted (eg, the number of nodes of the hidden layer, the learning rate, the training wheel) Number, etc., and continue the model training until the error between the output value of the training model and the real value converges and the value of the loss function is lower than the preset threshold, ending the model training.
  • the generated training model is used as the prediction model. Therefore, the structure of the prediction model is the same as that of the training model output after the end model training, and will not be described herein.
  • the embodiment obtains multiple research report data of the prediction object in the first preset time interval; analyzes and processes the plurality of research report data to obtain the research report prediction data of each research report; and the obtained research report prediction data It is input to a pre-established prediction model, and the result data output by the prediction model is used as a prediction value of the prediction object.
  • the prediction model of the embodiment is trained based on multiple historical research reports data, and not only can obtain more objective and accurate prediction results based on the research report data in real time, but also has high efficiency and labor saving. advantage.
  • the present application also proposes a prediction method based on research report data.
  • FIG. 4 is a schematic flowchart of a first embodiment of a method for predicting research report data according to the present application.
  • the method includes:
  • Step S10 Acquire multiple pieces of research report data of the prediction object in the first preset time interval.
  • the above forecast targets include securities trends, industry trends, etc., which usually refer to the forecast targets of the research report.
  • the forecast target is the amount of change in the securities of Automotive Electronics Corporation A.
  • the web crawler framework includes a downloader and a parser.
  • the downloader is responsible for crawling the web page, and the parser is responsible for parsing the web page into the library.
  • the information sources include, but are not limited to, websites (including financial websites, news websites, etc.), Weibo, WeChat, Post Bar, forums and the like.
  • Step S20 Perform analysis processing on the obtained plurality of research report data according to a predetermined analysis rule to obtain the research report prediction data of each research report.
  • the predetermined analysis rule includes:
  • Step S21 extracting research report text data corresponding to each research report based on the research report data.
  • the step S21 specifically includes:
  • the research report text data corresponding to each research report is directly extracted based on the research report data.
  • the research report data is subjected to character recognition processing, and the research report text data corresponding to the research report data is extracted.
  • the research report data for example, research report data in PDF format, etc.
  • character recognition processing of the research report data is required, for example, by OCR (Optical Character Recognition)
  • OCR Optical Character Recognition
  • step S22 the research report text data corresponding to each research report is subjected to word segmentation processing, and the word segmentation data corresponding to each research report is obtained.
  • the step S22 specifically includes:
  • the research report text data corresponding to each research report is processed in a sentence (for example, according to the character code, the punctuation in the research report text data is obtained, and according to the obtained punctuation, the research report text data is subjected to clause processing to segment At least one clause is obtained), and the clause data corresponding to each research report is obtained. Then, the clause data is subjected to word segmentation processing to obtain word segmentation data corresponding to each research report.
  • a word segmentation based on sequence labeling can be used. For example, in a paragraph of text, each word can be labeled according to their position in the word.
  • the commonly used labels are as follows: B, Begin, which means that the word is the first word of a word; M, Middle, indicates this Is the word in the middle of a word; E, End, which means that this is the tail of a word; S, Single, which means that this is a word.
  • the process of word segmentation is to input a character into the sequence labeling model, obtain the corresponding tag sequence, and then segment the word according to the tag sequence.
  • Commonly used Chinese word segmentation includes Jieba, Yaha, Finalseg, etc., mostly based on CRF (Conditional Random Field) algorithm, and some algorithms based on HMM (HiddenMarkov Model). .
  • Step S23 extracting keyword data in the word segmentation data corresponding to each research report.
  • the step S23 specifically includes:
  • the keyword library is set in advance.
  • the keyword library may also be replaced by a reference table including keywords.
  • the participle is marked as a keyword.
  • Keyword data in the word segmentation data is obtained.
  • the conclusive words in the research report can be set as keywords according to the needs. For example, set “Buy”, “Overweight”, “Neutral”, “Reduce”, etc. as keywords.
  • Step S24 parsing the keyword data corresponding to each research report, and outputting the analysis result corresponding to each research report as the research report prediction data corresponding to the research report.
  • a keyword analysis table is set in advance, and the keyword analysis table is queried to obtain an analysis result corresponding to each keyword.
  • the keyword analysis table can be referred to the following example.
  • the above keyword analysis representation example only shows partial keywords and their corresponding analysis results.
  • the keywords are far more than the above listed contents.
  • the keywords may also include “strongly recommended”, “recommended”, “Watching,” “selling,” “selling strongly,” etc., for words with similar meanings, the analytical results can be the same.
  • the analysis results of the above keywords reflect the predicted value of the research report on the predicted range of changes in the next six months, with positive values rising and negative values falling. The prediction period and the variation range can be set according to specific application scenarios.
  • step S30 all the obtained prediction data are input to a pre-established prediction model, and the result data output by the prediction model is used as a prediction value of the prediction object.
  • the above prediction model is a model based on historical data of the research report prediction data, which combines the prediction capabilities of a plurality of research institutions and can output more accurate prediction values.
  • the method may further include:
  • a prediction conclusion is generated based on the predicted value of the predicted object. For example, when the predicted value is greater than 15%, the predicted conclusion "buy” is generated. When the predicted value is within the interval [5%, 15%], the predicted conclusion “overweight” is generated, when the predicted value is in the interval [-5% When 5%], the prediction result is "neutral”, and when the predicted value is less than -5%, the prediction conclusion "underweight” is generated.
  • the method before step S30, or before steps S10 and S20, the method further includes:
  • the training model is constructed based on the neural network, and the historical research report prediction data corresponding to each sampling time interval is used as the sample of the sampling time interval, and the real value corresponding to the samples of each sampling time interval is determined, and each sampling is input.
  • the samples of the time interval and their corresponding real values are trained to the training model.
  • the method for determining the true value corresponding to the sample of the sampling time interval includes:
  • the actual variation range of the prediction object is taken as the real value of the sampling time interval sample prediction time (for example, the sixth month of the time point corresponding to the midpoint value in the sampling time interval).
  • the average value of the actual fluctuation range of the prediction object in the sampling time interval sample prediction time interval (for example, within the next six months of the time point corresponding to the midpoint value in the sampling time interval) is taken as the true value.
  • the above training model based on the neural network includes an input layer, a hidden layer and an output layer, and the specific structure thereof can be referred to the following example:
  • the input layer is a layer, including N nodes.
  • N may be set to be the total number of research teams that have made research reports on the industry in which the predicted object is located in the second preset time interval.
  • Each input should be an N-dimensional vector whose value is the historical report prediction data corresponding to the research report published by each research team within a sampling time interval. If a research team does not target the prediction object within the sampling time interval. Or if the industry in which it is published publishes a research report, the node input value corresponding to the predicted object is 0.
  • the above N may also be set to the total number of researchers who have made research reports on the industry in which the predicted object is located within the second preset time interval.
  • the hidden layer includes a first hidden layer, a second hidden layer, and a third hidden layer, wherein:
  • the third hidden layer (also referred to as a Softmax layer) includes N nodes, and the connection manner with the second hidden layer is a full connection, and the values of the nodes in the third hidden layer are normalized and processed by the Softmax function, and the output is performed.
  • the expression of the Softmax function is as follows:
  • S i is the output value of the value of the i-th node of the third hidden layer after normalization by Softmax
  • the output layer is a layer comprising one node.
  • the structure of the above training model based on neural network construction is for reference only, and the structure of the training model can be adjusted according to specific application scenarios to achieve better training results.
  • some hyper parameters can be adjusted.
  • the value of the training round can be set to be greater than 2000 epoch (the number of iterations).
  • learning rate can be set to be greater than 1. .
  • the training method of the model can adopt Stochastic gradient descent (SGD), which has a fast execution speed and can improve training efficiency.
  • SGD Stochastic gradient descent
  • Other model training methods may be used for training according to different application scenarios, which is not limited in this application.
  • the model training is ended, and the generated training model is used as the prediction model.
  • the hyperparameter is adjusted (eg, the number of nodes of the hidden layer, the learning rate, the training wheel) Number, etc., and continue the model training until the error between the output value of the training model and the real value converges and the value of the loss function is lower than the preset threshold, ending the model training.
  • the generated training model is used as the prediction model. Therefore, the structure of the prediction model is the same as that of the training model output after the end model training, and will not be described herein.
  • the embodiment obtains multiple research report data of the prediction object in the first preset time interval; analyzes and processes the plurality of research report data to obtain the research report prediction data of each research report; and the obtained research report prediction data It is input to a pre-established prediction model, and the result data output by the prediction model is used as a prediction value of the prediction object.
  • the prediction model of the embodiment is trained based on multiple historical research reports data, and not only can obtain more objective and accurate prediction results based on the research report data in real time, but also has high efficiency and labor saving. advantage.
  • the present application further provides a computer readable storage medium storing a prediction program based on a report data, the prediction program based on the report data may be executed by at least one processor, Causing the at least one processor to perform the prediction data based prediction method in any of the above embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Finance (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Game Theory and Decision Science (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Technology Law (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种电子装置、基于研报数据的预测方法、程序和计算机存储介质。该方法包括：获取第一预设时间区间内预测对象的多份研报数据（S10）；根据预先确定的分析规则，对获取的多份研报数据进行分析处理，以获得每份研报的研报预测数据（S20）；将获取的所有研报预测数据输入至预先建立的预测模型，并将预测模型输出的结果数据作为预测对象的预测值（S30）。相较于现有技术，该预测模型基于多份历史研报数据训练而成，不仅能够实时的基于研报数据获得更为客观、准确的预测结果，同时还具备效率高、节省人力等优点。

Description

电子装置、基于研报数据的预测方法、程序和计算机存储介质
优先权申明
本申请基于巴黎公约申明享有2018年05月10日递交的申请号为CN201810441411.2、名称为“电子装置、基于研报数据的预测方法和计算机存储介质”中国专利申请的优先权，该中国专利申请的整体内容以参考的方式结合在本申请中。
技术领域
本申请涉及计算机技术领域，特别涉及一种电子装置、基于研报数据的预测方法、程序和计算机存储介质。
背景技术
研报是券商的研究人员对证券及其相关产品的价值，或者影响其市场价格的因素进行分析，所作出的研究报告。
针对同一预测对象，不同研究团队可能给出不同的结论，若仅参考某一研究团队的结论风险较大，且缺乏准确性与客观性。若要参考该领域多个研究团队的结论，则需要分析海量研报数据，以提炼每份研报的核心信息。目前，研报的分析工作仍依赖人工，其不足之处在于耗时长、效率低、人工成本高，且难以根据研报作出实时的综合预测。
发明内容
本申请的主要目的是提供一种电子装置、基于研报数据的预测方法、程序和计算机存储介质，旨在解决当前基于研报数据的预测方法效率低且难以作出实时预测的问题。
为实现上述目的，本申请提出的一种电子装置，所述电子装置包括存储器和处理器，所述存储器上存储有基于研报数据的预测程序，所述基于研报数据的预测程序被所述处理器执行时实现如下步骤：
获取步骤：获取第一预设时间区间内预测对象的多份研报数据；
分析步骤：根据预先确定的分析规则，对获取的所述多份研报数据进行 分析处理，以获得每份研报的研报预测数据；
预测步骤：将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
此外，为实现上述目的，本申请提出一种基于研报数据的预测方法，该方法包括步骤：
获取步骤：获取第一预设时间区间内预测对象的多份研报数据；
分析步骤：根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据；
预测步骤：将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
此外，为实现上述目的，本申请提出一种基于研报数据的预测程序，所述基于研报数据的预测程序包括：
获取模块，用于获取第一预设时间区间内预测对象的多份研报数据；
分析模块，用于根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据；
预测模块，用于将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
此外，为实现上述目的，本申请提出一种计算机可读存储介质，所述计算机可读存储介质存储有基于研报数据的预测程序，所述基于研报数据的预测程序可被至少一个处理器执行，以使所述至少一个处理器执行如下步骤：
获取步骤：获取第一预设时间区间内预测对象的多份研报数据；
分析步骤：根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据；
预测步骤：将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
本申请获取第一预设时间区间内预测对象的多份研报数据；对该多份研报数据进行分析处理，以获得每份研报的研报预测数据；将获取的研报预测数据输入至预先建立的预测模型，并将预测模型输出的结果数据作为预测对象的预测值。相较于现有技术，本申请预测模型基于多份历史研报数据训练而成，不仅能够实时的基于研报数据获得更为客观、准确的预测结果，同时 还具备效率高、节省人力等优点。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图示出的结构获得其他的附图。
图1为本申请基于研报数据的预测程序第一实施例的运行环境示意图；
图2为本申请基于研报数据的预测程序第一实施例的程序模块图；
图3为本申请分析模块的细化程序模块示意图；
图4为本申请基于研报数据的预测方法第二实施例的流程示意图；
图5为本申请预先确定的分析规则的细化流程示意图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
以下结合附图对本申请的原理和特征进行描述，所举实例只用于解释本申请，并非用于限定本申请的范围。
此外，本申请提出一种基于研报数据的预测程序。
请参阅图1，是本申请基于研报数据的预测程序10第一实施例的运行环境示意图。
在本实施例中，基于研报数据的预测程序10安装并运行于电子装置1中。电子装置1可以是桌上型计算机、笔记本、掌上电脑及服务器等计算设备。该电子装置1可包括，但不仅限于，存储器11、处理器12及显示器13。图1仅示出了具有组件11-13的电子装置1，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。
存储器11在一些实施例中可以是电子装置1的内部存储单元，例如该电 子装置1的硬盘或内存。存储器11在另一些实施例中也可以是电子装置1的外部存储设备，例如电子装置1上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。进一步地，存储器11还可以既包括电子装置1的内部存储单元也包括外部存储设备。存储器11用于存储安装于电子装置1的应用软件及各类数据，例如基于研报数据的预测程序10的程序代码等。存储器11还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU)，微处理器或其他数据处理芯片，用于运行存储器11中存储的程序代码或处理数据，例如执行基于研报数据的预测程序10等。
显示器13在一些实施例中可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。显示器13用于显示在电子装置1中处理的信息以及用于显示可视化的用户界面。电子装置1的部件11-13通过程序总线相互通信。
请参阅图2，是本申请基于研报数据的预测程序10第一实施例的程序模块图。在本实施例中，基于研报数据的预测程序10可以被分割成一个或多个模块，一个或者多个模块被存储于存储器11中，并由一个或多个处理器(本实施例为处理器12)所执行，以完成本申请。例如，在图2中，基于研报数据的预测程序10可以被分割成获取模块101、分析模块102及预测模块103。本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段，比程序更适合于描述基于研报数据的预测程序10在电子装置1中的执行过程，其中：
获取模块101，用于获取第一预设时间区间内预测对象的多份研报数据。
上述预测对象包括证券走势、行业走势等，通常是指研报的预测标的。例如，设置预测对象为汽车电子公司A公司的证券变动幅度。
上述获取预测对象的研报数据的方法包括：
从研报数据库中下载研报数据，例如，通过Wind(万得)、Bloomberg(彭博)客户端下载。或者，接收用户输入的研报数据，例如，用户将研报纸件进行扫描，并上传研报的扫描文件。或者，通过网络爬虫在信息源获取研报数据，其中，上述网络爬虫可对分布于不同服务器上的网页进行多任务抓取， 提高研报数据抓取效率。该网络爬虫框架包括下载器和解析器。下载器负责抓取网页，解析器负责解析网页并入库。所述信息源包括但不限于网站(包括财经网站、新闻网站等)、微博、微信、贴吧、论坛等平台。
分析模块102，用于根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据。
优选地，参照图3，所述分析模块102包括第一提取单元1021、分词单元1022、第二提取单元1023及解析单元1024，其中：
第一提取单元1021，用于基于所述研报数据，提取每份研报对应的研报文本数据。
所述第一提取单元1021具体用于：
判断研报数据的文本数据是否可识别。
若是，则直接基于所述研报数据，提取每份研报对应的研报文本数据。
若否，则将该研报数据进行字符识别处理后，提取所述研报数据对应的研报文本数据。对于部分研报数据(例如，PDF格式的研报数据等)，若无法直接识别或编辑其文本数据，则需要对该研报数据进行字符识别处理，例如，通过OCR(Optical Character Recognition，光学字符识别)程序对研报数据进行字符识别，以获取每份研报对应的研报文本数据。
分词单元1022，用于将每份研报对应的研报文本数据进行分词处理，获得每份研报对应的分词数据。
所述分词单元1022具体用于：
首先，将每份研报对应的研报文本数据进行分句处理(例如，根据字符编码，获取研报文本数据中的标点，根据获取的标点，对研报文本数据进行分句处理，以分割出至少一个分句)，获得每份研报对应的分句数据。然后，将所述分句数据进行分词处理，获得每份研报对应的分词数据。
下面以中文文本为例，说明具体的分词方法：
由于中文文本是以字符为单位的文本，因此，可采用基于序列标注的分词方法。例如，在一段文字中，可将每个字按照他们在词中的位置进行标注，常用的标注有以下四个：B，Begin，表示这个字是一个词的首字；M，Middle，表示这是一个词中间的字；E，End，表示这是一个词的尾字；S，Single，表示这是单字成词。分词的过程就是将一段字符输入序列标注模型，得到相应 的标记序列，再根据标记序列进行分词。常用的中文分词库包括Jieba(结巴)、Yaha(哑哈)、Finalseg等，大多基于CRF(Conditionalrandom field，条件随机场)算法，也有一些基于HMM(HiddenMarkov Model，隐马尔科夫模型)等算法。
第二提取单元1023，用于提取每份研报对应的分词数据中的关键词数据。
所述第二提取单元1023具体用于：
首先，预先设置关键词库。其中，所述关键词库也可由包括关键词的参照表代替。
然后，确定所述分词数据中的分词是否存在于所述关键词库中。当一分词存在于所述关键词库中时，标记所述分词为关键词。
最后，获取所述分词数据中所有关键词数据。可根据需要，将研报中结论性词语设置为关键词。例如，设置“买入”、“增持”、“中性”、“减持”等作为关键词。
解析单元1024，对每份研报对应的关键词数据进行解析，输出每份研报对应的解析结果作为该研报对应的研报预测数据。
例如，预先设置关键词解析表，查询所述关键词解析表，获得各关键词对应的解析结果，关键词解析表可参照如下示例。
关键词 解析结果
买入 15％
增持 10％
中性 0
减持 -5％
上述关键词解析表示例仅示出部分关键词及其对应的解析结果，在实际应用当中，关键词远多于以上所列内容，例如，关键词还可包括“强烈推荐”、“推荐”、“观望”、“卖出”、“强烈卖出”等，对于词义相近的词，其解析结果可以相同。此外，上述关键词的解析结果体现的是研报对预测对象未来6 个月变动幅度的预测值，正值为上涨幅度，负值为下跌幅度。其中，预测周期以及变动幅度可根据具体的应用场景设定。
预测模块103，用于将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
上述预测模型是基于研报预测数据的历史数据训练而成的模型，该预测模型综合了多家研究机构的预测能力，可输出更为精准的预测值。
在一些实施例中，处理器执行所述基于研报数据的预测程序10时，还实现如下步骤：
根据所述预测对象的预测值，生成预测结论。例如，当预测值大于15％时，生成预测结论“买入”，当预测值在区间[5％，15％]内时，生成预测结论“增持”，当预测值在区间[-5％，5％]内时，生成预测结论“中性”，当预测值小于-5％，生成预测结论“减持”。
优选地，本实施例中，上述基于研报数据的预测程序10还包括模型构建模块(图中未示出)，该模型构建模块用于：
首先，采集第二预设时间区间内多个采样时间区间对应的所述预测对象所在行业的历史研报数据，并根据预先确定的分析规则(该分析规则与分析模块102执行的分析规则相同，在此不作赘述。)对采集的历史研报数据进行分析处理，以获得每份历史研报对应的历史研报预测数据。需要注意的是，本实施例中，由于每份研报的发表时间不同，因此在进行采样时，应当采集一个采样时间区间的样本数据，而不是采集一个采样时刻的样本数据。此外，将样本采集的范围扩大到预测对象所在行业，是为了采集更多样本，以提高预测模型的准确度。
然后，基于神经网络构建训练模型，将每一采样时间区间对应的历史研报预测数据作为该采样时间区间的样本，同时确定各所述采样时间区间的样本对应的真实值，输入各所述采样时间区间的样本及其对应的真实值至所述训练模型进行训练。其中，上述采样时间区间的样本对应的真实值的确定方法包括：
将该采样时间区间样本预测时刻(例如，采样时间区间中点值对应的时间点未来第六个月)预测对象的实际变动幅度作为真实值。
将该采样时间区间样本预测时间区间内(例如，采样时间区间中点值对 应的时间点未来六个月内)预测对象的实际变动幅度均值作为真实值。
其中，上述基于神经网络构建的训练模型包括输入层、隐藏层及输出层，其具体的结构可参照如下示例：
所述输入层为一层，包括N个节点，本实施例中，可设置N为第二预设时间区间内，对预测对象所处行业做出过研报的研究团队的总数量。每一条输入应当为一个N维的向量，该向量的值为一采样时间区间内各研究团队发表的研报对应的历史研报预测数据，若一研究团队在该采样时间区间内未针对预测对象或者其所在行业发表研报，则该预测对象对应的节点输入值为0。
在本实施例以外的其他实施例中，上述N也可设置为第二预设时间区间内，对预测对象所处行业做出过研报的研究员的总数量。
所述隐藏层包括第一隐藏层、第二隐藏层及第三隐藏层，其中：
第一隐藏层包括n 1(例如，n 1＝8)个节点，与输入层的连接方式为全连接(Full Connected)，所述第一隐藏层的激活函数为Sigmoid函数。
第二隐藏层包括n 2(例如，n 2＝5)个节点，与第一隐藏层的连接方式为全连接，所述第二隐藏层的激活函数为Sigmoid函数。
第三隐藏层(也可称为Softmax层)包括N个节点，与第二隐藏层的连接方式为全连接，所述第三隐藏层中节点的值经Softmax函数归一化处理后输出，该Softmax函数的表达式如下：
Figure PCTCN2018107508-appb-000001
其中，S i为第三隐藏层第i个节点的值经Softmax归一化处理后的输出值，V i为第三隐藏层第i个节点输入至Softmax函数的值，i∈[1，N]，j＝1，2，3…，N。
所述输出层为一层，包括1个节点。
上述基于神经网络构建的训练模型的结构仅供参考，可根据具体的应用场景调整训练模型的结构，以达到较优的训练结果。
此外，根据具体应用场景，还可对一些超参数进行调整，例如，可设置训练轮数的值大于2000epoch(迭代次数)，在样本较少时，可设置学习率(Learning Rate)的值大于1。
模型的训练方法可采用随机梯度下降法(Stochastic gradient descent，SGD)，该训练方法执行速度快，可提高训练效率。当然，根据不同的应用场景，也 可采用其他的模型训练方法进行训练，本申请对此不作限定。
最后，当所述训练模型的输出值与所述真实值之间的误差收敛且损失函数的值低于预设阈值时，结束模型训练，并将生成的训练模型作为所述预测模型。在一些实施例中，当训练模型的输出值与真实值之间的误差收敛且损失函数的值大于或等于预设阈值时，调整超参数(例如，隐藏层的节点数、学习率、训练轮数等)，并继续进行模型训练直至训练模型的输出值与真实值之间的误差收敛且损失函数的值低于预设阈值，结束模型训练。
需要注意的是，结束模型训练后，将生成的训练模型作为所述预测模型，因此，所述预测模型的结构与结束模型训练后输出的训练模型的结构相同，在此不作赘述。
本实施例获取第一预设时间区间内预测对象的多份研报数据；对该多份研报数据进行分析处理，以获得每份研报的研报预测数据；将获取的研报预测数据输入至预先建立的预测模型，并将预测模型输出的结果数据作为预测对象的预测值。相较于现有技术，本实施例预测模型基于多份历史研报数据训练而成，不仅能够实时的基于研报数据获得更为客观、准确的预测结果，同时还具备效率高、节省人力等优点。
本申请还提出一种基于研报数据的预测方法。
如图4所示，图4为本申请基于研报数据的预测方法第一实施例的流程示意图。
本实施例中，该方法包括：
步骤S10，获取第一预设时间区间内预测对象的多份研报数据。
上述预测对象包括证券走势、行业走势等，通常是指研报的预测标的。例如，设置预测对象为汽车电子公司A公司的证券变动幅度。
上述获取预测对象的研报数据的方法包括：
从研报数据库中下载研报数据，例如，通过Wind(万得)、Bloomberg(彭博)客户端下载。或者，接收用户输入的研报数据，例如，用户将研报纸件进行扫描，并上传研报的扫描文件。或者，通过网络爬虫在信息源获取研报数据，其中，上述网络爬虫可对分布于不同服务器上的网页进行多任务抓取，提高研报数据抓取效率。该网络爬虫框架包括下载器和解析器。下载器负责 抓取网页，解析器负责解析网页并入库。所述信息源包括但不限于网站(包括财经网站、新闻网站等)、微博、微信、贴吧、论坛等平台。
步骤S20，根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据。
优选地，参照图5，所述预先确定的分析规则包括：
步骤S21，基于所述研报数据，提取每份研报对应的研报文本数据。
所述步骤S21具体包括：
判断研报数据的文本数据是否可识别。
若是，则直接基于所述研报数据，提取每份研报对应的研报文本数据。
若否，则将该研报数据进行字符识别处理后，提取所述研报数据对应的研报文本数据。对于部分研报数据(例如，PDF格式的研报数据等)，若无法直接识别或编辑其文本数据，则需要对该研报数据进行字符识别处理，例如，通过OCR(Optical Character Recognition，光学字符识别)程序对研报数据进行字符识别，以获取每份研报对应的研报文本数据。
步骤S22，将每份研报对应的研报文本数据进行分词处理，获得每份研报对应的分词数据。
所述步骤S22具体包括：
首先，将每份研报对应的研报文本数据进行分句处理(例如，根据字符编码，获取研报文本数据中的标点，根据获取的标点，对研报文本数据进行分句处理，以分割出至少一个分句)，获得每份研报对应的分句数据。然后，将所述分句数据进行分词处理，获得每份研报对应的分词数据。
下面以中文文本为例，说明具体的分词方法：
由于中文文本是以字符为单位的文本，因此，可采用基于序列标注的分词方法。例如，在一段文字中，可将每个字按照他们在词中的位置进行标注，常用的标注有以下四个：B，Begin，表示这个字是一个词的首字；M，Middle，表示这是一个词中间的字；E，End，表示这是一个词的尾字；S，Single，表示这是单字成词。分词的过程就是将一段字符输入序列标注模型，得到相应的标记序列，再根据标记序列进行分词。常用的中文分词库包括Jieba(结巴)、Yaha(哑哈)、Finalseg等，大多基于CRF(Conditionalrandom field，条件随机场)算法，也有一些基于HMM(HiddenMarkov Model，隐马尔科夫模型) 等算法。
步骤S23，提取每份研报对应的分词数据中的关键词数据。
所述步骤S23具体包括：
首先，预先设置关键词库。其中，所述关键词库也可由包括关键词的参照表代替。
然后，确定所述分词数据中的分词是否存在于所述关键词库中。当一分词存在于所述关键词库中时，标记所述分词为关键词。
最后，获取所述分词数据中所有关键词数据。可根据需要，将研报中结论性词语设置为关键词。例如，设置“买入”、“增持”、“中性”、“减持”等作为关键词。
步骤S24，对每份研报对应的关键词数据进行解析，输出每份研报对应的解析结果作为该研报对应的研报预测数据。
例如，预先设置关键词解析表，查询所述关键词解析表，获得各关键词对应的解析结果，关键词解析表可参照如下示例。
关键词 解析结果
买入 15％
增持 10％
中性 0
减持 -5％
上述关键词解析表示例仅示出部分关键词及其对应的解析结果，在实际应用当中，关键词远多于以上所列内容，例如，关键词还可包括“强烈推荐”、“推荐”、“观望”、“卖出”、“强烈卖出”等，对于词义相近的词，其解析结果可以相同。此外，上述关键词的解析结果体现的是研报对预测对象未来6个月变动幅度的预测值，正值为上涨幅度，负值为下跌幅度。其中，预测周期以及变动幅度可根据具体的应用场景设定。
步骤S30，将获得的所有研报预测数据输入至预先建立的预测模型，并将 所述预测模型输出的结果数据作为所述预测对象的预测值。
上述预测模型是基于研报预测数据的历史数据训练而成的模型，该预测模型综合了多家研究机构的预测能力，可输出更为精准的预测值。
在一些实施例中，在步骤S30之后，该方法还可包括：
根据所述预测对象的预测值，生成预测结论。例如，当预测值大于15％时，生成预测结论“买入”，当预测值在区间[5％，15％]内时，生成预测结论“增持”，当预测值在区间[-5％，5％]内时，生成预测结论“中性”，当预测值小于-5％，生成预测结论“减持”。
优选地，本实施例中，在步骤S30之前，或者在步骤S10、S20之前，该方法还包括：
首先，采集第二预设时间区间内多个采样时间区间对应的所述预测对象所在行业的历史研报数据，并根据预先确定的分析规则(该分析规则与步骤S20中的分析规则相同，在此不作赘述。)对采集的历史研报数据进行分析处理，以获得每份历史研报对应的历史研报预测数据。需要注意的是，本实施例中，由于每份研报的发表时间不同，因此在进行采样时，应当采集一个采样时间区间的样本数据，而不是采集一个采样时刻的样本数据。此外，将样本采集的范围扩大到预测对象所在行业，是为了采集更多样本，以提高预测模型的准确度。
然后，基于神经网络构建训练模型，将每一采样时间区间对应的历史研报预测数据作为该采样时间区间的样本，同时确定各所述采样时间区间的样本对应的真实值，输入各所述采样时间区间的样本及其对应的真实值至所述训练模型进行训练。其中，上述采样时间区间的样本对应的真实值的确定方法包括：
将该采样时间区间样本预测时刻(例如，采样时间区间中点值对应的时间点未来第六个月)预测对象的实际变动幅度作为真实值。
将该采样时间区间样本预测时间区间内(例如，采样时间区间中点值对应的时间点未来六个月内)预测对象的实际变动幅度均值作为真实值。
其中，上述基于神经网络构建的训练模型包括输入层、隐藏层及输出层，其具体的结构可参照如下示例：
所述输入层为一层，包括N个节点，本实施例中，可设置N为第二预设 时间区间内，对预测对象所处行业做出过研报的研究团队的总数量。每一条输入应当为一个N维的向量，该向量的值为一采样时间区间内各研究团队发表的研报对应的历史研报预测数据，若一研究团队在该采样时间区间内未针对预测对象或者其所在行业发表研报，则该预测对象对应的节点输入值为0。
在本实施例以外的其他实施例中，上述N也可设置为第二预设时间区间内，对预测对象所处行业做出过研报的研究员的总数量。
所述隐藏层包括第一隐藏层、第二隐藏层及第三隐藏层，其中：
第一隐藏层包括n 1(例如，n 1＝8)个节点，与输入层的连接方式为全连接(Full Connected)，所述第一隐藏层的激活函数为Sigmoid函数。
第二隐藏层包括n 2(例如，n 2＝5)个节点，与第一隐藏层的连接方式为全连接，所述第二隐藏层的激活函数为Sigmoid函数。
第三隐藏层(也可称为Softmax层)包括N个节点，与第二隐藏层的连接方式为全连接，所述第三隐藏层中节点的值经Softmax函数归一化处理后输出，该Softmax函数的表达式如下：
Figure PCTCN2018107508-appb-000002
其中，S i为第三隐藏层第i个节点的值经Softmax归一化处理后的输出值，V i为第三隐藏层第i个节点输入至Softmax函数的值，i∈[1，N]，j＝1，2，3…，N。
所述输出层为一层，包括1个节点。
上述基于神经网络构建的训练模型的结构仅供参考，可根据具体的应用场景调整训练模型的结构，以达到较优的训练结果。
此外，根据具体应用场景，还可对一些超参数进行调整，例如，可设置训练轮数的值大于2000epoch(迭代次数)，在样本较少时，可设置学习率(Learning Rate)的值大于1。
模型的训练方法可采用随机梯度下降法(Stochastic gradient descent，SGD)，该训练方法执行速度快，可提高训练效率。当然，根据不同的应用场景，也可采用其他的模型训练方法进行训练，本申请对此不作限定。
最后，当所述训练模型的输出值与所述真实值之间的误差收敛且损失函数的值低于预设阈值时，结束模型训练，并将生成的训练模型作为所述预测模型。在一些实施例中，当训练模型的输出值与真实值之间的误差收敛且损 失函数的值大于或等于预设阈值时，调整超参数(例如，隐藏层的节点数、学习率、训练轮数等)，并继续进行模型训练直至训练模型的输出值与真实值之间的误差收敛且损失函数的值低于预设阈值，结束模型训练。
需要注意的是，结束模型训练后，将生成的训练模型作为所述预测模型，因此，所述预测模型的结构与结束模型训练后输出的训练模型的结构相同，在此不作赘述。
本实施例获取第一预设时间区间内预测对象的多份研报数据；对该多份研报数据进行分析处理，以获得每份研报的研报预测数据；将获取的研报预测数据输入至预先建立的预测模型，并将预测模型输出的结果数据作为预测对象的预测值。相较于现有技术，本实施例预测模型基于多份历史研报数据训练而成，不仅能够实时的基于研报数据获得更为客观、准确的预测结果，同时还具备效率高、节省人力等优点。
进一步地，本申请还提出一种计算机可读存储介质，所述计算机可读存储介质存储有基于研报数据的预测程序，所述基于研报数据的预测程序可被至少一个处理器执行，以使所述至少一个处理器执行上述任一实施例中的基于研报数据的预测方法。
以上所述仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是在本申请的发明构思下，利用本申请说明书及附图内容所作的等效结构变换，或直接/间接运用在其他相关的技术领域均包括在本申请的专利保护范围内。

Claims (20)

  1. 一种电子装置，所述电子装置包括存储器和处理器，其特征在于，所述存储器上存储有基于研报数据的预测程序，所述基于研报数据的预测程序被所述处理器执行时实现如下步骤：
    获取步骤：获取第一预设时间区间内预测对象的多份研报数据；
    分析步骤：根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据；
    预测步骤：将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
  2. 如权利要求1所述的电子装置，其特征在于，所述预先确定的分析规则包括：
    基于所述研报数据，提取每份研报对应的研报文本数据；
    将每份研报对应的研报文本数据进行分词处理，获得每份研报对应的分词数据；
    提取每份研报对应的分词数据中的关键词数据；
    对每份研报对应的关键词数据进行解析，输出每份研报对应的解析结果作为该研报对应的研报预测数据。
  3. 如权利要求1或2所述的电子装置，其特征在于，在所述预测步骤之前，所述处理器执行所述基于研报数据的预测程序，还实现以下步骤：
    采集第二预设时间区间内多个采样时间区间对应的所述预测对象所在行业的历史研报数据，并根据预先确定的分析规则对采集的历史研报数据进行分析处理，以获得每份历史研报对应的历史研报预测数据；
    基于神经网络构建训练模型，将每一采样时间区间对应的历史研报预测数据作为该采样时间区间的样本，同时确定各所述采样时间区间的样本对应的真实值，输入各所述采样时间区间的样本及其对应的真实值至所述训练模型进行训练；
    当所述训练模型的输出值与所述真实值之间的误差收敛且损失函数的值低于预设阈值时，结束模型训练，并将生成的训练模型作为所述预测模型。
  4. 如权利要求3所述的电子装置，其特征在于，所述预测模型包括一层输入层、三层隐藏层及一层输出层，所述三层隐藏层分别为第一隐藏层、第 二隐藏层及第三隐藏层；其中：
    所述输入层包括N个节点；
    所述第一隐藏层包括n 1个节点，以所述输入层的连接方式为全连接，所述第一隐藏层的激活函数为Sigmoid函数；
    所述第二隐藏层包括n 2个节点，与所述第一隐藏层的连接方式为全连接，所述第二隐藏层的激活函数为Sigmoid函数；
    所述第三隐藏层包括N个节点，与所述第二隐藏层的连接方式为全连接，所述第三隐藏层中节点的值经Softmax函数归一化处理后输出；
    所述输出层包括1个节点。
  5. 如权利要求4所述的电子装置，其特征在于，所述N为第二预设时间区间内对所述预测对象所在行业做出研报的研究团队的总数量。
  6. 一种基于研报数据的预测方法，其特征在于，该方法包括步骤：
    获取步骤：获取第一预设时间区间内预测对象的多份研报数据；
    分析步骤：根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据；
    预测步骤：将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
  7. 如权利要求6所述的基于研报数据的预测方法，其特征在于，所述预先确定的分析规则包括：
    基于所述研报数据，提取每份研报对应的研报文本数据；
    将每份研报对应的研报文本数据进行分词处理，获得每份研报对应的分词数据；
    提取每份研报对应的分词数据中的关键词数据；
    对每份研报对应的关键词数据进行解析，输出每份研报对应的解析结果作为该研报对应的研报预测数据。
  8. 如权利要求6或7所述的基于研报数据的预测方法，其特征在于，在所述预测步骤之前，该方法还包括：
    采集第二预设时间区间内多个采样时间区间对应的所述预测对象所在行业的历史研报数据，并根据预先确定的分析规则对采集的历史研报数据进行分析处理，以获得每份历史研报对应的历史研报预测数据；
    基于神经网络构建训练模型，将每一采样时间区间对应的历史研报预测数据作为该采样时间区间的样本，同时确定各所述采样时间区间的样本对应的真实值，输入各所述采样时间区间的样本及其对应的真实值至所述训练模型进行训练；
    当所述训练模型的输出值与所述真实值之间的误差收敛且损失函数的值低于预设阈值时，结束模型训练，并将生成的训练模型作为所述预测模型。
  9. 如权利要求8所述的基于研报数据的预测方法，其特征在于，所述预测模型包括一层输入层、三层隐藏层及一层输出层，所述三层隐藏层分别为第一隐藏层、第二隐藏层及第三隐藏层；其中：
    所述输入层包括N个节点；
    所述第一隐藏层包括n 1个节点，以所述输入层的连接方式为全连接，所述第一隐藏层的激活函数为Sigmoid函数；
    所述第二隐藏层包括n 2个节点，与所述第一隐藏层的连接方式为全连接，所述第二隐藏层的激活函数为Sigmoid函数；
    所述第三隐藏层包括N个节点，与所述第二隐藏层的连接方式为全连接，所述第三隐藏层中节点的值经Softmax函数归一化处理后输出；
    所述输出层包括1个节点。
  10. 如权利要求9所述的基于研报数据的预测方法，其特征在于，所述N为第二预设时间区间内对所述预测对象所在行业做出研报的研究团队的总数量。
  11. 一种基于研报数据的预测程序，其特征在于，所述基于研报数据的预测程序包括：
    获取模块，用于获取第一预设时间区间内预测对象的多份研报数据；
    分析模块，用于根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据；
    预测模块，用于将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
  12. 如权利要求11所述的基于研报数据的预测程序，其特征在于，所述预先确定的分析规则包括：
    基于所述研报数据，提取每份研报对应的研报文本数据；
    将每份研报对应的研报文本数据进行分词处理，获得每份研报对应的分词数据；
    提取每份研报对应的分词数据中的关键词数据；
    对每份研报对应的关键词数据进行解析，输出每份研报对应的解析结果作为该研报对应的研报预测数据。
  13. 如权利要求11或12所述的基于研报数据的预测程序，其特征在于，所述基于研报数据的预测程序还包括模型构建模块，用于：
    采集第二预设时间区间内多个采样时间区间对应的所述预测对象所在行业的历史研报数据，并根据预先确定的分析规则对采集的历史研报数据进行分析处理，以获得每份历史研报对应的历史研报预测数据；
    基于神经网络构建训练模型，将每一采样时间区间对应的历史研报预测数据作为该采样时间区间的样本，同时确定各所述采样时间区间的样本对应的真实值，输入各所述采样时间区间的样本及其对应的真实值至所述训练模型进行训练；
    当所述训练模型的输出值与所述真实值之间的误差收敛且损失函数的值低于预设阈值时，结束模型训练，并将生成的训练模型作为所述预测模型。
  14. 如权利要求13所述的基于研报数据的预测程序，其特征在于，所述预测模型包括一层输入层、三层隐藏层及一层输出层，所述三层隐藏层分别为第一隐藏层、第二隐藏层及第三隐藏层；其中：
    所述输入层包括N个节点；
    所述第一隐藏层包括n 1个节点，以所述输入层的连接方式为全连接，所述第一隐藏层的激活函数为Sigmoid函数；
    所述第二隐藏层包括n 2个节点，与所述第一隐藏层的连接方式为全连接，所述第二隐藏层的激活函数为Sigmoid函数；
    所述第三隐藏层包括N个节点，与所述第二隐藏层的连接方式为全连接，所述第三隐藏层中节点的值经Softmax函数归一化处理后输出；
    所述输出层包括1个节点。
  15. 如权利要求14所述的基于研报数据的预测程序，其特征在于，所述N为第二预设时间区间内对所述预测对象所在行业做出研报的研究团队的总数量。
  16. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质存储有基于研报数据的预测程序，所述基于研报数据的预测程序可被至少一个处理器执行，以使所述至少一个处理器执行如下步骤：
    获取步骤：获取第一预设时间区间内预测对象的多份研报数据；
    分析步骤：根据预先确定的分析规则，对获取的所述多份研报数据进行分析处理，以获得每份研报的研报预测数据；
    预测步骤：将获得的所有研报预测数据输入至预先建立的预测模型，并将所述预测模型输出的结果数据作为所述预测对象的预测值。
  17. 如权利要求16所述的计算机可读存储介质，其特征在于，所述预先确定的分析规则包括：
    基于所述研报数据，提取每份研报对应的研报文本数据；
    将每份研报对应的研报文本数据进行分词处理，获得每份研报对应的分词数据；
    提取每份研报对应的分词数据中的关键词数据；
    对每份研报对应的关键词数据进行解析，输出每份研报对应的解析结果作为该研报对应的研报预测数据。
  18. 如权利要求16或17所述的计算机可读存储介质，其特征在于，在所述预测步骤之前，所述处理器执行所述基于研报数据的预测程序，还实现以下步骤：
    采集第二预设时间区间内多个采样时间区间对应的所述预测对象所在行业的历史研报数据，并根据预先确定的分析规则对采集的历史研报数据进行分析处理，以获得每份历史研报对应的历史研报预测数据；
    基于神经网络构建训练模型，将每一采样时间区间对应的历史研报预测数据作为该采样时间区间的样本，同时确定各所述采样时间区间的样本对应的真实值，输入各所述采样时间区间的样本及其对应的真实值至所述训练模型进行训练；
    当所述训练模型的输出值与所述真实值之间的误差收敛且损失函数的值低于预设阈值时，结束模型训练，并将生成的训练模型作为所述预测模型。
  19. 如权利要求18所述的计算机可读存储介质，其特征在于，所述预测模型包括一层输入层、三层隐藏层及一层输出层，所述三层隐藏层分别为第 一隐藏层、第二隐藏层及第三隐藏层；其中：
    所述输入层包括N个节点；
    所述第一隐藏层包括n 1个节点，以所述输入层的连接方式为全连接，所述第一隐藏层的激活函数为Sigmoid函数；
    所述第二隐藏层包括n 2个节点，与所述第一隐藏层的连接方式为全连接，所述第二隐藏层的激活函数为Sigmoid函数；
    所述第三隐藏层包括N个节点，与所述第二隐藏层的连接方式为全连接，所述第三隐藏层中节点的值经Softmax函数归一化处理后输出；
    所述输出层包括1个节点。
  20. 如权利要求19所述的计算机可读存储介质，其特征在于，所述N为第二预设时间区间内对所述预测对象所在行业做出研报的研究团队的总数量。
PCT/CN2018/107508 2018-05-10 2018-09-26 电子装置、基于研报数据的预测方法、程序和计算机存储介质 Ceased WO2019214142A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810441411.2 2018-05-10
CN201810441411.2A CN108647822A (zh) 2018-05-10 2018-05-10 电子装置、基于研报数据的预测方法和计算机存储介质

Publications (1)

Publication Number Publication Date
WO2019214142A1 true WO2019214142A1 (zh) 2019-11-14

Family

ID=63754088

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/107508 Ceased WO2019214142A1 (zh) 2018-05-10 2018-09-26 电子装置、基于研报数据的预测方法、程序和计算机存储介质

Country Status (2)

Country Link
CN (1) CN108647822A (zh)
WO (1) WO2019214142A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114065849A (zh) * 2021-11-09 2022-02-18 东南数字经济发展研究院 一种检测蒸压釜工艺行为的数据分析方法

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109284504A (zh) * 2018-10-22 2019-01-29 平安科技（深圳）有限公司 利用深度学习模型的证券研报分析方法及装置
CN109472498A (zh) * 2018-11-15 2019-03-15 陶明 一种开放式定量浓缩证券分析报告的跟踪评价及交易系统
CN110110295B (zh) * 2019-04-04 2023-10-20 平安科技（深圳）有限公司 大样本研报信息提取方法、装置、设备及存储介质
CN110889291B (zh) * 2019-11-21 2023-06-27 山东浪潮科学研究院有限公司 一种基于深度学习和区块链的研报评价方法
CN113191886B (zh) * 2021-05-26 2025-08-12 中国工商银行股份有限公司 基于投研报告的投资交易监控方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106530082A (zh) * 2016-10-25 2017-03-22 清华大学 基于多机器学习的股票预测方法及系统
CN107122351A (zh) * 2017-05-02 2017-09-01 灯塔财经信息有限公司 一种应用于股票新闻领域的态度倾向分析方法与系统
CN107239855A (zh) * 2017-05-23 2017-10-10 华中科技大学 一种基于lstm模型的股票预测方法和系统
US20180032870A1 (en) * 2015-10-22 2018-02-01 Tencent Technology (Shenzhen) Company Limited Evaluation method and apparatus based on text analysis, and storage medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180032870A1 (en) * 2015-10-22 2018-02-01 Tencent Technology (Shenzhen) Company Limited Evaluation method and apparatus based on text analysis, and storage medium
CN106530082A (zh) * 2016-10-25 2017-03-22 清华大学 基于多机器学习的股票预测方法及系统
CN107122351A (zh) * 2017-05-02 2017-09-01 灯塔财经信息有限公司 一种应用于股票新闻领域的态度倾向分析方法与系统
CN107239855A (zh) * 2017-05-23 2017-10-10 华中科技大学 一种基于lstm模型的股票预测方法和系统

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114065849A (zh) * 2021-11-09 2022-02-18 东南数字经济发展研究院 一种检测蒸压釜工艺行为的数据分析方法

Also Published As

Publication number Publication date
CN108647822A (zh) 2018-10-12

Similar Documents

Publication Publication Date Title
Kumar et al. The AI-powered evolution of big data
WO2019214142A1 (zh) 电子装置、基于研报数据的预测方法、程序和计算机存储介质
WO2019200806A1 (zh) 文本分类模型的生成装置、方法及计算机可读存储介质
CN104364781B (zh) 用于计算类别比例的系统和方法
WO2019214145A1 (zh) 文本情绪分析方法、装置及存储介质
CN108021651B (zh) 一种网络舆情风险评估方法及装置
US20130268457A1 (en) System and Method for Extracting Aspect-Based Ratings from Product and Service Reviews
WO2020000717A1 (zh) 网页分类方法、装置及计算机可读存储介质
CN113986864A (zh) 日志数据处理方法、装置、电子设备及存储介质
WO2015180622A1 (zh) 一种确定搜索查询词类别属性的方法和装置
US20220350814A1 (en) Intelligent data extraction
WO2023159756A1 (zh) 价格数据的处理方法和装置、电子设备、存储介质
CN108416034B (zh) 基于金融异构大数据的信息采集系统及其控制方法
CN113935314B (zh) 基于异构图网络的摘要抽取方法、装置、终端设备及介质
CN115640439A (zh) 一种网络舆情监控的方法、系统及存储介质
CN114358024A (zh) 日志分析方法、装置、设备、介质和程序产品
WO2018205391A1 (zh) 信息检索准确性评估方法、系统、装置及计算机可读存储介质
US20230401426A1 (en) Prediction method, prediction apparatus and program
Wu et al. Bootstrap variability studies in ROC analysis on large datasets
JP2024003750A (ja) 言語モデルの訓練方法、装置、電子デバイス及び記憶媒体
US10127075B2 (en) Model driven optimization of annotator execution in question answering system
CN119357411B (zh) 一种安全管理知识图谱数据的标注方法
US20170337484A1 (en) Scalable web data extraction
CN114201376A (zh) 基于人工智能的日志解析方法、装置、终端设备及介质
CN119577078A (zh) 问答方法、装置、计算机设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18918230

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18918230

Country of ref document: EP

Kind code of ref document: A1