JPH04277860A - Event numerical correlation display system, event numerical correlation model preparing system and intension decision supporting system - Google Patents

Abstract

PURPOSE:To handle both numerical information and even information and to eliminate necessity for making the thinking process of a specialist into a rule. CONSTITUTION:This system is composed of an event time sequence data base 2, numerical time sequence data base 3, event information retrieving device 4, processor 6 to execute event numerical correlation model preparation or intension decision support due to event numerical display or neural network 70, display device 7, neuro learning processor 8, receiver 9 to receive the event information through a communication line, receiver 10 to receive real time numerical information through the communication line, and even numerical relation table 11 collecting events to affect the numerical information. Thus, correlation between the event information and the numerical information can be visually easily grasped and even when the thinking process of the specialist is not made into the rule, an event numerical correlation model can be prepared to handle both the event information and the numerical information. Further, prediction or the like fetching the event information can be observed at real time.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an event numerical correlation display system, an event numerical correlation model creation system, and a decision support system. More specifically, we will create an event numerical correlation display system that associates and displays event information such as economic events and numerical information such as stock prices, and a correlation model between event information such as economic events and numerical information such as stock prices. The present invention relates to an event numerical correlation model creation system and a decision support system that incorporates real-time event information into the event numerical correlation model and shows the influence on the numerical information to a decision maker.

0002

[Prior Art] As a statistical modeling technique for stock market prices, etc., Box &Jenkins' Time S
eriesAnalysis" is known. This is a model that predicts fluctuations in stock prices, etc. from past time-series data. On the other hand, Japanese Unexamined Patent Publication No. 64-2166 discloses a "data processing system for business analysis" using an expert system. This supports decision making by modeling the thinking process of experts using IF-THEN rules.

0003

[Problem to be solved by the invention] Stocks, bonds, exchange rates, etc. are not only affected by numerical information such as ordinary profits, GNP, trade balance, etc., but also by numerical information such as statements by the US president and announcements of oil price increases. It is known that events are influenced by event information that cannot be expressed by Therefore, in order to analyze stocks, bonds, exchange rates, etc., it is necessary to know not only the influence of numerical information, but also what kind of event information has an influence and how.

However, the conventional statistical modeling techniques described above have a problem in that they cannot handle event information that cannot be expressed numerically.

[0005] Furthermore, in the "data processing system for business analysis" disclosed in the above-mentioned Japanese Patent Application Laid-Open No. 64-2166, it is necessary to accurately convert the thought process of experts into IF-THEN rules, but this is not practical. It is difficult to In addition, it is necessary to create combination evaluation rules in addition to information evaluation rules, resulting in a problem that the configuration is complicated.

Accordingly, the present invention provides an event-numeric correlation display system, an event-numeric correlation model creation system, and an event-numeric correlation model creation system that can handle both numerical information and event information, and that do not require rules for the thought process of experts. The purpose is to provide a decision support system.

[0007]

[Means for Solving the Problems] In a first aspect, the present invention provides an event time series database that stores event information such as news in the form of text data together with time information, and a database that stores event information such as news in the form of text data and time information. search processing means for performing a full text search on the event time series database to determine whether the included event information exists in the analysis interval;
Provided is an event numerical value correlation display system characterized by comprising a display means for displaying text data including event information when the event information exists.

In a second aspect, the present invention provides an event time series database that stores event information such as news together with time information, a numerical time series database that stores numerical information such as prices together with time information, and a numerical time series database that stores event information such as news with time information. An output layer associated with information, an event information input layer associated with event information that affects the target numerical information, a numerical information input layer associated with the numerical information that influences the target numerical information, and an intermediate layer. a neural network having layers;
Search the event time series database to determine whether event information that affects the target numerical information exists during the learning period, and provide the result to the event information input layer, as well as numeric values that affect the target numerical information. Information is obtained by referring to the numerical time series database and given to the numerical information input layer, and numerical information of the object affected by the event information and numerical information in the learning period is obtained by referring to the numerical time series database, and the numerical information is obtained by referring to the numerical time series database. An event numerical correlation model creation system is provided, characterized by comprising a learning processing means for calculating weights of a neural network applied to an output layer.

In a third aspect, the present invention provides an event information receiving means for receiving event information such as news via a communication line, an event time series database for storing the received event information together with time information, and a communication Numerical information receiving means that receives numerical information such as prices via a line, a numerical time series database that stores the received numerical information together with time information, an output layer associated with the target numerical information, and the target numerical value. A neural network has an event information input layer associated with event information that affects information, a numerical information input layer and an intermediate layer that are associated with numerical information that affects target numerical information, and Search the event time series database for influencing event information and provide the results to the event information input layer, and obtain numerical information that influences the target numerical information by referring to the numerical time series database. Provided is a decision-making support system characterized by comprising processing display means for applying numerical information to the numerical information input layer and extracting and displaying the output of the output layer at that time.

0010

[Operation] In the event numerical correlation display system of the present invention, the event time series database stores event information such as news as text data together with time information. Such text data event information is available online from newspaper companies and the like. Specify or preset keywords and analysis intervals related to event information that may affect the numerical information to be analyzed, and search to see if event information that includes those keywords exists in the analysis interval. A processing means performs a full text search on the event timeline database. Then, when event information exists, the display means displays text data including the event information. In this way, the correlation between event information and numerical information can be easily grasped visually.

In the event numerical correlation model creation system of the present invention, the neural network has an event information input layer that is associated with event information that affects target numerical information. Then, the learning processing means searches the event time series database to find out whether there is event information that affects the target numerical information during the specified learning period, and provides the result to the event information input layer. and train the neural network. Therefore, it is possible to create an event-numeric correlation model that can handle both event information and numerical information without creating rules for experts' thought processes.

In the decision support system of the present invention, the event information receiving means receives event information such as news in real time via a communication line. Further, the numerical information receiving means receives numerical information such as prices in real time via a communication line. Then, the processing display means inputs this real-time information into the neural network,
Display the output at that time. Therefore, predictions incorporating event information can be viewed in real time, making it possible to provide accurate and timely support for decision-making.

[0013]

[Examples] Examples of the present invention will be described below. Note that the present invention is not limited thereby.

FIG. 1 shows a configuration diagram of an event numerical correlation analysis system 1 according to an embodiment of the present invention.

This event numerical correlation analysis system 1 includes an event time series database 2 that stores event information such as news in the form of text data together with time information, a numerical time series database 3 that stores stock price information together with time information, and a numerical time series database 3 that stores stock price information together with time information. an event information search device 4 that searches for event information in the numerical time series database 2 by full text search; a database management device 5 that manages reading and writing of stock price information in the numerical time series database 3; and an engineering workstation. A processing device 6 that creates an event numerical correlation model and performs decision-making support processing using a neural network (see Figure 5) using (EWS) etc., a display device 7 that serves as an interface with a decision maker, and a learning system for the neural network. A neuro-learning processor 8 performs processing, a receiving device 9 receives event information such as the latest news via a communication line, a receiving device 10 receives real-time stock price information via a communication line, and a receiving device 10 receives real-time stock price information via a communication line. It consists of an event numerical value related table 11 that summarizes influencing events, and a communication network 12 that connects each of the devices 2 to 11 with an Ethernet network or the like.

FIG. 2 shows the configuration of the event time series database 2. Text data such as news that occurred (or was announced) at each time is stored corresponding to time parameters t1, t2, . . . .

FIG. 3 shows the configuration of the numerical time series database 3. The stock price of each stock name at each time is stored corresponding to time parameters t1, t2, . . . .

FIG. 4 shows the structure of the event numerical value association table 11. Items of events that cause fluctuations in the stock price of each stock issue are tabulated. Each element written in one item column is combined with an AND condition, and each item column is combined with an OR condition. For example, the fluctuation factor of the stock name M1 is the occurrence of event information such as "increase in US GNP" or "decrease in interest rates." The event numerical value related table 11 is set by an expert.

The event information search device 4 searches the event time series database 2 for events stored in the event numerical value association table 11. In this case, if "United States" is used, thesauruses such as "America" and "USA" are also searched.

FIG. 5 shows the structure of the neural network. This neural network 70 has a set of weights αij for each stock name.

[0021] The neuro-learning processor 8 provides data to the input layer and provides desirable data for the input data to the output layer, with a set of weights αij for a certain stock name set in the neural network 70. The input/output relationship is fed back to the weights αij using the backprogation method, and the set of weights αij is updated (the neural network 70 is made to learn). The reason why the neuro learning processor 8 is used is to perform learning calculations at high speed.

Conventionally, only numerical information data such as short-term averages and long-term averages of stock prices was handled as input data, but in the present invention, as shown in FIG. It also handles event information data. That is, numerical information that influences market fluctuations is input to input layers a, b, c, . . . as in the conventional case. In addition, input layers φ1, φ2
, φ3,..., if the event information of the item number 1, 2, 3,... of the event numerical value related table 11 is found in the event information of the noted time ti of the event time series database 2, for example, it is "1", and if it is not found, If not, enter "0", for example. The output data is the rate of change in stock prices.

The processing functions of the event numerical correlation analysis system 1 are comprised of the following three. (1) Event numerical correlation display that displays the correlation between event information and numerical information (2) Event numerical correlation model that creates a stock price dynamic characteristic model that takes events into consideration from event time series database 2 and numerical time series database 3 Creation (3) Real-time decision support using the created event numerical correlation model The above functions can be selected from the display device 7 using a menu method. FIG. 6 is an example of the menu display.

First, the processing function of event numerical correlation display will be explained. The purpose of the event numerical correlation display is to allow the decision maker to judge the correlation by outputting the relationship between the event information and the numerical information in the analysis interval (period) specified by the decision maker to the display device 7.

FIG. 7 shows an example of screen output for event numerical correlation display. A stock price fluctuation chart 70 in the analysis interval designated by the decision maker is displayed based on the numerical time series database 3. Further, if an event listed in the event numerical value related table 11 is found in the analysis section of the event time series database 2, a mark ▲ is displayed on the time fluctuation chart 70 corresponding to the event. When one of the ▲ marks is pointed with the pointer m, the corresponding event information is displayed in the prompter area 81.

FIG. 8 shows a flowchart of the processing performed by the processing device 6 that performs the display shown in FIG. In step 101, the decision maker specifies the stock brand to be analyzed and the analysis interval. The specified stock brand and analysis interval analysis are transferred from the processing device 6 to the event information search device 4 and the database management device 5. In step 102, the processing device 6 takes in the event for the stock brand to be analyzed from the event value related table 11.

In step 103, the processing device 6 transfers the event captured in step 102 to the event information retrieval device 4. The event information search device 4 retrieves the transferred event (
Full-text search is performed for event information in the analysis section of the event time series database 2 using keywords (and thesaurus). Then, event information containing keywords (sentences containing keywords or paragraphs containing those sentences, etc.) is extracted,
Time information corresponding to the event information is transferred to the processing device 6.

In step 104, the processing device 6 generates an extraction information table 90 as shown in FIG. 9 based on the transferred event information and time information.

In step 105 , the database management device 5 reads the stock price data in the analysis interval of the stock issue to be analyzed from the numerical time series database 3 and transfers it to the processing device 6 .

In step 106, the processing device 6 converts the stock price data transferred from the database management device 5 into a stock price chart 70 and displays it on the display device 7. Further, a mark ▲ of an appropriate color is displayed on the stock price chart 70 corresponding to time Ti in the extraction information table 90.

[0031] In step 107, input using the mouse is accepted. There are two types of mouse input: a ▲ mark instruction and an end instruction. If the instructions marked ▲ are followed,
Proceed to step 108. If there is a termination instruction, it will terminate.

In step 108, the color of the designated ▲ mark is changed (the color of the ▲ marks other than the designated ▲ mark is made the same as the color added in step 106).

In step 109, the prompter area 8
1 is cleared, the event information corresponding to the indicated ▲ mark is read from the extraction information table 90, and the prompter area 8 is cleared.
Display on 1. After displaying, the process returns to step 107.

FIG. 10 shows another example of screen output for event numerical correlation display. A stock price fluctuation chart 70 in the analysis interval designated by the decision maker is displayed based on the numerical time series database 3. In addition, a window 92 for inputting a keyword is displayed, and an event search section for searching for the occurrence of an event related to the keyword input in the window 92 is displayed as a ▲ mark on the fluctuation chart 70. Further, the input keyword is displayed in the window 92, and the time at which an event related to the keyword occurs is displayed as a ▽ mark on the fluctuation chart 70. moreover,
When one of the ▽ marks is pointed with the pointer m, the corresponding event information is displayed in the prompter area 81.

FIG. 11 shows a flowchart of the processing of the processing device 6 that performs the display shown in FIG. In step 121, the decision maker specifies the stock brand to be analyzed and the analysis interval. The specified stock brand and analysis interval analysis is performed by the processing device 6.
event information retrieval device 4 and database management device 5
will be forwarded to. In step 122 , the database management device 5 reads the stock price data in the analysis interval of the stock brand to be analyzed from the numerical time series database 3 and transfers it to the processing device 6 .

In step 123, the processing device 6 converts the stock price data transferred from the database management device 5 into a stock price chart 70 and displays it on the display device 7.

In step 124, the decision maker specifies an event search interval on the stock price chart 70. In step 125, input via the keyboard is accepted. There are two types of keyboard input: keyword input and termination instruction. If a keyword is entered,
Proceed to step 126. If there is a termination instruction, it will terminate. In step 126, keywords are read. In step 127, the keyword and event search section are transferred to the event information search device 4. In step 128, the event information search device 4 searches the event time series database for event information on the stock issue to be analyzed, and searches the event information database for event information related to the keyword (
(including the keyword) and during the event search period. Then, the time corresponding to the extracted event information is displayed on the fluctuation chart 70 as a ▽ mark. Further, when any of the ▽ marks is pointed with the pointer m, the corresponding event information is displayed in the prompter area 81.

Next, the processing function for creating an event numerical correlation model will be explained. The purpose of creating the event numerical correlation model is to obtain a set of weights αij of the neural network shown in FIG. 5 for the target stock issue.

FIG. 12 is a flowchart of the process of creating an event numerical correlation model. In step 131, the decision maker specifies a stock brand and a learning interval for which an event numerical correlation model is to be created.

In step 132, the processing device 6 takes in event information for the target stock brand from the event value related table 11. Then, it is transferred to the event information search device 4.

In step 133, the event information search device 4
However, a full text search is performed to determine whether the captured event information exists during the learning period of the event information time series database 2. Then, based on the search results, input data of event information to be input into the neural network is created. That is, each time tA, tA+1,..., tB during the learning period
“1” if event information exists, otherwise “0”
is transferred from the event information search device 4 to the processing device 6. The processing device 6 fills the event information input data column of the learning data table 130 in FIG. 13 with the transferred event information input data.

In step 134, based on the contents of the numerical time series database 3, the moving average value of the stock price, trading volume, RSI index value, etc. are calculated at each time tA, tA+1 during the learning period.
,...,tB, and set the maximum value during the learning period to "1".
”, the minimum value is normalized as “0” and input data of numerical information is created. Then, the learning data table 130
Fill in the input data field with numerical information.

In step 135, the stock price fluctuation rate is calculated for each time tA, tA+1,..., tB during the learning period based on the contents of the numerical time series database 3, and the stock price fluctuation rate is β1% or more. is “1” for node A, less than β1% to more than β2% is “1” for node B, and so on.
Create output data. Then, the output data column of the output layer of the learning data table 130 is filled in.

In step 136, the neuro-learning processor 8 calculates the value of the neural network weight αij based on the contents of the learning data table 130 by the back-propagation method. Since the backpropagation method requires repeated calculations, a high-speed neurolearning processor 8 is used. The obtained value of the neural network weight αij is transferred to the processing device 6.

In step 137, the processing device 6 converts the transferred neural network weight αij into a set of weights αij for the target stock brand.
save. As a result, an event numerical correlation model for the target stock issue has been created. FIG. 14 illustrates a neural network parameter table that stores a set of weights αij for each stock issue.

Next, the processing function of real-time decision support will be explained. Real-time decision support aims to provide real-time input data to a neural network, obtain output data, and present it to a decision maker.

FIG. 15 is a real-time decision support table 150 in which real-time input data and output data obtained from the neural network for each stock issue M1, M2, . . . are tabulated.

FIG. 16 is a flowchart of processing for managing the real-time decision support table 150 in the latest state. In step 171, the processing device 6
Accepting input interrupts from the receiving device 9 for receiving real-time event information and input interrupts from the receiving device 10 for receiving real-time numerical information,
Permit the transfer of information.

In step 172, it is determined whether the transferred information is event information or numerical information. If it is event information, the process proceeds to step 173; if it is numerical information, the process proceeds to step 176.

In step 173, it is checked whether the input event information exists in the event value related table 11, with the stock issue M1 as the first target. If there is, the value in the corresponding column of the input data of the real-time decision support table 150 is set to "1". Also, review other columns of input data and change “1” to “0” if necessary (
the effect of the event has expired).

In step 174, the processing device 6 calculates the output data of the neural network based on the values stored in the input data column, and stores it in the output data column of the real-time decision support table 150.

In step 175, it is checked whether or not all stock issues have been completed. If not, the count is increased from i to i+1 and the process is repeated in step 1.
Return to 73. When the process is completed for all stock issues, the process ends.

On the other hand, in step 176, it is checked whether there is a column corresponding to the input numerical data in the input data column of the stock brand M1 in the real-time decision support table 150, and if there is a column, the value is stored.

In step 177, the processing device 6 calculates the output data of the neural network based on the values stored in the input data column, and stores it in the output data column of the real-time decision support table 150.

In step 178, it is checked whether all stock issues have been completed. If not, the count is increased from i to i+1 and the process is repeated in step 1.
Return to 76. When the process is completed for all stock issues, the process ends.

The output data column of the real-time decision support table 150 obtained above is the expected stock price fluctuation rate by the event numerical value correlation analysis system 1. Decision makers can refer to this to make decisions regarding stock trading.

[0057] In the above embodiment, news and stock prices were used as examples, but models related to factors of both event information and numerical information, such as local events and product sales, or background music and accident rates, may also be used. The present invention is applicable to such cases.

[0058]

According to the event numerical value correlation display system of the present invention, the correlation between event information and numerical information can be easily grasped visually. Further, according to the event-numeric correlation model creation system of the present invention, an event-numeric correlation model that can handle both event information and numerical information can be created without creating rules for the thinking process of experts.

Furthermore, according to the decision-making support system of the present invention, predictions incorporating event information can be viewed in real time, making it possible to support decision-making accurately and without time delay.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram of an event numerical correlation analysis system 1 according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram of the configuration of an event time series database.

FIG. 3 is an explanatory diagram of the configuration of a numerical time series database.

FIG. 4 is a configuration explanatory diagram of an event numerical value related table.

FIG. 5 is a configuration diagram of a neural network.

FIG. 6 is an exemplary diagram of a function selection menu.

FIG. 7 is a diagram illustrating an event numerical correlation display.

FIG. 8 is a flowchart of processing for displaying FIG. 7;

FIG. 9 is a configuration explanatory diagram of an extraction information table.

FIG. 10 is another exemplary diagram of event numerical correlation display.

FIG. 11 is a flowchart of processing for displaying FIG. 10;

FIG. 12 is a flowchart of processing for creating an event numerical correlation model.

FIG. 13 is a configuration explanatory diagram of a learning data table.

FIG. 14 is an explanatory diagram of a set of neural network weights.

FIG. 15 is a configuration explanatory diagram of a real-time decision support table.

FIG. 16 is a flow diagram of processing for managing the real-time decision support table of FIG. 15;

[Explanation of symbols]

1 Event numerical correlation analysis system 2 Event time series database 3 Numerical time series database 4 Event information search device 5 Database management device 6 Processing equipment 7 Display device 8 Neuro learning processor 9 Receiving device 10 Receiving device 11 Event numerical value related table 12 Network 50 Neural network

Claims (5)

[Claims] Claim 1: An event time series database that stores event information such as news in text data together with time information, and an event information database that stores event information such as news as text data together with time information; What is claimed is: 1. An event numerical correlation display system comprising: a search processing means for performing a full text search on a time series database; and a display means for displaying text data including event information when event information exists. Claim 2: An event time series database that stores event information such as news in the form of text data together with time information; a numerical time series database that stores numerical information such as prices together with time information; a search processing means for performing a full text search on the event time series database to determine whether event information including a keyword exists in the analysis interval; and a search processing means for obtaining numerical information to be analyzed in the analysis interval from the numerical time series database, and a display means for displaying a mark on the chart at a position corresponding to the time of the event information when the event information is found by the search. 3. The event numerical correlation analysis system according to claim 2, further comprising mark selection means for selecting one of the marks, and display means for displaying text data including event information corresponding to the selected mark. An event numerical correlation display system characterized by: Claim 4: An event time series database that stores event information such as news together with time information, a numerical time series database that stores numerical information such as prices together with time information, and an output layer that is associated with the target numerical information. and a neural network having an event information input layer associated with event information that affects the numerical information of the target, and a numerical information input layer and an intermediate layer associated with the numerical information that influences the numerical information of the target; Search the event time series database to find out whether event information that affects the numerical information of the subject exists in the learning period, and provide the result to the event information input layer, as well as numerical information that affects the numerical information of the target. is obtained by referring to the numerical time series database and given to the numerical information input layer, and the numerical information of the target affected by the event information and numerical information in the learning period is obtained by referring to the numerical time series database and the output is obtained. A system for creating an event numerical correlation model, comprising a learning processing means for calculating weights for a neural network by applying weights to layers. 5. Event information receiving means for receiving event information such as news via a communication line; an event time series database for storing the received event information together with time information; and an event information receiving means for receiving event information such as news via a communication line; A numerical information receiving means for receiving information, a numerical time series database for storing the received numerical information together with time information, an output layer associated with the target numerical information, and event information that affects the target numerical information. A neural network having a numerical information input layer and an intermediate layer associated with a corresponding event information input layer and numerical information that affects the target numerical information, and event information that influences the target numerical information. Searching the event time series database and providing the results to the event information input layer, and obtaining numerical information that affects the target numerical information by referring to the numerical time series database and providing the result to the numerical information input layer. A decision support system characterized by further comprising processing display means for extracting and displaying the output of the output layer at that time.