JP5075962B2 - Attention target prediction device - Google Patents

Description

  Embodiments described herein relate generally to a target object prediction apparatus that predicts a target object from among a plurality of targets.

  In general, it may be necessary to select a target to be noticed in the next period from a plurality of targets.

  For example, in the securities market, it is necessary to make a decision on buying and selling a brand by selecting a brand to be noticed from many brands. For example, in plant monitoring, it is necessary to select a plant having a possibility of failure or the like from many plants and create a maintenance plan for the plant.

JP 2002-269357 A JP 2007-087354 A

  However, it is very difficult to select a target to be noticed from the large number of objects in a large amount of information for the large number of objects. Therefore, an apparatus (attention target prediction apparatus) that can automatically predict a target to be noticed from a large number of objects and present the target to be noticed is desired.

  Further, it is desirable for a user who uses such a target-of-interest prediction apparatus to easily and easily select a target to be noted from a large number of targets.

  Therefore, a problem to be solved by the present invention is to provide an attention target prediction device that allows a user to easily and easily select a target to be noticed from among a large number of objects.

  According to the embodiment, an attention object prediction device that predicts an object to be noticed from among a plurality of objects is provided.

  An attention target prediction apparatus according to the embodiment includes a target expression storage unit, a first data collection unit, a first evaluation value calculation unit, a second data collection unit, a change rate calculation unit, and a third Data collection means, second evaluation value calculation means, attention level calculation means, and presentation means are provided.

  The object expression storage means stores a character string related to the object for each object.

The first data collection unit collects text information including character strings within a plurality of predetermined periods, and acquires time-series text data including the collected text information in time-series order.

  The first evaluation value calculation means includes a character string related to the target stored in the target expression storage means for each target and time series text data acquired by the first data collection means. Based on the text information in the sequence order, the evaluation value in the time series order is calculated for each target.

The second data collection means collects, for each target, numerical information including numerical values related to the target within the predetermined plurality of periods, and acquires time-series numerical data including the numerical information in time series order. .

  The rate-of-change calculating means is configured to calculate the time for each target based on numerical values included in numerical information before and after the time series in the time-series order included in the time-series numerical data acquired by the second data collecting means for each target. The rate of change in sequence order is calculated.

Third data collection means collects the text information including a text string after a plurality of periods said predetermined.

  The second evaluation value calculating means applies a character string related to the object stored in the object expression storing means for each object and a character string included in the text information collected by the third data collecting means. Based on this, an evaluation value is calculated for each object.

  The degree-of-attention calculation means includes the evaluation values in the time series order calculated by the first evaluation value calculation means for each target, and the change rates in the time series order calculated by the change rate calculation means for each target. The attention level for each target is calculated using the evaluation value calculated by the second evaluation value calculation means for each target.

  The presenting means presents the attention level for each target calculated by the attention level calculation means.

The block diagram which shows the hardware constitutions of the attention object prediction apparatus which concerns on embodiment. The block diagram which mainly shows a function structure of the attention object prediction apparatus 30 shown in FIG. The figure which shows an example of the data structure of the object expression storage part 22 shown in FIG. The flowchart which shows the process sequence of the attention object prediction apparatus 30 which concerns on this embodiment. The flowchart which shows the process sequence of a time series prediction model update process. The flowchart which shows the process sequence of a text analysis process. The flowchart which shows the process sequence of object numerical analysis process. The figure which shows an example of the time series text data which contains the text information in a learning data period in time series order. The figure which shows an example of the object expression evaluation value of the time series order of object "A company" stored in the object expression evaluation part 32 inside. The figure which shows an example of the target expression evaluation value of the time series order of the object "B company" stored in the inside of the target expression evaluation part 32. FIG. The figure which shows an example of the object expression evaluation value of the time series order of object "C company" stored in the object expression evaluation part 32 inside. The figure which shows an example of the object expression evaluation value of the time series order of object "Company D" stored in the object expression evaluation part 32 inside. The figure which shows an example of the time series numerical data containing the numerical information collected about object "A company" in time series order. The figure which shows an example of the time series numerical data containing the numerical information collected about object "B company" in time series order. The figure which shows an example of the time series numerical data which contains the numerical information collected about object "C company" in time series order. The figure which shows an example of the time series numerical data containing the numerical information collected about object "D company" in time series order. The figure which shows an example of the target numerical value change rate of the target "A company" stored in the inside of the target numerical value change rate evaluation part 33 in the time series order. The figure which shows an example of the target numerical value change rate of the target "company B" stored in the inside of the target numerical value change rate evaluation part 33 in the time series order. The figure which shows an example of the target numerical value change rate of the target "Company C" stored in the inside of the target numerical value change rate evaluation part 33 in the time series order. The figure which shows an example of the target numerical value change rate of the target "Company D" stored in the inside of the target numerical value change rate evaluation part 33 in the time series order. The figure which shows an example of the time series prediction model produced | generated by the time series prediction model production | generation part. The figure which shows an example of the object expression evaluation value of object "A company" calculated by the object expression evaluation part 32. FIG. The figure which shows an example of the object expression evaluation value of object "B company" calculated by the object expression evaluation part 32. FIG. The figure which shows an example of the object expression evaluation value of object "Company C" calculated by the object expression evaluation part 32. FIG. The figure which shows an example of the object expression evaluation value of object "Company D" calculated by the object expression evaluation part 32. FIG. The figure which shows an example of the object expression evaluation value of the time series order of object "A company" after an update. The figure which shows an example of the object expression evaluation value of the time series order of object "B company" after an update. The figure which shows an example of the object expression evaluation value of the time series order of object "C company" after an update. The figure which shows an example of the object expression evaluation value of the time series order of object "D company" after an update. The figure which shows an example of the time series numerical data containing the numerical information collected about object "A company." The figure which shows an example of the time series numerical data containing the numerical information collected about object "B company." The figure which shows an example of the time series numerical data containing the numerical information collected about object "C company." The figure which shows an example of the time series numerical data containing the numerical information collected about object "D company." The figure which shows an example of the target numerical value change rate of the object "Company A" calculated by the target numerical value change rate evaluation part 33. The figure which shows an example of the target numerical value change rate of the object "Company B" calculated by the target numerical value change rate evaluation part 33. The figure which shows an example of the target numerical value change rate of the object "Company C" calculated by the target numerical value change rate evaluation part 33. The figure which shows an example of the target numerical value change rate of the object "Company D" calculated by the target numerical value change rate evaluation part 33. The figure which shows an example of the object numerical value change rate of the time series order of object "A company" after an update. The figure which shows an example of the object numerical value change rate of the time series order of the object "B company" after an update. The figure which shows an example of the object numerical value change rate of the time series order of the object "C company" after an update. The figure which shows an example of the target numerical change rate of the target "Company D" after an update in the time series order. The flowchart which shows the process sequence of object ranking process. The figure which shows an example of each attention degree of object "A company", "B company", "C company", and "D company" calculated by the object attention degree calculation part 35. FIG. The figure which shows an example of the ranking result of the attention degree for every object displayed by the object ranking display part 36. FIG. The figure which shows an example of the ranking result of the attention degree for every object displayed by the object ranking display part 36 after the parameter update.

  Hereinafter, embodiments will be described with reference to the drawings.

  FIG. 1 is a block diagram illustrating a hardware configuration of the attention target prediction apparatus according to the present embodiment. As shown in FIG. 1, the computer 10 is connected to an external storage device 20 such as a hard disk drive (HDD). The external storage device 20 stores a program 21 executed by the computer 10. The computer 10 and the external storage device 20 constitute an attention target prediction device 30.

  This attention object prediction device 30 has a function of predicting an object to be noticed from among a plurality of objects in, for example, a securities market. In this case, examples of the target include companies.

  FIG. 2 is a block diagram mainly showing a functional configuration of the attention target prediction apparatus 30 shown in FIG. As illustrated in FIG. 2, the attention target prediction device 30 includes a data collection unit 31, a target expression evaluation unit 32, a target numerical change rate evaluation unit 33, a time series prediction model generation unit 34, a target attention level calculation unit 35, and a target ranking. A display unit 36 is included. In the present embodiment, these units 31 to 36 are realized by the computer 10 illustrated in FIG. 1 executing the program 21 stored in the external storage device 20. This program 21 can be stored in advance in a computer-readable storage medium and distributed. Further, this program 21 may be downloaded to the computer 10 via, for example, a network.

  The attention target prediction device 30 includes a target expression storage unit 22. In the present embodiment, the target expression storage unit 22 is stored in, for example, the external storage device 20 illustrated in FIG.

  For each target described above, the target expression storage unit 22 stores a character string related to the target (hereinafter referred to as target expression). This object expression includes the name (abbreviation), nickname, and brand name of the object (company).

  The data collection unit 31 collects text information including a character string within a predetermined period (hereinafter referred to as a learning data collection period), and acquires time-series text data including the text information in time-series order. The text information collected by the data collection unit 31 includes, for example, a headline of news transmitted from a website or the like.

  In addition, the data collection unit 31 collects numerical information including numerical values related to the target within the learning data collection period described above for each target, and acquires time-series numerical data including the numerical information in time series order. The numerical information collected by the data collection unit 31 includes the stock price of the target (company). The data collection unit 31 collects numerical information from, for example, a website.

  Furthermore, the data collection unit 31 collects text information including character strings within a period from the learning data collection period to the present (hereinafter referred to as a predicted data collection period).

  The target expression evaluation unit 32 includes, for each target, the target expression (character string related to the target) stored in the target expression storage unit 22 and the time series order included in the time series text data acquired by the data collection unit 31. Based on the text information, an evaluation value in time series order is calculated for each target.

  Further, the target expression evaluation unit 32 sets the target expression for each target based on the target expression stored in the target expression storage unit 22 for each target and the text information collected by the data collection unit 31 within the prediction data collection period. The evaluation value is calculated.

  The target numerical change rate evaluation unit 33 sets the time series for each target based on the numerical values included in the numerical information before and after in the time series order included in the time series numerical data acquired by the data collection unit 31 for each target. Calculate the rate of change in order.

  The time-series prediction model generation unit 34 sets the time-series order evaluation values calculated by the target expression evaluation unit 32 for each target and the time-series order change rates calculated by the target numerical value change rate evaluation unit 33 for each target. Based on this, a time series prediction model to be described later is generated. The time series prediction model generated by the time series prediction model generation unit 34 is used when calculating the degree of attention for each target. The time series prediction model generated by the time series prediction model generation unit 34 is stored in the time series prediction model generation unit 34, for example.

  The target attention level calculation unit 35 calculates the attention level for each target. The target attention level calculation unit 35 calculates the evaluation value in time series order calculated by the target expression evaluation unit 32 for each target, the change rate in time series order calculated by the target numerical value change rate evaluation unit 33 for each target, and for each target. The degree of attention for each target is calculated using the evaluation value calculated by the target expression evaluation unit 32. Specifically, the target attention level calculation unit 35 is a time-series prediction model stored in the time-series prediction model generation unit 34 (a time-series prediction model generated by the time-series prediction model generation unit 34), for each target. The degree of attention for each target is calculated using the evaluation value calculated by the target expression evaluation unit 32 and a plurality of predetermined parameters.

  The target ranking display unit 36 presents the attention level for each target calculated by the target attention level calculation unit 35. Specifically, the target ranking display unit 36 rearranges the plurality of targets in the order of the attention degrees calculated by the target attention level calculation unit 35, and displays the plurality of targets in a ranking format.

  FIG. 3 shows an example of the data structure of the target expression storage unit 22 shown in FIG. As shown in FIG. 3, the target expression storage unit 22 stores a target and a character string (target expression) related to the target in association with each other. Here, it is assumed that “Company A”, “Company B”, “Company C”, and “Company D” exist as targets.

  In the example illustrated in FIG. 3, the target expression storage unit 22 stores the target “Company A” and the target expressions “Company A”, “X1 brand”, and “X2 brand” in association with each other. This indicates that the character strings related to the target “Company A” are “Company A”, “X1 brand”, and “X2 brand”.

  The target expression storage unit 22 stores the target “B company” and the target expressions “B company”, “B1 company”, and “Y brand” in association with each other. According to this, it is indicated that the character strings related to the target “Company B” are “Company B”, “Company B1”, and “Y brand”.

  The target expression storage unit 22 stores the target “Company C” and the target expressions “C company” and “Z brand” in association with each other. This indicates that the character strings related to the target “C company” are “C company” and “Z brand”.

  The target expression storage unit 22 stores the target “Company D” and the target expressions “Company D” and “D1 company” in association with each other. According to this, it is indicated that the character strings related to the target “Company D” are “Company D” and “Company D1”.

  Next, with reference to the flowchart of FIG. 4, the processing procedure of the attention object prediction apparatus 30 which concerns on this embodiment is demonstrated.

  First, initialization of the attention object prediction device 30 is executed (step S1). Here, the data collection unit 31 sets, for example, a website for collecting text information (news headlines), sets, for example, a website for collecting numerical information (target stock price), and a unit time for the collection (data collection). (Unit time) and the maximum number of times of collection (maximum number of data collection) are set.

  Note that the values of these setting items may be described in a file or the like in advance, or may be input (set) by a user through an interface.

  The data collection unit 31 sets a mode according to the user's operation (step S2). At this time, the user can select any one of the learning mode, the prediction mode, and the end mode, for example. The learning mode is a mode for collecting text information and numerical information and generating the above-described time series prediction model. The prediction mode is a mode for predicting a target to be noticed from a plurality of targets and presenting the target to be noticed. The end mode is a mode for ending the operation (processing) of the target prediction device 30.

  In step S2, for example, the mode selection from the user is waited for a certain time, and if the mode is not selected by the user even after the certain time has elapsed, the learning mode may be automatically set.

  Note that, as will be described later, the operation of the target object prediction device 30 differs depending on the mode set by the data collection unit 31.

  Next, the data collection unit 31 determines whether or not the learning mode is set (step S3). When it is determined that the learning mode is set (YES in step S3), a time series prediction model update process described later is executed (step S4). The time series prediction model update process is executed by the data collection unit 31, the target expression evaluation unit 32, the target numerical change rate evaluation unit 33, and the time series prediction model generation unit 34. If the process of step S4 is performed, it will return to above-mentioned step S2 and a process will be repeated.

  Note that, as described above, when the mode is not selected by the user in step S2, the learning mode is set. Therefore, unless the prediction mode or the end mode is explicitly selected by the user, the time series prediction is performed. The model update process is repeated. In this case, the time-series prediction model update process is repeated, for example, at intervals of the data collection unit time set in the initialization process of the attention target prediction device 30 described above.

  On the other hand, when it is determined that the learning mode is not set (NO in step S3), the data collection unit 31 determines whether the prediction mode is set (step S5).

  When it is determined that the prediction mode is set (YES in step S5), a target ranking process described later is executed (step S6). The target ranking process is executed by the data collection unit 31, the target expression evaluation unit 32, the target attention level calculation unit 35, and the target ranking display unit 36. If the process of step S6 is performed, it will return to above-mentioned step S2 and a process will be repeated.

  On the other hand, when it is determined that the prediction mode is not set (NO in step S5), that is, when the end mode is set, the operation (process) of the attention target prediction device 30 is ended.

  Next, a processing procedure of the above-described time series prediction model update process (the process of step S4 shown in FIG. 4) will be described with reference to the flowchart of FIG. The time series prediction model update process is a process executed when the learning mode is set as described above.

  First, the data collection unit 31 sets a learning data collection period (step S11). For example, when the time series prediction model update process has not been executed before, that is, when the current time series prediction model update process is the first time, the initialization process of the target prediction device 30 (steps shown in FIG. 4) The time obtained by multiplying the data collection unit time set in S1) by the value obtained by subtracting 1 from the maximum number of data collections (ie, “data collection unit time * (maximum number of data collections−1)”) is traced back from the present. The period from the point in time (immediately after) to the present is set as the learning data collection period. Specifically, 1 day is set as the data collection unit time, 5 times is set as the maximum number of data collection, and the current date is “2010/1/8”, the current date The period from the date “2010/1/8” to the time “1 day * (5-1)” from the present to the present, that is, “2010/1/4 to 2010/1/8” A period of five days (that is, the maximum number of data collection times of the data collection unit time) is set as the learning data collection period.

  On the other hand, when the time series prediction model update process has been executed previously, that is, when the current time series prediction model update process is the second time or later, the data collection unit 31 updates the previous time series prediction model update process. A period from the learning data collection period set in the process (that is, after the end of the time series prediction model update process) to the present (that is, a period of one data collection unit time) is set as the learning data collection period. Specifically, the learning data collection period set in the previous time series prediction model update process is “2010/1/4 to 2010/1/8”, and the current date is “2010/1/9”. In some cases, a period of one day of “2010/1/9” (that is, one data collection unit time) is set as the learning data collection period.

  That is, when the current time series prediction model update process is the first time, a period longer than one data collection unit time (a period corresponding to the maximum number of data collections of the data collection unit time) is set as the learning data collection period. If the time series prediction model update process is the second or later, one data collection unit time is set as the learning data collection period.

  Next, text analysis processing is executed based on the learning data collection period set by the data collection unit 31 (step S12). In this text analysis process, time-series text data within a learning data collection period set by the data collection unit 31 is acquired, and an evaluation value for each target (hereinafter referred to as target expression) based on the acquired time-series text data. Evaluation value and notation) are calculated.

  When the learning data collection period set by the data collection unit 31 is longer than one data collection unit time (that is, when the current time series prediction model update process is the first time), in the text analysis process, A target expression evaluation value for each data collection unit time within the data collection period (that is, a target expression evaluation value in time series order) is calculated for each target. On the other hand, when the learning data collection period set by the data collection unit 31 is one data collection unit time (that is, when the current time series prediction model update process is the second or later), in the text analysis process, A target expression evaluation value for one data collection unit time within the learning data collection period is calculated for each target. Details of the text analysis process will be described later.

  When the text analysis process is executed, the target expression evaluation unit 32 updates the target expression evaluation value in time series order for each target based on the target expression evaluation value for each target calculated in the text analysis process. (Step S13). Note that the target expression evaluation value in time series order for each target is stored in the target expression evaluation unit 32, for example.

  As described above, when the learning data collection period is longer than one data collection unit time, in step S13, the target expression evaluation values in the chronological order for each target calculated in the text analysis process (that is, the learning data collection period). The object expression evaluation value for each data collection unit time) is stored as it is. On the other hand, when the learning data collection period is one data collection unit time, in step S13, the oldest date in the target expression evaluation value in time series order for each target already stored in the previous time series prediction model update process The target expression evaluation value of (time zone) is deleted, and the target expression evaluation value calculated in the text analysis process (that is, the target expression evaluation value for one collection unit time) is the last of the target expression evaluation values in the time series order Stored in That is, the same number of target expression evaluation values as the maximum number of data collections are stored for each target in the target expression evaluation unit 32.

  Next, the target numerical analysis process is executed based on the learning data collection period set by the data collection unit 31 (step S14). In this target numerical analysis process, time-series numerical data within the learning data collection period set by the data collection unit 31 is acquired, and the rate of change for each target (hereinafter referred to as target) based on the acquired time-series numerical data. Numerical value change rate).

  As in the text analysis process described above, when the learning data collection period set by the data collection unit 31 is longer than one data collection unit time, in the target numerical analysis process, the data collection unit within the learning data collection period The target numerical change rate for each time (that is, the target numerical change rate in time series order) is calculated for each target. On the other hand, when the learning data collection period set by the data collection unit 31 is one data collection unit time, in the target numerical analysis process, the target numerical value change rate for one data collection unit time within the learning data collection period is Calculated for each subject. Details of the target numerical analysis process will be described later.

  When the target numerical value analysis process is executed, the target expression evaluation unit 32 calculates the target numerical value change rate in time series order for each target based on the target numerical value change rate for each target calculated in the target numerical value analysis process. Update (step S15). Note that the target numerical value change rate in time series order for each target is stored in the target numerical value change rate evaluation unit 33, for example.

  As described above, when the learning data collection period is longer than one data collection unit time, in step S15, the target numerical value change rate in chronological order for each target calculated in the target numerical value analysis processing (that is, learning data collection) The target numerical value change rate for each data collection unit time within the period) is stored as it is. On the other hand, when the learning data collection period is one data collection unit time, in step S15, the oldest date in the target numerical change rate in time series order for each target already stored in the previous time series prediction model update process The target numerical change rate in (time zone) is deleted, and the target numerical change rate calculated in the target numerical analysis process (that is, the target expression evaluation value for one collection unit time) is the target numerical change rate in the time series order. Stored last.

  As will be described later, the target numerical change rate is calculated based on the numerical values included in the numerical information before and after the numerical information in the time series order included in the time-series numerical data. The target numerical value change rate, which is one less than the number of target expression evaluation values for each target (that is, the maximum number of data collection times) stored in the target expression evaluation unit 32, is stored for each target. .

Next, the time-series prediction model generation unit 34 calculates the target expression evaluation value (target expression evaluation value stored in the target expression evaluation unit 32) and the target numerical value in time series order for each target calculated in the text analysis process. A time series prediction model is generated based on the target numerical rate of change in time series order for each target calculated in the analysis process (target numerical rate of change stored in the target numerical rate change rate evaluation unit 33) (step S16). ). In this case, the time series prediction model generation unit 34 generates a time series prediction model using, for example, a formula (1) prepared in advance.

In Expression (1), p represents a value obtained by subtracting 1 from the maximum number of data collections set in the initialization process of the target object prediction device 30 (that is, “maximum number of data collections−1”). T represents a target. Further, i represents each date when the oldest date (first date) is set to 1 in the time series order of the target expression evaluation value and the target numerical value change rate.

  In Expression (1), α represents an attenuation rate and is a parameter indicating how much past results are reflected. α takes a value in the range [0, 1], and the larger this value, the more the past result is reflected. When α = 0, it is regarded that “0 to the 0th power is 1.”

  On the other hand, in Expression (1), β represents the degree of importance of expression, and is a parameter representing how much importance is given to the target expression evaluation value compared to the target numerical value change rate when evaluating the target t. β takes a value in the range of [0, 1], and the larger this value, the more important the target expression evaluation value.

  As described above, the time series prediction model is generated by Expression (1) using the target expression evaluation value in the time series order for each target and the target numerical value change rate in the time series order for each target. That is, the time series prediction model includes each value calculated for each target with respect to a specific set of parameters (α and β) described in Equation (1).

  The time series prediction model generated in step S16 is stored in the time series prediction model generation unit 34, for example. In this way, the time series prediction model is updated each time the time series prediction model update process is executed.

  When the time series prediction model update process is executed as described above, for example, the learning data collection period set in the time series prediction model update process is set in the attention target prediction apparatus 30.

  Next, the processing procedure of the above-described text analysis process (the process of step S12 shown in FIG. 5) will be described with reference to the flowchart of FIG. This text analysis process is executed by the data collection unit 31 and the target expression evaluation unit 32.

  First, the data collection unit 31 collects text information including character strings within the learning data collection period set in step S11 shown in FIG. 5 (step S21). In this case, the data collection unit 31 accesses the website (website for collecting text information) set in the initialization process of the attention target prediction apparatus 30 described above, and is given a date within the learning data collection period, for example. Collect (download) text information. The date given to the text information collected by the data collection unit 31 represents the date on which the text information (for example, news headline) was transmitted. In addition, when a plurality of websites are set in the initialization process of the attention target prediction device 30, text information is collected from each website.

  The data collection unit 31 arranges the collected text information in the date order (that is, time series order) given to the text information, and acquires time series text data including the text information in time series order.

  The data collection unit 31 passes the acquired time series text data to the target expression evaluation unit 32. Note that the time-series text data passed to the target expression evaluation unit 32 is stored in the target expression evaluation unit 32, for example.

  Hereinafter, the target expression evaluation unit 32 performs the processing of steps S22 to S24 for each piece of text information in time series order included in the time series text data passed from the data collection unit 31.

  The target expression evaluation unit 32 extracts one of the time-series text information included in the time-series text data from the time-series text data passed from the data collection unit 31 (step S22).

  The target expression evaluation unit 32 determines whether or not the extracted text information includes the target expression (character string related to the target) stored in the target expression storage unit 22 (step S23). In step S <b> 23, processing is executed for each target expression stored in the target expression storage unit 22.

  When it is determined that the text information includes the target expression (YES in step S23), the target expression evaluation unit 32 is associated with the target expression in the target expression storage unit 22 on the date given to the text information. 1 is added to the score (number of cases) of the target (step S24).

  On the other hand, when it is determined that the text information does not include the target expression (NO in step S23), the process in step S24 is not executed.

  Next, the target expression evaluation unit 32 determines whether or not the processing in steps S22 to S24 described above has been executed for all text information included in the time-series text data passed from the data collection unit 31 (step S25). ).

  When it is determined that the processing has not been executed for all text information (NO in step S25), the process returns to the above-described step S22 and is repeated. In this case, text information for which the processing of steps S22 to S24 has not been executed is extracted in step S22. As described above, by repeating the processing of steps S22 to S24 for each piece of text information, it is possible to obtain a score for each object on each date within the learning data collection period.

On the other hand, when it is determined that the processing has been executed for all the text information (YES in step S25), the target expression evaluation unit 32 calculates a target expression evaluation value for each target on each date within the learning data collection period ( Step S26).

  When all of the target expression evaluation values for each target on each date within the learning data collection period are calculated by the target expression evaluation unit 32, the text analysis process ends.

The target expression evaluation values for each target on each date within the learning data collection period calculated by the target expression evaluation unit 32 are aligned for each target and in chronological order (date order), as described above. It is stored inside the evaluation unit 32.

  In the process of step S23 shown in FIG. 6 described above, it has been described as simply determining whether or not the text information includes the target expression. For example, a proper noun is extracted from the text information based on natural language processing technology, The proper noun may be configured to determine whether or not there is a thing that matches the target expression. Further, it is determined whether or not the textual information includes the textual expression by adding information such as character delimiter information or part of speech to the textual expression and comparing it with the morphological analysis result of the text information, for example. It may be a configuration.

  Further, in step S24, the number of text information including the target expression is simply counted as a score. For example, the weight of the text information using the time associated with the text information or information on the site is taken into consideration. A value that takes into account may be used as the score.

  Next, the procedure of the target numerical value analysis process (the process of step S14 shown in FIG. 5) will be described with reference to the flowchart of FIG. This target numerical value analysis process is executed by the data collection unit 31 and the target numerical value change rate evaluation unit 33.

  First, the data collection unit 31 collects, for each target, numerical information including numerical values related to the target within the learning data collection period set in step S11 shown in FIG. 5 (step S31). In this case, the data collection unit 31 accesses a website (website that collects numerical information) set in the initialization process of the target prediction device 30, and is given a date within the learning data collection period, for example. Collect (download) numerical information. The date given to the numerical information collected by the data collection unit 31 represents the date when the numerical information (for example, the target stock price) is transmitted. In addition, when a plurality of websites are set in the initialization process of the attention target prediction device 30, text information is collected from each website.

  The data collection unit 31 arranges the numerical information collected for each target in the date order (that is, the time series order) given to the numerical information, and the time series for each target includes the numerical information in the time series order. Get numeric data.

  The data collection unit 31 passes the acquired time-series numerical data for each target to the target numerical value change rate evaluation unit 33. The time-series numerical data for each target passed to the target numerical value change rate evaluation unit 33 is stored, for example, inside the target numerical value change rate evaluation unit 33. In addition, the target numerical analysis process described here is executed in the time series prediction model update process as described above. However, if the current time series prediction model update process is the second time or later, Time series numerical data (hereinafter referred to as previous time series numerical data) including numerical information collected for each target in step S31 of the target numerical analysis process in the series prediction model update process is a target numerical change rate evaluation unit. It is assumed that it is stored inside 33.

  Hereinafter, the target numerical value change rate evaluation unit 33 performs the processing of steps S32 to S36 for each time-series numerical data for each target passed from the data collection unit 31.

  The target numerical value change rate evaluation unit 33 extracts time-series numerical data of one target from the time-series numerical data for each target passed from the data collection unit 31 (step S32).

  Next, the target numerical change rate evaluation unit 33 performs the processes of steps S33 to S35 for each piece of numerical information included in the extracted time-series numerical data in time-series order. The target numerical change rate evaluation unit 33 processes numerical information included in the extracted time-series numerical data, for example, in time-series order.

  The target numerical change rate evaluation unit 33 extracts one of the time-series numerical information included in the time-series numerical data from the time-series numerical data extracted in step S32 (step S33).

  The target numerical change rate evaluation unit 33 determines whether there is numerical information before the numerical information extracted in step S33 based on the time-series numerical data stored in the target numerical change rate evaluation unit 33. Determination is made (step S34). Here, even if the numerical information extracted in step S33 is the first numerical information in the time-series order of the numerical information included in the time-series numerical data extracted in step S32, as described above, the previous time-series numerical value is obtained. When data remains, it is determined that there is numerical information before the numerical information extracted in step S33.

When it is determined in step S33 that there is numerical information before the extracted numerical information, the target numerical change rate evaluation unit 33 determines the numerical information immediately before the numerical information ( Hereinafter, based on the previous numerical value information), the target numerical value change rate of the target from which the time-series numerical data is extracted in step S32 on the date given to the previous numerical value information is calculated (step S35).

  Next, the target numerical change rate evaluation unit 33 determines whether or not the above-described processing of steps S33 to S35 has been executed for all numerical information included in the time-series numerical data extracted in step S32 (step S36). ).

  When it is determined in step S32 that processing has not been executed for all numerical information included in the time-series numerical data extracted in step S32 (NO in step S36), the process returns to step S33 described above and is repeated. In this case, numerical information for which the processing of steps S33 to S35 has not been executed is extracted in step S33.

Here, when the processing of steps S33 to S35 is repeated for all the numerical information included in the time-series numerical data extracted in step S32 as described above, the time in step S32 on each date within the learning data collection period. The target numerical value change rate of the target from which the series numerical data is extracted is calculated. The target numerical change rates of the target on each date within the learning data collection period calculated as described above are stored in the target numerical change rate evaluation unit 33 in the order of the date (time-series order).

  On the other hand, when it is determined that the processing has been executed for all the numerical information included in the time-series numerical data extracted in step S32 (YES in step S36), the target numerical change rate evaluation unit 33 reads from the data collection unit 31. It is determined whether or not the processing in steps S32 to S36 has been executed for all of the time-series numerical data for each passed target (step S37).

  When it is determined that the process has not been executed for all the time-series numerical data for each target (NO in step S37), the process returns to the above-described step S32 and is repeated. In this case, target time-series numerical data for which the processing of steps S32 to S36 has not been executed is extracted in step S32.

  On the other hand, when it is determined that the processing has been executed for all of the time-series numerical data for each target (YES in step S37), the target numerical analysis process is ended.

  By executing the target numerical value analysis process in this way, the target numerical value change rate in time series order is calculated for each target. The target numerical value change rate calculated in the target numerical value analysis process is stored in the target numerical value change rate evaluation unit 33 as described above.

  Next, the time series prediction model update process described above will be specifically described with reference to FIGS. Here, it is assumed that the plurality of targets (enterprises) include A company, B company, C company, D company, and the like.

  It is assumed that the data collection unit time set in the initialization process of the target prediction device 30 is one day, and the maximum number of data collection is five. Further, it is assumed that the target expression storage unit 22 has the data structure shown in FIG.

  In the time series prediction model update process, first, a learning data collection period is set by the data collection unit 31. Here, it is assumed that the current time-series prediction model update process is the first time. Here, when the current date is “2010/1/8”, the data collection unit time (1 day) and the maximum number of data collections (5 times) indicate that “2010/1/4 to 2010 / “1/8” is set as the learning data collection period.

  Next, text analysis processing is executed based on the learning data collection period set by the data collection unit 31.

  In the text analysis process, first, text information within the learning data collection period is collected by the data collection unit 31. The data collection unit 31 collects, for example, news headlines as text information from the website set in the initialization process of the attention target prediction device 30. Thereby, the data collection unit 31 acquires time series text data including the collected text information in time series order.

  Here, FIG. 8 is a time series text including text information (news headlines) in the learning data collection period “2010/1/4 to 2010/1/8” collected by the data collection unit 31 in time series. An example of data is shown.

  As shown in FIG. 8, the text information included in the time-series text data is given, for example, date, time, and transmission site (information indicating).

  Specifically, for example, the date “2010/4”, the time “8:00”, and the transmission site “z site” are assigned to the text information “B1 company announces new product”. This shows that the text information “B1 company announces new product” was sent from “z site” at “8:00” of “2010/4”.

  Next, the target expression evaluation unit 32 extracts one of the time-series text information included in the time-series text data from the time-series text data acquired by the data collection unit 31. Here, it is assumed that text information “B1 company announces new product” is extracted from the time-series text information included in the time-series text data shown in FIG.

  The target expression evaluation unit 32 determines whether or not the extracted text information includes the target expression stored in the target expression storage unit 22. Here, referring to the target expression storage unit 22 shown in FIG. 3, the text information “B1 company announces new product” is stored in the target expression storage unit 22 in association with the target “B company”. “B1 company” is included.

  In this case, the target expression evaluation unit 32 sets the target “B1” associated with the target expression “B1 company” on the date “2010/4” given to the text information “B1 company announces new product”. 1 is added to the score of “Company”.

  Here, the text information “B1 company announces new product” included in the time-series text data shown in FIG. 8 has been described, but the same applies to other text information.

  As described above, when the processing is executed for all the text information included in the time-series text data illustrated in FIG. 8, the target expression evaluation unit 32 sets the target expression evaluation value for each target on each date within the learning data collection period. Is calculated. The target expression evaluation value is calculated using the score for each target on each date within the learning data collection period and the above-described equation (2).

  Here, for example, calculation of the target expression evaluation value of the target “Company B” on the date “2010/4” will be specifically described. The target expression evaluation value of the target “Company B” on this date “2010/4” is calculated based on the date “2010” with respect to the sum of the scores of all targets on the date “2010/4” according to the above formula (2). / 1/4 ”is calculated as the ratio of the score of the target“ Company B ”.

  Here, the calculation of the target expression evaluation value of the target “Company B” on the date “2010/4” has been described, but other target expression evaluation values are calculated in the same manner.

  As described above, since the current time-series prediction model update process is the first time, when the target expression evaluation value for each target on each date in the learning data collection period is calculated in this way, the calculated target expression The evaluation values are stored in the target expression evaluation unit 32 as they are, arranged for each target and in time series order (date order).

  Here, FIGS. 9 to 12 show examples of the target expression evaluation values in order of time series for each target stored in the target expression evaluation unit 32.

  FIG. 9 shows an example of the target expression evaluation value in the time series order of the target “Company A” among a plurality of targets. FIG. 10 shows an example of the target expression evaluation value in the time series order of the target “Company B” among a plurality of targets. FIG. 11 shows an example of the target expression evaluation value in the time series order of the target “Company C” among a plurality of targets. FIG. 12 shows an example of the target expression evaluation value in the time series order of the target “Company D” among a plurality of targets.

  As shown in FIGS. 9 to 12, each date within the learning data collection period “2010/1/4 to 2010/1/8” is included in the target expression evaluation value in the time series order of each target. The assigned object expression evaluation value is included.

  Here, the target expression evaluation values in the chronological order of the targets “Company A”, “Company B”, “Company C”, and “Company D” have been described. The target expression evaluation values in the target time series order are also stored in the same manner.

  Next, the target numerical analysis process is executed based on the learning data collection period set by the data collection unit 31.

  In the target numerical analysis processing, first, numerical data including numerical values related to the target within the learning data collection period is collected by the data collection unit 31 for each target. The data collection unit 31 collects, for example, the stock price of each target (company) as numerical information from the website set in the initialization process of the attention target prediction device 30. Thereby, the data collection part 31 acquires time series numerical data including the collected numerical information in time series order.

  Here, FIGS. 13 to 16 show time-series numerical information (target stock prices) within the learning data collection period “2010/1/4 to 2010/1/8” collected by the data collection unit 31 for each target. An example of the time series numerical data included in order is shown.

  FIG. 13 shows an example of time-series numerical data including numerical information (company A's stock price) collected for a target “Company A” among a plurality of targets in time-series order. FIG. 14 shows an example of time-series numerical data including numerical information (stock price of company B) collected for a target “Company B” among a plurality of targets in time-series order. FIG. 15 shows an example of time-series numerical data including numerical information (stock price of company C) collected for a target “Company C” among a plurality of targets in time-series order. FIG. 16 shows an example of time-series numerical data including numerical information (stock price of company D) collected for a target “Company D” out of a plurality of targets in time-series order.

  Note that, as shown in FIGS. 13 to 16, date (information indicating) is given to the numerical information included in the time-series numerical data.

  Specifically, for example, the date “2010/4” is assigned to the numerical information “1020” included in the time-series numerical data shown in FIG. According to this, it is indicated that the numerical information “1020” is the stock price of the company A on the date “2010/1/4”.

  Here, the target numerical value change rate evaluation unit 33 extracts time-series numerical data of one target from the time-series numerical data for each target acquired by the data collection unit 31. Here, it is assumed that the time-series numerical data of the target “Company A” shown in FIG. 13 is extracted.

  Next, the target numerical change rate evaluation unit 33 extracts one of the time-series numerical information included in the time-series numerical data from the extracted time-series numerical data of the target “Company A”. Here, it is assumed that the numerical information “1020” is extracted from the numerical information in time series order included in the time series numerical data of the target “Company A” shown in FIG.

  Here, since the current time-series prediction model update process is the first time as described above, there is no numerical information before the numerical information “1020” extracted by the target numerical change rate evaluation unit 33.

  In addition, since all the numerical information included in the time series numerical data of the target “Company A” has not been processed, the target numerical change rate evaluation unit 33, for example, the numerical value next to the numerical information “1020” in time series order. The information “1000” is extracted from the time series numerical data of the target “Company A”.

  The numerical information “1020” precedes the numerical information “1000” extracted here. Therefore, the target numerical value change rate evaluation unit 33 uses the extracted numerical information “1000”, the numerical information “1020” that is the previous numerical information, and the numerical information “1020” to be assigned to the numerical information “1020”. The target numerical value change rate of the target “Company A” on the date “2010/4” is calculated. In this case, the target numerical value change rate of the target “Company A” on the date “2010/4” is | 1000−1020 | /1020≈0.02 according to Equation (3).

  Here, calculation of the target numerical value change rate of the target “Company A” on the date “2010/4” has been described, but other target numerical value change rates are calculated in the same manner.

  As described above, since this time series prediction model update process is the first time, when the target numerical value change rate for each target in each date within the learning data collection period is calculated in this way, the calculated target numerical value is calculated. The rate of change is aligned with the target word and in chronological order (date order) and stored as it is in the target numerical value change rate evaluation unit 33.

  Here, FIG. 17 to FIG. 20 show an example of the target numerical value change rate in order of time series for each target stored in the target numerical value change rate evaluation unit 33.

  FIG. 17 shows an example of the target numerical value change rate in the time series order of the target “Company A” among a plurality of targets. FIG. 18 shows an example of the target numerical value change rate in the time series order of the target “Company B” among a plurality of targets. FIG. 19 shows an example of the target numerical value change rate in the time series order of the target “Company C” among a plurality of targets. FIG. 20 shows an example of the target numerical value change rate in the time series order of the target “Company D” among a plurality of targets.

  As shown in FIG. 17 to FIG. 20, each date within the learning data collection period “2010/1/4 to 2010/1/8” is included in the target numerical value change rate in the time series order of each target. The assigned target numerical rate of change is included.

  Specifically, for example, the target numerical value change rate in the chronological order of the target “Company A” illustrated in FIG. 17 includes the target numerical value change rate “0.02” to which the date “2010/1/4” is assigned. It is. The date “2010/4” is the date given to the immediately preceding numerical information (numerical information “1020” shown in FIG. 13) used when calculating the target numerical value change rate “0.02”. It is.

  The date “2010/1/8” is within the learning data collection period “2010/1/4 to 2010/1/8”, but the target numerical value to which the date “2010/1/8” is given. Since the change rate is calculated after the numerical information of the date “2010/1/9” is collected, it is blank as shown in FIGS.

  Here, the target numerical change rates of the target “Company A”, “B Company”, “C Company”, and “Company D” are described in time series order. The target numerical value change rates in the time series order of other targets are also stored in the same manner.

  Next, the time-series prediction model generation unit 34 includes a target expression evaluation value (for example, FIG. 9 to FIG. 12) and a target numerical value change rate evaluation unit for each target stored in the target expression evaluation unit 32. A time series prediction model is generated based on the target numerical change rate (for example, FIG. 17 to FIG. 20) in the order of time series for each target stored inside 33. In this case, the time series prediction model generation unit 34 generates a time series prediction model using the above equation (1).

Here, for example, a case where a time series prediction model is generated for the target “Company A” will be specifically described. In this case, the target t in the formula (1) is the target “Company A”. In addition, since the maximum number of data collection times is 5, p is 4 (maximum number of data collection times-1). I = 1 represents the date “2010/4”. Similarly, i = 2 represents the date “2010/1/5”, i = 3 represents the date “2010/1/6”, i = 4 represents the date “2010/1/7”, and i = 5 represents the date “2010/1/8”.

  FIG. 21 shows an example of a time series prediction model generated by the time series prediction model generation unit 34.

  As described above, the time series prediction model includes values calculated for each target with respect to a specific set of parameters (α and β) in Equation (1).

  In the example shown in FIG. 21, the parameter α is 0.00, 0.25, 0.50, 0.75, and 1.00, and the parameter β is 0.25, 0.50, and 0.75. In the case of each value, a time-series prediction model including values calculated for each target by Expression (1) for each set of the parameters α and β is shown.

  As described above, when the time series prediction model is generated by the time series prediction model generation unit 34, the time series prediction model update process is ended. The time series prediction model generated by the time series prediction model generation unit 34 is stored inside the time series prediction model generation unit 34.

  Here, as described above with reference to FIG. 4, when the time-series prediction model update process is ended, and the prediction mode or the end mode is not set according to the user's operation (that is, the learning mode is set). In the case where the time series prediction model is updated, the time series prediction model update process (that is, the second time series prediction model update process) is executed again. In this case, the time series prediction model update process is executed after one data collection unit time has elapsed from the learning data collection period set in the previous time series prediction model update process.

  Hereinafter, after the first time series prediction model update process described above (hereinafter referred to as the previous time series prediction model update process) is executed, the time series prediction model update process (that is, the second time series prediction model update process) is executed again. The case where the model update process is executed will be specifically described. It is assumed that the current date is “2010/1/9”.

  First, the data collection unit 31 sets a learning data collection period. Here, since the current time series prediction model update process is the second time and the current date is “2010/1/9”, immediately after the learning data collection period set in the previous time series prediction model update process. To the current date, that is, “2010/1/9 (˜2010 / 1/9)” is set as the learning data collection period. When the previous time-series prediction model update process is executed, for example, the learning data collection period “2010/1/4 to 2010/1/8” set in the previous time-series prediction model update process (learning It is assumed that the previous data collection date in the mode is set.

  Next, text analysis processing is executed based on the learning data collection period set by the data collection unit 31.

  In the text analysis process, first, text information within the learning data collection period is collected by the data collection unit 31. Here, since the learning data collection period is “2010/1/9”, news headlines for “2010/1/9” are collected as text information. Thereby, the data collection part 31 acquires the time series text data containing the collected text information. Here, the time-series text data acquired by the data collection unit 31 includes only text information assigned with the date “2010/1/9”.

  Hereinafter, the target expression evaluation unit 32 uses the text information included in the time-series text data acquired by the data collection unit 31 and the target expression stored in the target expression storage unit 22 according to the above equation (2). A target expression evaluation value for each target on a date within the learning data collection period (here, date “2010/1/9”) is calculated. Since the calculation of the target expression evaluation value is as described above, detailed description thereof is omitted.

  Here, FIG. 22 to FIG. 25 show an example of the target expression evaluation value for each target on the date “2010/1/9” calculated by the target expression evaluation unit 32.

  FIG. 22 shows an example of the target expression evaluation value of the target “Company A” among a plurality of targets. FIG. 23 shows an example of the target expression evaluation value of the target “Company B” among a plurality of targets. FIG. 24 shows an example of the target expression evaluation value of the target “Company C” among a plurality of targets. FIG. 25 shows an example of a target expression evaluation value of a target “Company D” among a plurality of targets.

  As shown in FIGS. 22 to 25, the target expression evaluation value of each target includes the date (that is, “2010/1/9”) within the learning data collection period “2010/1/9” described above. Has been granted.

  Next, the target expression evaluation value of each target shown in FIGS. 22 to 25 is stored in the target expression evaluation unit 32, thereby updating the target expression evaluation value in time series order of each target. .

  Here, for example, the update of the target expression evaluation value in the time series order of the target “Company A” will be specifically described. In this case, among the target expression evaluation values in the time series order of the target “Company A” shown in FIG. 9 described above, the target expression evaluation value “0.01” assigned the oldest date is the target expression evaluation unit 32. Deleted from inside. Further, the target expression evaluation value “0.40” of the target “Company A” shown in FIG. 22 is stored in the target expression evaluation unit 32. Accordingly, the target expression evaluation value in the time series order of the target “Company A” shown in FIG. 9 is updated to the target expression evaluation value in the time series order of the target “Company A” shown in FIG.

  Similarly, the target expression evaluation value in the time series order of the target “company B” shown in FIG. 10 is updated to the target expression evaluation value in the time series order of the target “company B” shown in FIG. Further, the target expression evaluation value in the time series order of the target “Company C” shown in FIG. 11 is updated to the target expression evaluation value in the time series order of the target “Company C” shown in FIG. The target expression evaluation value in the time series order of the object “Company D” shown in FIG. 12 is updated to the target expression evaluation value in the time series order of the object “D company” shown in FIG.

  Here, the target “Company A”, “Company B”, “Company C”, and “Company D” have been described, but the target expression evaluation values of other targets are similarly updated.

  Next, the target numerical analysis process is executed based on the learning data collection period set by the data collection unit 31.

  In the target numerical analysis process, first, numerical information within the learning data collection period is collected by the data collection unit 31. Here, since the learning data collection period is “2010/1/9”, the data collection unit 31 collects the stock price for each target (company) of “2010/1/9” as numerical information. Thereby, the data collection unit 31 acquires time-series numerical data including the collected numerical information. Here, the time-series numerical data acquired by the data collecting unit 31 includes only numerical information to which the date “2010/1/9” is assigned.

  Here, FIG. 30 to FIG. 33 are examples of time-series numerical data including numerical information (target stock price) within the learning data collection period “2010/1/9” collected for each target by the data collecting unit 31. Show.

  FIG. 30 shows an example of time-series numerical data including numerical information (stock price of Company A) on the date “2010/1/9” collected for the target “Company A” among a plurality of targets. FIG. 31 shows an example of time-series numerical data including numerical information (stock price of B company) on the date “2010/1/9” collected for the target “Company B” among a plurality of targets. FIG. 32 shows an example of time-series numerical data including numerical information (stock price of company C) on the date “2010/1/9” collected for the object “company C” among a plurality of objects. FIG. 33 shows an example of time-series numerical data including numerical information (stock price of company D) on the date “2010/1/9” collected for the object “company D” among a plurality of objects.

  It should be noted that the date “2010/1/9” within the learning data collection period “2010/1/9” is given to the numerical information included in the time-series numerical data shown in FIGS.

  Next, the target numerical change rate evaluation unit 33 uses the numerical information included in the time-series numerical data for each target acquired by the data collection unit 31 and the numerical information immediately before the numerical information (immediate numerical information). The target numerical value change rate for each target is calculated by the above equation (3).

  For example, the target numerical change rate of the target “Company A” is the numerical information “550” of the date “2010/1/9” of the target “Company A” and the numerical information immediately before the numerical information “550” shown in FIG. Here, the numerical value information “750” of the date “2010/1/8” of the target “Company A” shown in FIG.

  Here, the target “Company A” has been described, but the target numerical change rate is similarly calculated for other targets such as the targets “Company B”, “Company C”, and “Company D”.

  Here, FIG. 34 to FIG. 37 show an example of the target numerical value change rate for each target calculated by the target numerical value change rate evaluation unit 33.

  FIG. 34 shows an example of the target numerical value change rate of the target “Company A” among a plurality of targets. FIG. 35 shows an example of the target numerical value change rate of the target “Company B” among a plurality of targets. FIG. 36 shows an example of the target numerical value change rate of the target “Company C” among a plurality of targets. FIG. 37 shows an example of the target numerical value change rate of the target “Company D” among a plurality of targets.

  As shown in FIGS. 34 to 37, the target numerical value change rate of each target includes the date (here, the date “2010” given to the immediately preceding numerical information used for calculation of the target numerical value change rate. / 1/8 ").

  Next, the target numerical change rate of each target shown in FIGS. 34 to 37 is stored in the target numerical change rate evaluation unit 33, so that the target numerical change rate of each target in chronological order is updated. Is done.

  Here, for example, the update of the target numerical value change rate in the time series order of the target “Company A” will be specifically described. In this case, the target numerical value change rate “0.02” to which the oldest date is assigned among the target numerical value change rates of the target “Company A” shown in FIG. 33 is deleted from the inside. Further, the target numerical value change rate “0.27” of the target “Company A” shown in FIG. 34 is stored in the target numerical value change rate evaluation unit 33. The date “2010/1/8” is assigned to the target numerical value change rate “0.27” of the target “Company A”. Accordingly, the target numerical value change rate in the time series order of the target “Company A” shown in FIG. 17 is updated to the target numerical value change rate in the time series order of the target “Company A” shown in FIG. Note that the target numerical value change rate to which the date “2010/1/9” is assigned is calculated after the numerical information of the date “2010/1/10” is collected, and is blank as shown in FIG. Yes.

  Similarly, the target numerical value change rate in the time series order of the target “Company B” shown in FIG. 18 is updated to the target numerical value change rate in the time series order of the target “Company B” shown in FIG. Further, the target numerical value change rate in the time series order of the target “Company C” shown in FIG. 19 is updated to the target numerical value change rate in the time series order of the target “Company C” shown in FIG. The target numerical value change rate in the time series order of the target “Company D” shown in FIG. 20 is updated to the target numerical value change rate in the time series order of the target “Company D” shown in FIG.

  Here, the target “Company A”, “Company B”, “Company C”, and “Company D” have been described, but the target numerical value change rates of other targets are similarly updated.

  Next, the time-series prediction model generation unit 34 includes a target expression evaluation value (for example, FIGS. 26 to 29) and a target numerical value change rate evaluation unit for each target stored in the target expression evaluation unit 32. A time series prediction model is generated by the above-described equation (1) using the target numerical value change rate (for example, FIG. 38 to FIG. 41) in the time series order for each target stored in the inside 33. Note that the time series prediction model generation processing is as described above, and thus detailed description thereof is omitted.

  The time series prediction model generated by the time series prediction model generation unit 34 is stored (that is, overwritten) inside the time series prediction model generation unit 34. Thereby, in the time series prediction model update process, the time series prediction model is updated.

  Next, a processing procedure of the above-described target ranking process (the process of step S6 shown in FIG. 4) will be described with reference to the flowchart of FIG. The target ranking process is a process executed when the prediction mode is set as described above.

  First, the data collection unit 31 determines whether or not the previous data collection date (the learning data collection period set in the time series prediction model update process) is set in the learning mode, that is, the time series prediction model update process is executed. Whether or not a time series prediction model has been generated is determined (step S41).

  When it is determined that the previous data collection date in the learning mode is set (YES in step S41), the data collection unit 31 sets a predicted data collection period (step S42). In this case, the data collection unit 31 sets the period from immediately after the previous data collection date in the learning mode (that is, after the learning data collection period set in the time series prediction model update process) to the current date as the prediction data collection period. Set. Since the target ranking process is executed during the time series prediction model update process executed when the learning mode is set, the prediction data collection period set by the data collection unit 31 is the target prediction Although it is shorter than the data collection unit time set in the initialization process of the apparatus 30, the period is assumed to be one unit.

  Next, text analysis processing is executed based on the learning data collection period set by the data collection unit 31 (step S43). The text analysis process is the same as the process in step S12 shown in FIG. 5 (the process shown in FIG. 6) when the learning data collection period is the predicted data collection period, and detailed description thereof is omitted. . That is, according to the text analysis processing in step S43, the target expression evaluation value (target expression evaluation value for one data collection unit time) for each target on the date within the predicted data collection period is calculated.

  The target attention level calculation unit 35 includes a target expression evaluation value for each target calculated by the text analysis processing in step S43 and a time series prediction model stored in the time series prediction model generation unit 34 (the time series prediction described above). Based on the time-series prediction model generated in the model update process), the degree of attention for each target is calculated.

  In other words, the target attention level calculation unit 35 calculates the attention level in the next period of the target by applying the time series prediction model to the target expression evaluation value for each target calculated by the text analysis processing in step S43. .

In this case, the target attention level calculation unit 35 calculates the attention level for each target using, for example, a formula (4) prepared in advance.

  Here, the time series prediction model shown in FIG. 21 described above is stored in the time series prediction model generation unit 34, and the object “company A” shown in FIG. Assume that the expression evaluation value “0.40” is calculated. In this case, if 0.25 is specified as the parameter α (attenuation rate) and 0.50 is specified as the parameter β (expression importance level) by the user, the attention level of the target “Company A” is 0.461191 ( = 0.40 + 0.25 * 0.244763) is calculated. Note that the value “0.244763” used for calculating the degree of attention of the target “Company A” is α = 0.25 and β = 0.50 included in the time series prediction model shown in FIG. Is a value calculated by the above-described equation (1).

  Here, the attention level of the target “Company A” has been described, but the attention levels are similarly calculated for the other targets “Company B”, “C Company”, “D Company”, and the like. Note that the attention level calculated for each target by the target attention level calculation unit 35 indicates that the larger the value, the higher the target will be in the next term.

  Here, FIG. 43 shows an example of the respective attention degrees of the targets “Company A”, “Company B”, “Company C”, and “Company D” calculated by the target attention degree calculation unit 35.

  In the example shown in FIG. 43, the degree of attention of the target “Company A” is 0.461191. The degree of attention “0.461191” of the target “Company A” is a case where α = 0.25 and β = 0.50 are designated by the user as described above, and are shown in FIG. The degree of attention calculated when the time series prediction model shown in FIG. 21 is applied to the target expression evaluation value “0.40” of the target “Company A”.

  The degree of attention of the target “Company B” is 0.005530. The degree of attention “0.005530” of the target “Company B” is a case where α = 0.25 and β = 0.50 are designated by the user as described above, and are shown in FIG. 23 described above. The degree of attention calculated when the time series prediction model shown in FIG. 21 is applied to the target expression evaluation value “0.00” of the target “Company B”.

  The degree of attention of the target “Company C” is 0.171399. The degree of attention “0.1711399” of the object “Company C” is a case where α = 0.25 and β = 0.50 are designated by the user as described above, and are shown in FIG. 24 described above. This is the degree of attention calculated when the time series prediction model shown in FIG. 21 is applied to the target expression evaluation value “0.15” of the target “Company C”.

  Further, the degree of attention of the target “Company D” is 0.0229296. The degree of attention “0.022962” of the target “Company D” is a case where α = 0.25 and β = 0.50 are designated by the user as described above, and are shown in FIG. 25 described above. The degree of attention calculated when the time series prediction model shown in FIG. 21 is applied to the target expression evaluation value “0.02” of the target “Company D”.

  In FIG. 43, the attention levels of the objects other than the objects “Company A”, “Company B”, “Company C”, and “Company D” are omitted.

  Returning to FIG. 42 again, the target ranking display unit 36 rearranges the targets in order of the attention level based on the attention level (size) calculated for each target by the target attention level calculation unit 35 (step S45). ).

  The target ranking display unit 36 presents (displays) to the user the result of rearranging the targets in order of attention (that is, the ranking result of the attention level for each target) (step S46).

  Here, FIG. 44 shows an example of the ranking result of the attention level for each target displayed by the target ranking display unit 36. In FIG. 44, objects other than the targets “Company A”, “Company B”, “Company C”, and “Company D” are omitted.

  As shown in FIG. 44, in the ranking result of the attention level for each target, the targets are displayed in the ranking format in descending order of the attention level shown in FIG.

  In the example shown in FIG. 44, the objects “A company”, “C company”, “D company”, and “B company” are displayed together with the degree of attention in this order.

  In FIG. 44, it has been described that all of the targets “Company A”, “Company B”, “Company C”, and “Company D” are displayed. For example, the top n (n is 1 or more) Only a target of (integer) may be displayed.

  Here, as shown in FIG. 44, when the ranking result of the attention level for each target is displayed, the user gives an instruction to update (change) the parameters (α and β) used to calculate the attention level. be able to. In this case, the user can specify the updated parameter value.

  Next, it is determined whether or not the parameter has been updated in accordance with such a user instruction (step S47).

  If it is determined that the parameter has been updated (YES in step S47), the process returns to step S44 and the process is repeated. In this case, the target attention level calculation unit 35 calculates the attention level for each target again based on the value of the parameter specified by the user. After the attention level for each target is calculated again in this way, the processes of steps S45 and S46 described above are executed.

  Here, FIG. 45 shows an example of the ranking result of the attention level for each target displayed by the target ranking display unit 36 after the parameter is updated. Note that the degree of attention for each target shown in FIG. 45 is the degree of attention calculated when the parameter α is updated from 0.25 to 1.00 and the parameter β is updated from 0.50 to 0.25. .

  In this way, by updating the parameters according to the user's instructions, the ranking result of the attention level for each target in the case of a set of various values of the attenuation rate (α) and the expression importance level (β) is used. Can be presented to the person.

  As described above, in the present embodiment, when the learning mode is set according to the user's operation, the text information is collected within the learning data collection period (predetermined period), and the collected information is collected. Time-series text data including text information in time-series order is acquired and included in the target expression (character string related to the target) stored in the target expression storage unit 22 for each target and the acquired time-series text data. Based on the text information in time series order, calculate the target expression evaluation value in time series order for each target, collect numerical information in the learning data collection period (predetermined period) for each target, Based on the numerical values included in the numerical information before and after in the time series order included in the time series numerical data included in the time series numerical data acquired for each target To calculate the target numerical value rate of change in chronological order for each elephant. In the present embodiment, when the prediction mode is set according to the user's operation, text information is collected within the prediction data collection period (the period from the learning data collection period to the present), and the target A target expression evaluation value is calculated for each target based on the target expression stored in the target expression storage unit 22 and the collected text information for each target, and the target expressions in time series order calculated for each target Using the evaluation value, the target numerical rate of change in time series calculated for each target, and the target expression evaluation value calculated for each target, the degree of attention for each target is calculated, Present attention to the user.

  As a result, in the present embodiment, a target to be noted can be predicted from a plurality of targets, and the target to be noticed can be presented to the user. It is possible to easily and easily select the target to be performed.

  In other words, in the present embodiment, not only the change of numerical information corresponding to the target up to the present time but also the appearance frequency of the target expression (character string related to the target) that affects the change of the numerical information (described above) By referring to the target score), for example, it is possible to support selection of a target to be noticed based on information from a plurality of information sources without assuming knowledge by an expert or the like.

  Specifically, in this embodiment, for example, by using news headlines as text information and company stock prices as numerical information, the user should pay attention from a large number of companies (brands) in the securities market. Brands can be easily and conveniently selected. The present embodiment can also be applied to the field of plant monitoring, for example. Specifically, by using daily information recorded by maintenance personnel in the field of plant monitoring as text information and measurement data collected from the plant as numerical information, users should pay attention from a large number of plants. Plants (for example, plants with a possibility of failure, etc.) can be selected.

  Further, in the present embodiment, when the learning mode is set, the time series prediction is performed based on the target expression evaluation value in the time series order calculated for each target and the rate of change in the time series order calculated for each target. By generating the model, it is possible to reduce the processing amount necessary for calculating the attention level for each target when the prediction mode is set.

  In the present embodiment, it has been described that only one type of numerical information (for example, a stock price) exists for each target. However, there is a plurality of numerical information for each target, and the plurality of numerical information. A time series prediction model may be generated from a target numerical change rate in a time series order for each of a plurality of objects calculated from the above. That is, the time series prediction model may be generated by adding, for example, numerical information of the volume of the stock price to the numerical information of the stock price of the target (company) in the present embodiment.

  In the present embodiment, the data collection unit 31 has been described as collecting each data from a website that provides text information and a website that provides numerical information. However, for example, other information connected in a network environment Data may be collected from the system or the like.

  Further, the present invention is not limited to the above-described embodiments as they are, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Moreover, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

  DESCRIPTION OF SYMBOLS 10 ... Computer, 20 ... External storage device, 22 ... Object expression storage part, 30 ... Attention object prediction apparatus, 31 ... Data collection part, 32 ... Object expression evaluation part, 33 ... Object numerical value change rate evaluation part, 34 ... Time series Prediction model generation unit, 35 ... target attention degree calculation unit, 36 ... target ranking display unit.

Claims (4)

In a target object prediction device that predicts a target object of interest from a plurality of targets,
For each target, target expression storage means for storing a character string related to the target;
First data collection means for collecting text information including character strings within a plurality of predetermined periods, and acquiring time-series text data including the collected text information in time-series order;
Based on the character string related to the target stored in the target expression storage unit for each target and the time-series text information included in the time-series text data acquired by the first data collection unit, First evaluation value calculating means for calculating evaluation values in the time series order for each of the objects;
Second data collection means for collecting, for each target, numerical information including numerical values related to the target in the plurality of predetermined periods, and acquiring time-series numerical data including the numerical information in time-series order;
Based on the numerical values included in the numerical information before and after the time-series order included in the time-series numerical data acquired by the second data collection unit for each target, the rate of change in the time-series order for each target A change rate calculating means for calculating;
Third data collection means for collecting text information including character strings after the plurality of predetermined periods;
An evaluation value for each target based on a character string related to the target stored in the target expression storage unit for each target and a character string included in the text information collected by the third data collection unit Second evaluation value calculating means for calculating
The evaluation value in the time series order calculated by the first evaluation value calculating unit for each target, the change rate in the time series order calculated by the change rate calculating unit for each target, and the target value for each target Attention level calculation means for calculating the attention level for each target using the evaluation value calculated by the second evaluation value calculation means;
An attention target prediction apparatus comprising: presentation means for presenting the attention level for each target calculated by the attention level calculation means. The first evaluation value calculation means is stored in the object expression storage means for each object in the text information in time series order included in the time series text data acquired by the first data collection means. It is determined whether a character string related to the target is included, and the evaluation value in the time series order of the target is calculated according to the number of the text information including the character string related to the target,
Whether the second evaluation value calculation means includes text information related to the object stored in the object expression storage means for each object in the text information collected by the third data collection means The attention object prediction apparatus according to claim 1, wherein the evaluation value of the target is calculated according to the number of the text information including the character string related to the target. Based on the evaluation value in the time series order calculated by the first evaluation value calculation unit for each target and the change rate in the time series order calculated by the change rate calculation unit for each target Further comprising a time-series prediction model generating means for generating a time-series prediction model used for calculating the degree of attention for each;
The attention level calculation means uses the evaluation value calculated by the second evaluation value calculation means for each target and the time series prediction model stored in the time series prediction model storage means, and receives attention for each target. The attention object prediction device according to claim 1 which calculates a degree. The degree-of-attention calculation means uses a plurality of predetermined parameters to calculate the degree of attention for each target for each parameter,
The attention object prediction apparatus according to claim 1, wherein the presenting means presents a degree of attention for each object calculated using a parameter designated by a user.

Images