Information processing method and information processing program

Abstract

To make it easy to learn a discrete choice model capable of accurately estimating which option is chosen by a choosing subject from among a plurality of options.SOLUTION: An information processing apparatus sets an expected value of a probability that a choosing subject that has chosen an option J in the past belongs to a class 501 such that the choosing subject definitely belongs to class 501. The information processing apparatus sets an expected value of a probability that a choosing subject that has not chosen the option J in the past belongs to the class 501 such that the choosing subject probabilistically belongs to class 501 and class 502 and that the probability that the choosing subject belongs to the class 501 decreases as the effect of the option J increases. The information processing apparatus estimates parameters of a choice model 511 and a choice model 512 based on the record representing an option chosen by each choosing subject in the past and each set expected value.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present invention relates to an information processing method and an information processing program. [Background technology]

[0002] Conventionally, there exists a discrete choice model that estimates which option a selector will select from among multiple options. Specifically, the discrete choice model has a function of estimating which option a selector will select by using a formula that represents the effect of each option. There is a technology that trains the discrete choice model based on multiple records that represent options that different selectors have selected from among multiple options in the past.

[0003] Prior art, for example, classifies images that have positive comments from a collection of images that are categorized into categories that match or are close to a theme desired by a user, into images that have positive comments and images that do not have positive comments. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2019-114243 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the conventional technology, it is difficult to learn a discrete choice model capable of accurately estimating which option a selector will select from among multiple options. Specifically, there may be a case where a first selector who recognizes one of the options and a second selector who does not recognize any of the options are mixed among multiple selectors. In this case, if a first record representing an option selected in the past by the first selector and a second record representing an option selected in the past by the second selector are treated in the same way, it becomes difficult to properly learn a discrete choice model.

[0006] In one aspect, the present invention aims to facilitate learning of a discrete choice model that can accurately estimate which option a selector will select from among a plurality of options. [Means for solving the problem]

[0007] According to one embodiment, an information processing method and an information processing program are proposed that acquire a plurality of records each representing an option previously selected from a plurality of options, set a first value representing a definite membership as an expected value of the probability that the selecting subject belongs to a first class that recognizes the first option for each of the acquired plurality of records, and set a second value representing a probabilistic membership as the expected value for each of the remaining records among the acquired plurality of records that are different from the one or more records based on a standard value of the probability that the selecting subject belongs to the first class and a formula representing the effect of each of the multiple options, and learn the standard value and the parameters included in the formula based on each of the acquired plurality of records and the expected value set for each of the acquired plurality of records. Effect of the Invention

[0008] According to one aspect, it is possible to facilitate learning of a discrete choice model that can accurately estimate which option a selector will select from among a plurality of options. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram illustrating an example of an information processing method according to an embodiment. [Diagram 2] FIG. 2 is an explanatory diagram illustrating an example of an information processing system 200. As shown in FIG. [Diagram 3] FIG. 3 is a block diagram showing an example of a hardware configuration of the information processing device 100. As shown in FIG. [Figure 4] FIG. 4 is a block diagram showing an example of a functional configuration of the information processing device 100. As shown in FIG. [Diagram 5] FIG. 5 is an explanatory diagram showing the flow of operations of the information processing device 100. [Figure 6] FIG. 6 is an explanatory diagram (part 1) showing a first operation example of the information processing device 100. [Figure 7] FIG. 7 is an explanatory diagram (part 2) showing a first operation example of the information processing device 100. In FIG. [Figure 8] FIG. 8 is an explanatory diagram (part 3) showing a first operation example of the information processing device 100. [Figure 9] FIG. 9 is an explanatory diagram (part 4) showing a first operation example of the information processing device 100. [Figure 10] FIG. 10 is an explanatory diagram (part 5) showing a first operation example of the information processing device 100. [Figure 11] FIG. 11 is an explanatory diagram (part 6) showing a first operation example of the information processing device 100. [Figure 12] FIG. 12 is an explanatory diagram (part 7) showing a first operation example of the information processing device 100. [Figure 13] FIG. 13 is an explanatory diagram (part 8) showing a first operation example of the information processing device 100. [Figure 14] FIG. 14 is a flowchart illustrating an example of a procedure of the first learning process. [Figure 15] FIG. 15 is a flowchart illustrating an example of a procedure of the first prediction process. [Figure 16] FIG. 16 is an explanatory diagram (part 1) showing a second operation example of the information processing device 100. [Figure 17] FIG. 17 is an explanatory diagram (part 2) showing a second operation example of the information processing device 100. In FIG. [Figure 18] FIG. 18 is an explanatory diagram (part 3) showing a second operation example of the information processing device 100. [Figure 19] FIG. 19 is an explanatory diagram (part 4) showing a second operation example of the information processing device 100. [Figure 20]FIG. 20 is an explanatory diagram (part 5) showing a second operation example of the information processing device 100. [Figure 21] FIG. 21 is an explanatory diagram (part 6) showing a second operation example of the information processing device 100. [Figure 22] FIG. 22 is a flowchart illustrating an example of a procedure of the second prediction process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment of an information processing method and an information processing program according to the present invention will be described in detail with reference to the drawings.

[0011] (An example of an information processing method according to an embodiment) 1 is an explanatory diagram showing an example of an information processing method according to an embodiment. An information processing device 100 is a computer for learning a discrete choice model. The information processing device 100 is, for example, a server or a PC (Personal Computer).

[0012] The discrete choice model is a model that estimates the probability that a selector will select each of a plurality of options. The discrete choice model has a function of estimating the probability that a selector will select each option by using a formula that represents the effect of each option in response to input of feature quantities related to each option. Specifically, the discrete choice model is a multinomial logit model, a nested logit model, a mixed logit model, or the like.

[0013] For example, a discrete choice model may be trained based on a plurality of records representing options selected from a plurality of options by different selection agents in the past. The records may include, for example, feature quantities related to each option selected by the selection agents in the past from a plurality of options. Here, it is desirable to train a discrete choice model capable of accurately estimating which option the selection agent will select from a plurality of options.

[0014] However, there are cases where it is difficult to learn a discrete choice model capable of accurately estimating which option a selector will select from among multiple options. For example, there may be a case where a first selector who recognizes one of the options and a second selector who does not recognize any of the options are mixed among multiple selectors. In this case, if a first record representing an option selected in the past by the first selector and a second record representing an option selected in the past by the second selector are treated in the same way, it becomes difficult to properly learn a discrete choice model.

[0015] For example, a record showing a second option previously selected by a first selection subject who is aware of the first option may be treated as a record showing that the effect of the first option is relatively smaller than the effect of the second option for the first selection subject. On the other hand, it is not preferable for a record showing a second option previously selected by a second selection subject who is not aware of the first option to be treated as a record showing that the effect of the first option is relatively smaller than the effect of the second option for the second selection subject.

[0016] For example, in the past, on the assumption that all selection subjects recognize all options, the selection subjects tend to select which option when what feature values ​​appear for each option is reflected in the learning discrete choice model. Therefore, it is not possible to reflect in the learning discrete choice model that the selection subjects were unable to select any of the options because they did not recognize the options, regardless of the feature values ​​related to the options. Based on a record representing a second option selected in the past by a selection subject who does not recognize the first option, the discrete choice model may reflect an erroneous determination that the effect of the first option is relatively smaller than the effect of the second option for the selection subject. For this reason, it is difficult to learn a discrete choice model that can accurately estimate which option the selection subject will select from multiple options. For example, the learned discrete choice model may erroneously estimate the effect of any option to be smaller than the actual effect.

[0017] Specifically, each of the multiple options may be an option representing a means of transportation. The multiple options may be, for example, walking, bicycle, car, and electric kick scooter. For example, the bicycle may be available for rental. For example, the car may be a taxi. For example, the electric kick scooter may be available for rental.

[0018] Here, in a case where a moving person, who is the selection subject, is trying to select a means of transportation, it is possible to train a discrete choice model on the premise that all moving persons recognize walking, bicycles, cars, and electric kick scooters as available means of transportation. However, it is possible that one of the selection subjects does not recognize the electric kick scooter as an available means of transportation. In this case, a discrete choice model in an incorrect state that underestimates the effect of the electric kick scooter may be trained.

[0019] Therefore, in this embodiment, an information processing method that can facilitate learning of a discrete choice model that can accurately estimate which option a selector will select from among a plurality of options will be described.

[0020] In FIG. 1, there are a plurality of options 110. The options 110 represent, for example, means of transportation. Assume that each of the plurality of selection subjects selected one of the plurality of options 110 in the past. The selection subject is, for example, a traveler. Assume that the plurality of options 110 includes, for example, a first option 111.

[0021] Here, it is assumed that there are multiple records 120 each representing an option 110 previously selected from among the multiple options 110. The multiple records 120 may include, for example, two or more records 120 representing options 110 previously selected by the same selection subject. The multiple records 120 may include, for example, two or more records 120 representing options 110 previously selected by different selection subjects.

[0022] The information processing device 100 has, for example, a reference value of the probability that the selecting subject belongs to the first class 130 that has recognized the first option 111. The information processing device 100 has, for example, a mathematical formula that expresses the effect of each option 110 of the multiple options 110. The mathematical formula, for example, is an element forming a model that estimates the probability that the selecting subject will select each option of the multiple options. Specifically, the mathematical formula is an element forming a model that estimates the probability that the selecting subject will select each option in response to input of feature amounts related to each option.

[0023] (1-1) The information processing device 100 acquires a plurality of records 120. For example, the information processing device 100 acquires the plurality of records 120 by receiving them from another computer. For example, the information processing device 100 may acquire the plurality of records 120 by accepting input of the plurality of records 120 based on an operational input by a user.

[0024] (1-2) The information processing device 100 identifies, from among the acquired multiple records 120, one or more records 121 corresponding to a selecting subject who has recognized the first option 111. For example, from among the multiple records 120, the information processing device 100 identifies one or more records 121 each representing an option selected by a selecting subject who has previously selected the first option 111. For each record 121 of the identified one or more records 121, the information processing device 100 sets a first value representing definite membership as an expected value of the probability that the selecting subject belongs to the first class 130.

[0025] (1-3) The information processing device 100 identifies remaining records 122 that are different from one or more records 121 from among the acquired multiple records 120. The information processing device 100 sets a second value representing probabilistic belonging as an expected value for each of the identified remaining records 122 based on a reference value and a formula.

[0026] This allows the information processing device 100 to handle the records 121 and 122 separately depending on whether the selection subject has recognized the first option 111. With regard to the record 122, the information processing device 100 can take into consideration both the case where the selection subject has recognized the first option 111 and the case where the selection subject has not recognized the first option 111.

[0027] (1-4) The information processing device 100 learns a reference value and parameters included in a formula based on each of the acquired records 120 and an expected value. The information processing device 100 learns a model that estimates the probability of selecting each option of a plurality of options based on the parameters of the set formula. The model is, for example, a discrete choice model.

[0028] Thereby, the information processing device 100 can set the reference value and the parameters included in the formula to appropriate values ​​taking into consideration the case where the selecting subject has recognized the first option 111 and the case where the selecting subject has not recognized the first option 111. The information processing device 100 can appropriately set the formula that is an element forming a model that estimates the probability that the selecting subject will select each option of multiple options based on the parameters of the set formula. Therefore, the information processing device 100 can easily learn a model that accurately estimates the probability that the selecting subject will select each option of multiple options. The information processing device 100 can easily learn a model that is in line with the actual situation or circumstances of the selecting subject who performs the selection act, for example.

[0029] Here, the case where the information processing device 100 learns the model has been described, but the present invention is not limited to this. For example, another computer may learn the model. In this case, the information processing device 100 transmits the parameters of the set formula to the other computer.

[0030] Here, the case where the information processing device 100 is realized by a single computer has been described, but is not limited to this. For example, the functions of the information processing device 100 may be realized by a plurality of computers. Specifically, the functions of the information processing device 100 may be realized on a cloud.

[0031] (An example of the information processing system 200) Next, an example of an information processing system 200 to which the information processing device 100 shown in FIG. 1 is applied will be described with reference to FIG.

[0032] 2 is an explanatory diagram showing an example of an information processing system 200. In FIG. 2, the information processing system 200 includes an information processing device 100 and a client device 201.

[0033] In the information processing system 200, the information processing device 100 and the client device 201 are connected via a wired or wireless network 210. The network 210 is, for example, a local area network (LAN), a wide area network (WAN), or the Internet.

[0034] The information processing device 100 is a computer for facilitating learning of a model that estimates the probability that a selection subject will select each of a plurality of options. The model is, for example, a discrete choice model. The selection subject is, for example, a traveler. The options are, for example, means of transportation.

[0035] The model includes, for example, a formula expressing the effect of each of a plurality of options. If the option is a means of transportation, the effect corresponds to the cost of the means of transportation. The cost is a financial, time, or physical cost. The formula has a function of calculating an index value expressing the magnitude of the effect of each of the plurality of options based on, for example, a feature amount related to each option. For example, the index value is larger as the cost is smaller.

[0036] The information processing device 100 has a reference value of the probability that a first option, among a plurality of options, belongs to a first recognized class. The reference value is, for example, a priori probability. The information processing device 100 has a mathematical formula expressing the effect of each option of the plurality of options. The information processing device 100 has a plurality of records each expressing an option previously selected by a selection subject, among the plurality of options.

[0037] The information processing device 100 sets a first value representing definite belonging to the first class as an expected value of the probability that the selecting subject belongs to the first class for each record of one or more records corresponding to the selecting subject who has recognized the first option among the multiple records. The information processing device 100 sets a second value representing probabilistic belonging to the first class as an expected value of the probability that the selecting subject belongs to the first class for each record of the remaining records based on the reference value and the formula. The information processing device 100 learns the reference value and parameters included in the formula based on each record of the multiple records and the expected value. The information processing device 100 learns a model that estimates the probability of selecting each option of the multiple options based on the parameters of the set formula.

[0038] The information processing device 100 receives a processing request including a feature amount related to each option from the client device 201. The information processing device 100 estimates the probability that the selection subject will select each option by using a learned model based on the feature amount related to each option included in the processing request. The information processing device 100 generates notification information based on the estimated probability of selecting each option. The notification information is, for example, information indicating a plurality of options ordered according to the estimated probability of selecting each option. The information processing device 100 transmits the generated notification information to the client device 201. The information processing device 100 is, for example, a server or a PC.

[0039] The client device 201 is a computer used by the selection subject. The client device 201 acquires feature quantities related to each option based on an operation input by the selection subject. The client device 201 generates a processing request including feature quantities related to each option, and transmits the request to the information processing device 100. The client device 201 receives notification information from the information processing device 100. The client device 201 outputs the received notification information so that the selection subject can refer to it. The client device 201 is, for example, a PC, a tablet terminal, or a smartphone.

[0040] Here, the case where the information processing device 100 is a computer different from the client device 201 has been described, but this is not limited thereto. For example, the information processing device 100 may have a function as the client device 201 and may also operate as the client device 201.

[0041] (Example of hardware configuration of information processing device 100) Next, an example of the hardware configuration of the information processing device 100 will be described with reference to FIG.

[0042] Fig. 3 is a block diagram showing an example of a hardware configuration of the information processing device 100. In Fig. 3, the information processing device 100 has a CPU (Central Processing Unit) 301, a memory 302, a network I / F (Interface) 303, a recording medium I / F 304, and a recording medium 305. In addition, each component is connected to each other via a bus 300.

[0043] Here, the CPU 301 is responsible for the overall control of the information processing device 100. The memory 302 has, for example, a read only memory (ROM), a random access memory (RAM), and a flash ROM. Specifically, for example, the flash ROM and the ROM store various programs, and the RAM is used as a work area for the CPU 301. The programs stored in the memory 302 are loaded into the CPU 301, causing the CPU 301 to execute the coded processes.

[0044] The network I / F 303 is connected to the network 210 through a communication line, and is connected to other computers via the network 210. The network I / F 303 manages an internal interface with the network 210, and controls input and output of data from other computers. The network I / F 303 is, for example, a modem or a LAN adapter.

[0045] The recording medium I / F 304 controls reading / writing data from / to the recording medium 305 under the control of the CPU 301. The recording medium I / F 304 is, for example, a disk drive, a solid state drive (SSD), a universal serial bus (USB) port, etc. The recording medium 305 is a non-volatile memory that stores data written under the control of the recording medium I / F 304. The recording medium 305 is, for example, a disk, a semiconductor memory, a USB memory, etc. The recording medium 305 may be detachable from the information processing device 100.

[0046] In addition to the above-mentioned components, the information processing device 100 may have, for example, a keyboard, a mouse, a display, a printer, a scanner, a microphone, a speaker, etc. Furthermore, the information processing device 100 may have a plurality of recording medium I / Fs 304 and recording media 305. Furthermore, the information processing device 100 may not have the recording medium I / Fs 304 and the recording media 305.

[0047] (Example of Hardware Configuration of Client Device 201) Specifically, an example of the hardware configuration of the client device 201 is similar to the example of the hardware configuration of the information processing device 100 shown in FIG. 3, and therefore a description thereof will be omitted.

[0048] (An example of a functional configuration of the information processing device 100) Next, an example of a functional configuration of the information processing device 100 will be described with reference to FIG.

[0049] 4 is a block diagram showing an example of a functional configuration of the information processing device 100. The information processing device 100 includes a storage unit 400, an acquisition unit 401, a first setting unit 402, a second setting unit 403, a learning unit 404, a prediction unit 405, and an output unit 406.

[0050] The storage unit 400 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 shown in Fig. 3. In the following, a case where the storage unit 400 is included in the information processing device 100 will be described, but this is not limiting. For example, the storage unit 400 may be included in a device different from the information processing device 100, and the stored contents of the storage unit 400 may be accessible from the information processing device 100.

[0051] The acquiring unit 401 to the output unit 406 function as an example of a control unit. Specifically, the acquiring unit 401 to the output unit 406 realize their functions by, for example, causing the CPU 301 to execute a program stored in a storage area such as the memory 302 or the recording medium 305 shown in Fig. 3, or by the network I / F 303. The processing results of each functional unit are stored in, for example, a storage area such as the memory 302 or the recording medium 305 shown in Fig. 3.

[0052] The storage unit 400 stores various information that is referred to or updated in the processing of each functional unit. The storage unit 400 stores, for example, a plurality of records. The record is, for example, information representing an option previously selected by a selection subject from among a plurality of options. The option represents, for example, a means of transportation. The plurality of options includes, for example, a first option.

[0053] The record may include, for example, an attribute value of the selection subject. The attribute value is, for example, an identifier that identifies the selection subject. The attribute value may be, for example, the age, sex, or affiliation of the selection subject. The attribute value may be, for example, the location of the selection subject, or the current time. The record is acquired, for example, by the acquisition unit 401. The record may be, for example, set by the user in advance.

[0054] The storage unit 400 stores, for example, a reference value of the probability that a selection subject belongs to a first class. The first class represents, for example, that the first option has been recognized. The first class is a class for classifying selection subjects who have recognized the first option. The reference value is acquired, for example, by the acquisition unit 401. The reference value may be set in advance by a user, for example.

[0055] The storage unit 400 stores, for example, a mathematical formula expressing the effect of each of the multiple options. The mathematical formula is, for example, an element forming a model that estimates the probability that a selection subject will select each of the multiple options. The model is, for example, a discrete choice model. The mathematical formula has a function of calculating an index value indicating the effect of each of the multiple options. More specifically, the mathematical formula has a function of calculating an index value indicating the effect of each of the multiple options based on a feature amount related to each option. More specifically, the mathematical formula may have a function of calculating an index value indicating the effect of each of the multiple options based on a feature amount related to each option and an attribute value of the selection subject.

[0056] The storage unit 400 specifically stores a first formula expressing the effect of each of the multiple options when the selecting subject belongs to the first class. The storage unit 400 specifically stores a second formula expressing the effect of each of the remaining options, which are different from the first option, among the multiple options when the selecting subject does not belong to the first class. The formula is acquired by the acquisition unit 401, for example. The formula may be set in advance by the user, for example.

[0057] The storage unit 400 stores, for example, feature amounts related to each option. The feature amounts are acquired, for example, by the acquisition unit 401. The feature amounts may be set in advance by the user, for example.

[0058] The storage unit 400 stores, for example, an attribute value of the selection subject. The attribute value is acquired, for example, by the acquisition unit 401. The attribute value may be set in advance by a user, for example. The attribute value may be included in a record, for example.

[0059] The storage unit 400 stores, for example, a model that estimates the probability that a subject of selection will select each option from among a plurality of options. The model is generated by the learning unit 404, for example.

[0060] The acquisition unit 401 acquires various information used for processing by each functional unit. The acquisition unit 401 stores the acquired various information in the storage unit 400 or outputs it to each functional unit. The acquisition unit 401 may also output the various information stored in the storage unit 400 to each functional unit. The acquisition unit 401 acquires various information based on, for example, an operation input by a user. The acquisition unit 401 may receive various information from, for example, a device different from the information processing device 100.

[0061] The acquiring unit 401 acquires, for example, a plurality of records. Specifically, the acquiring unit 401 acquires the plurality of records by receiving them from another computer. Specifically, the acquiring unit 401 may acquire the plurality of records by accepting input of the plurality of records based on an operational input by a user.

[0062] The acquiring unit 401 acquires, for example, a reference value of the probability that the selecting subject belongs to the first class. Specifically, the acquiring unit 401 acquires the reference value by receiving it from another computer. Specifically, the acquiring unit 401 may acquire the reference value by accepting an input of the reference value based on an operation input by a user.

[0063] The acquiring unit 401 acquires, for example, a formula expressing the effect of each of a plurality of options. Specifically, the acquiring unit 401 acquires the formula by receiving it from another computer. Specifically, the acquiring unit 401 may acquire the formula by accepting input of the formula based on an operation input by a user.

[0064] The acquiring unit 401 stores, for example, feature amounts related to each option. Specifically, the acquiring unit 401 acquires the feature amounts by receiving them from another computer. Specifically, the acquiring unit 401 may acquire the feature amounts by accepting input of the feature amounts based on an operation input by a user.

[0065] The acquiring unit 401 stores, for example, an attribute value of a selection subject. Specifically, the acquiring unit 401 acquires the attribute value by receiving it from another computer. Specifically, the acquiring unit 401 may acquire the attribute value by accepting an input of the attribute value based on an operation input by a user.

[0066] The acquisition unit 401 acquires a learning request that requests learning of a model. Specifically, the acquisition unit 401 acquires the learning request by receiving it from another computer. Specifically, the acquisition unit 401 may acquire the learning request by accepting input of the learning request based on an operation input by a user.

[0067] The acquisition unit 401 acquires a prediction request requesting the use of a model. The prediction request, for example, requests estimating the probability that a selecting subject will select each option. The prediction request may include, for example, an attribute value of the selecting subject. The prediction request may include a feature amount related to each option. Specifically, the acquisition unit 401 acquires the prediction request by receiving it from another computer. Specifically, the acquisition unit 401 may acquire the prediction request by accepting an input of the prediction request based on an operation input by a user.

[0068] The acquiring unit 401 may acquire the attribute value of the selecting subject by, for example, extracting the attribute value of the selecting subject included in the prediction request. The acquiring unit 401 may acquire the feature amount of each option by, for example, extracting the feature amount of each option included in the prediction request.

[0069] The acquisition unit 401 may receive a start trigger for starting the processing of any of the functional units. The start trigger may be, for example, a predetermined operation input by a user. The start trigger may be, for example, a receipt of predetermined information from another computer. The start trigger may be, for example, a case where any of the functional units outputs predetermined information. Specifically, the acquisition unit 401 may receive the acquisition of a learning request as a start trigger for starting the processing of the first setting unit 402, the second setting unit 403, and the learning unit 404. Specifically, the acquisition unit 401 may receive the acquisition of a prediction request as a start trigger for starting the processing of the prediction unit 405.

[0070] The first setting unit 402 specifies one or more records corresponding to a selection subject who has recognized the first option, from among the multiple records acquired by the acquisition unit 401. The one or more records are, for example, a collection of records representing the first option. The one or more records may include, for example, records representing other options different from the first option selected by each selection subject who has selected the first option at least once in the past. The one or more records may include, for example, records representing other options different from the first option selected by other selection subjects having the same or similar attribute values ​​as each selection subject who has selected the first option at least once in the past. In this way, the first setting unit 402 can specify one or more records for which it is preferable to treat the selection subject as a record of a situation in which the first option has been recognized.

[0071] The first setting unit 402 sets a first value, which indicates definite belongingness, as an expected value of the probability that the selecting subject belongs to the first class, for each of the one or more identified records. For example, if the probability is a value between 0 and 1, the first value is 1. This makes it possible for the first setting unit 402 to appropriately take into consideration, when learning parameters of the formula, which option the selecting subject tends to select when the selecting subject has already recognized the first option, based on the one or more identified records.

[0072] The second setting unit 403 identifies a remaining record that is different from one or more records corresponding to a selection subject who has recognized the first option, from among the multiple records acquired by the acquisition unit 401. This allows the second setting unit 403 to identify a remaining record that is preferably treated as a record of a situation in which there is a mixture of a possibility that the selection subject has recognized the first option and a possibility that the selection subject has not recognized the first option.

[0073] The second setting unit 403 sets a second value representing probabilistic belonging to the first class as an expected value of the probability that the selecting subject belongs to the first class for each record of the identified remaining records based on the reference value and the formula. For example, if the probability is 0 to 1, the second value is equal to or greater than 0 and less than 1. The second value may be equal to or greater than 0 and less than 1. This allows the first setting unit 402 to appropriately take into consideration, when learning the parameters of the formula, which option the selecting subject tends to select when there is a possibility that the selecting subject has not yet recognized the first option, based on the identified remaining records.

[0074] The second setting unit 403 may, for example, calculate a similarity between each of the remaining records and the record representing the first option. For example, the second setting unit 403 sets a second value representing a probabilistic belonging as an expected value for each of the remaining records based on a reference value and a formula, such that the expected value increases as the calculated similarity increases. This allows the first setting unit 402 to appropriately take into consideration, when learning parameters of the formula, which option the selecting subject tends to select when there is a possibility that the selecting subject has not yet recognized the first option, based on the identified remaining records.

[0075] The learning unit 404 learns the reference value and the parameters included in the formula based on the respective records and the expected values ​​acquired by the acquisition unit 401. For example, the learning unit 404 learns the reference value, the parameters included in the first formula, and the parameters included in the second formula based on the respective records and the expected values ​​acquired by the acquisition unit 401. In this way, the learning unit 404 can accurately learn the reference value and the parameters included in the formula, taking into consideration the possibility that the selection subject has not yet recognized the first option.

[0076] The learning unit 404 learns the reference value and the parameters of the formula, thereby learning a model that uses the formula to estimate the probability that each of the multiple options will be selected in response to the input of the feature amount related to each of the multiple options. This allows the learning unit 404 to learn the model with high accuracy, taking into account the possibility that the selector may not have recognized the first option.

[0077] The learning unit 404 learns a first model for cases where the selecting subject belongs to the first class, for example, by learning a reference value and a parameter of the first formula. The first model estimates the probability that each of the multiple options will be selected using the first formula, for example, in response to input of a feature amount related to each of the multiple options. This allows the learning unit 404 to accurately learn the first model for cases where the selecting subject has already recognized the first option.

[0078] The learning unit 404 learns the second model for cases where the selector does not belong to the first class, for example, by learning the reference value and the parameters of the second formula. The second model, for example, estimates the probability that each of the remaining options will be selected using the second formula in response to input of a feature value for each of the remaining options different from the first option among a plurality of options. This allows the learning unit 404 to accurately learn the second model for cases where it is unclear whether the selector has already recognized the first option.

[0079] The prediction unit 405 acquires the attribute value of the selector and the feature amount related to each option based on the prediction request. The prediction unit 405 estimates the probability that each option will be selected, for example, based on the attribute value of the selector and the feature amount related to each option, using the model learned by the learning unit 404. This allows the prediction unit 405 to accurately estimate the probability that each option will be selected.

[0080] Specifically, the prediction unit 405 may estimate a first probability that each option will be selected based on the attribute value of the selection subject and the feature amount related to each option, using a first model trained by the learning unit 404. Specifically, the prediction unit 405 may estimate a second probability that each option will be selected based on the attribute value of the selection subject and the feature amount related to each option, using a second model trained by the learning unit 404. This allows the prediction unit 405 to accurately estimate the probability that each option will be selected.

[0081] The output unit 406 outputs the processing result of at least any one of the functional units. The output format is, for example, display on a display, printout on a printer, transmission to an external device via the network I / F 303, or storage in a storage area such as the memory 302 or the recording medium 305. In this way, the output unit 406 can notify the user of the processing result of each functional unit, and can support the management and operation of the information processing device 100, for example, updating of setting values ​​of the information processing device 100, thereby improving the convenience of the information processing device 100.

[0082] The output unit 406 outputs, for example, the model learned by the learning unit 404. Specifically, the output unit 406 transmits the model to another computer. Specifically, the output unit 406 outputs the model so that the user can refer to it. In this way, the output unit 406 can make the model available externally.

[0083] The output unit 406 outputs, for example, the probability estimated by the prediction unit 405. Specifically, the output unit 406 transmits the probability to another computer. Specifically, the output unit 406 outputs the probability so that it can be referenced by a user. In this way, the output unit 406 can make the probability available externally.

[0084] (Operation flow of information processing device 100) Next, the flow of operations of the information processing device 100 will be described with reference to FIG.

[0085] Fig. 5 is an explanatory diagram showing the flow of operations of the information processing device 100. In Fig. 5, the information processing device 100 stores an option J, which may not be recognized by the selection subject, among a plurality of options. The option J is set in advance by a user, for example. The information processing device 100 sets an expected value of the probability that a selection subject who has selected option J in the past belongs to class 501, so that the selection subject definitely belongs to class 501 that recognizes option J.

[0086] The information processing device 100 sets an expected value of the probability that a selecting subject who has not selected option J in the past belongs to class 501 so that the selecting subject probabilistically belongs to class 501 that recognizes option J and class 502 that does not recognize option J. For example, the information processing device 100 sets an expected value of the probability that a selecting subject who has not selected option J in the past belongs to class 501 so that the greater the effect of option J, the smaller the probability that the selecting subject who has not selected option J in the past belongs to class 501.

[0087] The information processing device 100 estimates parameters of the selection models 511 and 512 based on records representing options previously selected by each selection subject and the respective expected values ​​that have been set. The selection model 511 corresponds to, for example, a case where option J is included in a selectable option set. The selection model 512 corresponds to, for example, a case where option J is not included in a selectable option set. An example of estimating parameters of the selection models 511 and 512 is shown in a first operation example described later with reference to Figs. 6 to 13, or a second operation example described later with reference to Figs. 16 to 21.

[0088] (First operation example of information processing device 100) Next, a first operation example of the information processing device 100 will be described with reference to FIGS.

[0089] 6 to 13 are explanatory diagrams showing a first operation example of the information processing device 100. Fig. 6 shows a specific example of a functional configuration for realizing a first learning process in which the information processing device 100 in the first operation example learns a selection model. In Fig. 6, the information processing device 100 includes an acquisition unit 601, a recording unit 602, a model setting unit 603, a recognition flag assignment unit 604, a learning data generation unit 605, and a learning processing unit 606.

[0090] The acquiring unit 601 acquires selection history data including an observation number identifying one selection action of any selecting subject, option attributes of each option in the situation of the selection action, and the option selected in the selection action. The option attributes are, for example, time or monetary costs. The selection history data may include personal information regarding the selecting subject. The personal information may include, for example, an individual number identifying the selecting subject. The personal information may include, for example, personal attributes representing the characteristics of the selecting subject. The personal attributes are, for example, age, sex, location, or affiliation. The acquiring unit 601 transmits the selection history data to the recording unit 602.

[0091] The recording unit 602 receives selection history data from the acquisition unit 601. The recording unit 602 records the received selection history data in a selection history data management table 700, which will be described later with reference to Fig. 7. Now, moving on to the explanation of Fig. 7, an example of the contents stored in the selection history data management table 700 will be described. The selection history data management table 700 is realized, for example, by a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in Fig. 3.

[0092] Fig. 7 shows an example of the contents stored in the selection history data management table 700. As shown in Fig. 7, the selection history data management table 700 has fields for observation number, individual number, option attribute, and selection. The selection history data management table 700 stores selection history data as record 700-a by setting information in each field for each selection action. a is an arbitrary integer.

[0093] In the observation number field, an observation number is set that identifies one selection action of any one of the selection subjects. In the individual number field, an individual number is set that identifies any one of the selection subjects that performed the selection action. In the option attribute field, an option attribute is set. An option attribute is, for example, time or monetary cost.

[0094] The field of the option attribute has, for example, fields for walking, car, and e-scooter. The field for walking has, for example, fields for cost and required time. Specifically, the cost field has an index value that represents the monetary cost incurred when the selecting subject selects walking as a means of transportation. Specifically, the required time field has an index value that represents the time cost incurred when the selecting subject selects walking as a means of transportation.

[0095] The automobile field has, for example, fields for cost and required time. Specifically, the cost field has an index value set therein that represents the monetary cost incurred when the selecting subject selects an automobile as a means of transportation. Specifically, the required time field has an index value set therein that represents the time cost incurred when the selecting subject selects an automobile as a means of transportation.

[0096] The e-scooter field has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation. The selection field is set with the option selected in the above selection action.

[0097] Returning to the explanation of FIG. 6, the model setting unit 603 sets one of the multiple options as a specific option J that is not necessarily recognized by the selection subject. The specific option J is, for example, an e-scooter. The model setting unit 603 sets an initial value of a parameter of a utility function and a priori probability of a class probability, which are referred to by the selection model learned by the learning processing unit 606. The initial value is, for example, 0. The initial value may be, for example, a random real number. The class probability is, for example, the probability that the selection subject belongs to a class indicating that the specific option J has been recognized. The class probability is, for example, a value between 0 and 1. The priori probability may be, for example, a random real number.

[0098] Specifically, the information processing device 100 defines the personal number n that identifies the selecting subject as shown in the following formula (1).

[0099]

number

[0100] Specifically, the information processing device 100 defines a latent class index k for identifying a latent class into which a selector n is classified, as shown in the following formula (2). When k has a value of 1, it indicates a latent class representing that a specific option J has been recognized. When k has a value of 2, it indicates a latent class representing that a specific option J has not been recognized.

[0101]

number

[0102] Specifically, the information processing device 100 defines an option index for identifying each option when k=1 as shown in the following formula (3).

[0103]

number

[0104] Specifically, the information processing device 100 defines an option index for identifying each option when k=2 as shown in the following formula (4).

[0105]

number

[0106] Specifically, the information processing device 100 defines the number of selections t of each selecting subject n as shown in the following formula (5).

[0107]

number

[0108] The information processing device 100 calculates an indicator variable q that indicates whether the selecting subject n belongs to the latent class k. nk is defined as shown in the following formula (6) and formula (7).

[0109]

number

[0110]

number

[0111] The information processing device 100 calculates an indicator variable y indicating the option j selected by the selection subject n with the selection count t. njt is defined as shown in the following formula (8) and formula (9).

[0112]

number

[0113]

number

[0114] The information processing device 100 converts the option attribute vector into X njt The information processing device 100 defines the prior probability of the class probability of the latent class k as π k The information processing device 100 defines π k is defined as shown in the following formula (10) and formula (11).

[0115]

number

[0116]

number

[0117] The information processing device 100 calculates the random utility U of the option j for the number of selections t of the selection subject n in the latent class k. njt|k is defined as shown in the following formula (12). njt|k is a random term. The information processing device 100 calculates the fixed utility V njt|k is defined as shown in the following formula (13). jk are the parameters of the utility function of choice j in latent class k. The utility function may be common to latent classes k=1,2, for example.

[0118]

number

[0119]

number

[0120] The information processing device 100 defines the class 1 specific selection model, which is a selection model for calculating the probability p(·) of a selector n in class k=1 selecting an option j with a selection count t, as shown in the following formula (14). Here, q nk , "q nk|k=1 " should be written.

[0121]

number

[0122] The information processing device 100 defines the class 2 specific selection model, which is a selection model for calculating the probability p(·) of a selector n in class k=2 selecting an option j with a selection count t, as shown in the following formula (15). Here, q nk , "q nk|k=2 " should be written.

[0123]

number

[0124] 8, an example of a graphical model 800 corresponding to the class 1 eigenselection model and the class 2 eigenselection model is described. k Determined by q nk The node that calculates X njt A node that calculates q nk and β jk and X njt y is determined by njt The node that computes

[0125] Returning to the explanation of FIG. 6, the information processing device 100 specifically calculates the determined utility V njt|k =V walk is defined as shown in the following equation (16). walk For example, is common to latent classes k=1,2.

[0126]

number

[0127] Specifically, the information processing device 100 calculates the determined utility V njt|k =V autois defined as shown in the following equation (17). auto For example, is common to latent classes k=1,2.

[0128]

number

[0129] Specifically, the information processing device 100 calculates the determined utility V njt|k =V e-scooter is defined as shown in the following equation (18). e-scooter For example, is common to latent classes k=1,2.

[0130]

number

[0131] The cognition flag assignment unit 604 refers to the selection history data management table 700 and assigns a cognition flag of 1 to selection history data in which a specific option J has been selected. The cognition flag assignment unit 604 refers to the selection history data management table 700 and assigns a cognition flag of 0 to selection history data in which a specific option J has not been selected. The cognition flag assignment unit 604 may refer to the selection history data management table 700 and assign a cognition flag of 1 to selection history data including a personal number identifying a selection subject who has selected a specific option J. The cognition flag assignment unit 604 transmits the selection history data to which the cognition flag has been assigned to the learning data generation unit 605.

[0132] The learning data generation unit 605 generates learning data to be used by the learning processing unit 606. The learning data generation unit 605 receives selection history data to which a recognition flag has been added from the recognition flag adding unit 604. The learning data generation unit 605 generates learning data by replacing a character string representing a means of transportation in the received selection history data with a number corresponding to the means of transportation.

[0133] The learning data generating unit 605 records the generated learning data in a learning data management table 900, which will be described later with reference to Fig. 9. Now, moving on to the explanation of Fig. 9, an example of the contents stored in the learning data management table 900 will be described. The learning data management table 900 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in Fig. 3.

[0134] Fig. 9 shows an example of the contents stored in the learning data management table 900. As shown in Fig. 9, the learning data management table 900 has fields for observation number, individual number, option attribute, selection, and recognition flag. The learning data management table 900 stores learning data as record 900-b by setting information in each field for each selection action. b is an arbitrary integer.

[0135] In the observation number field, an observation number is set that identifies one selection action of any one of the selection subjects. In the individual number field, an individual number is set that identifies any one of the selection subjects that performed the selection action. In the option attribute field, an option attribute is set. An option attribute is, for example, time or monetary cost.

[0136] The field of the option attribute has, for example, a field for the number 1 indicating walking, a field for the number 2 indicating a car, and a field for the number 3 indicating an e-scooter. The field for the number 1 has, for example, fields for cost and required time. Specifically, the cost field has an index value set therein that indicates the monetary cost incurred when the selecting subject selects walking as a means of transportation. Specifically, the required time field has an index value set therein that indicates the time-related cost incurred when the selecting subject selects walking as a means of transportation.

[0137] The field 2 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an automobile as the means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an automobile as the means of transportation.

[0138] The field 3 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation. The selection field is set with a number indicating the option selected in the above selection action. The recognition flag field is set with a recognition flag.

[0139] Returning to the description of FIG. 6, the learning processing unit 606 acquires a convergence condition. The convergence condition is, for example, that the difference value of the log likelihood is less than a threshold. The convergence condition is, for example, that the parameter β of the utility function is jk may be learned a predetermined number of times. The learning processing unit 606 acquires learning data by referring to the learning data management table 900. The learning processing unit 606 learns parameters that define the class 1 eigenselection model and the class 2 eigenselection model based on the acquired learning data until the convergence condition is satisfied. For example, the learning processing unit 606 learns parameters β of the utility function that defines the class 1 eigenselection model and the class 2 eigenselection model until the convergence condition is satisfied. jk Learn.

[0140] Specifically, the learning processing unit 606 calculates the parameter β jk The initialization is, for example, setting 0. The initialization may be, for example, setting a random real number. Specifically, the learning processing unit 606 initializes any of the parameters β jkIf any of the parameters β jk A constant may be set to x, y ...

[0141] (6-1) More specifically, the learning processing unit 606 sets, in accordance with the following formula (19), an expected value of the class probability corresponding to each piece of learning data including a personal number identifying a selection subject who has never selected a specific option J. More specifically, the learning processing unit 606 sets, in accordance with the following formula (20), an expected value of the class probability corresponding to each piece of learning data including a personal number identifying a selection subject who has selected a specific option J.

[0142]

number

[0143]

number

[0144] (6-2) More specifically, the learning processing unit 606 calculates the expected value of the log-likelihood according to the following formula (21). The Roman numeral 2 symbol represents an indicator function. More specifically, the learning processing unit 606 calculates the parameter β of the utility function according to the following formula (22) based on the expected value of the log-likelihood. k More specifically, the learning processing unit 606 updates the class probabilities according to the following formulas (23) and (24).

[0145]

number

[0146]

number

[0147]

number

[0148]

number

[0149] (6-3) More specifically, the learning processing unit 606 repeats the processes shown in (6-1) and (6-2) until the convergence condition is satisfied. As a result, the learning processing unit 606 adjusts the parameter β k Therefore, the learning processing unit 606 can appropriately update the parameter β k The learning processing unit 606 can appropriately learn a selection model that refers to the parameter β k and the class probabilities to the recording unit 602.

[0150] The recording unit 602 records the parameter β of the utility function, which is a parameter of the selection model learned by the learning processing unit 606. k The recording unit 602 records the class probability learned by the learning processing unit 606 for any one of the selecting subjects as a posterior probability in association with the personal number that identifies that one of the selecting subjects. This allows the recording unit 602 to make the selection model and the posterior probability available. Next, we move on to the explanation of FIG. 10.

[0151] 10 illustrates a specific example of a functional configuration for implementing a first prediction process in which the information processing device 100 in the first operation example uses a selection model. In FIG. 10, the information processing device 100 includes an acquisition unit 1001, a recording unit 1002, a prediction data generation unit 1003, and a prediction processing unit 1004.

[0152] The acquiring unit 1001 acquires prediction target data including a prediction number that identifies one selection action of any selecting subject, and option attributes of each option in the situation of the selection action. The selection action may be, for example, a selection action that is planned to be performed, but may not have been performed yet. The prediction target data may include an individual number that identifies the selecting subject. The individual number in the prediction target data is common to the individual number in the prediction target data. The prediction target data may include individual attributes that represent the characteristics of the selecting subject. The acquiring unit 1001 transmits the prediction target data to the recording unit 1002.

[0153] The recording unit 1002 receives prediction target data from the acquisition unit 1001. The recording unit 1002 records the received prediction target data in a prediction target data management table 1100, which will be described later with reference to FIG. 11. Now, moving to the explanation of FIG. 11, an example of the storage contents of the prediction target data management table 1100 will be explained. The prediction target data management table 1100 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in FIG. 3.

[0154] Fig. 11 shows an example of the contents stored in the prediction target data management table 1100. As shown in Fig. 11, the prediction target data management table 1100 has fields for a prediction number, a personal number, and an option attribute. The prediction target data management table 1100 stores prediction target data as a record 1100-c by setting information in each field for each selection action that is planned to be performed. c is an arbitrary integer.

[0155] In the prediction number field, a prediction number is set that identifies one selection action of any one of the selection subjects. In the personal number field, a personal number is set that identifies any one of the selection subjects that performed the selection action. In the option attribute field, an option attribute is set. An option attribute is, for example, a time or monetary cost.

[0156] The field of the option attribute has, for example, fields for walking, car, and e-scooter. The field for walking has, for example, fields for cost and required time. Specifically, the cost field has an index value that represents the monetary cost incurred when the selecting subject selects walking as a means of transportation. Specifically, the required time field has an index value that represents the time cost incurred when the selecting subject selects walking as a means of transportation.

[0157] The automobile field has, for example, fields for cost and required time. Specifically, the cost field has an index value set therein that represents the monetary cost incurred when the selecting subject selects an automobile as a means of transportation. Specifically, the required time field has an index value set therein that represents the time cost incurred when the selecting subject selects an automobile as a means of transportation.

[0158] The e-scooter field has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation.

[0159] Returning to the explanation of Fig. 10, the calculation data generating unit 1003 refers to the prediction target data management table 1100, and generates calculation data to be referenced by the prediction processing unit 1004, based on the prediction target data. The calculation data generating unit 1003 generates calculation data by replacing a character string representing a means of transportation in the received selection history data with a number corresponding to the means of transportation.

[0160] The prediction data generating unit 1003 assigns a class probability to the prediction data. When the prediction data includes a personal number, the prediction data generating unit 1003 uses the recording unit 1002 to assign a posterior probability corresponding to the personal number as a class probability to the prediction data. When the prediction data generating unit 1003 uses the recording unit 1002 to determine that there is no posterior probability corresponding to the personal number, the prediction data generating unit 1003 assigns a priori probability as a class probability to the prediction data. The prediction data generating unit 1003 may normalize the value of the prediction data.

[0161] The calculation data generating unit 1003 records the generated calculation data in a calculation data management table 1200, which will be described later with reference to Fig. 12. Now, moving to the explanation of Fig. 12, an example of the stored contents of the calculation data management table 1200 will be explained. The calculation data management table 1200 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in Fig. 3.

[0162] Fig. 12 shows an example of the contents stored in the prediction data management table 1200. As shown in Fig. 12, the prediction data management table 1200 has fields for a prediction number, a personal number, an option attribute, and a class probability. The prediction data management table 1200 stores prediction data as a record 1200-d by setting information in each field for each selection action. d is an arbitrary integer.

[0163] In the prediction number field, a prediction number is set that identifies one selection action of any one of the selection subjects. In the personal number field, a personal number is set that identifies any one of the selection subjects that performed the selection action. In the option attribute field, an option attribute is set. An option attribute is, for example, a time or monetary cost.

[0164] The field of the option attribute has, for example, a field for the number 1 indicating walking, a field for the number 2 indicating a car, and a field for the number 3 indicating an e-scooter. The field for the number 1 has, for example, fields for cost and required time. Specifically, the cost field has an index value set therein that indicates the monetary cost incurred when the selecting subject selects walking as a means of transportation. Specifically, the required time field has an index value set therein that indicates the time-related cost incurred when the selecting subject selects walking as a means of transportation.

[0165] The field 2 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an automobile as the means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an automobile as the means of transportation.

[0166] The field 3 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with a required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation. The class probability field is set with a class probability.

[0167] Returning to the description of Fig. 10, the prediction processing unit 1004 acquires parameters of the selection model from the recording unit 1002. The prediction processing unit 1004 refers to the calculation data management table 1200 and calculates the selection probability that each option will be selected for each piece of calculation data.

[0168] The prediction processing unit 1004 calculates, for example, a first selection probability at which each option is selected using a model 1 specific selection model for the prediction data. The prediction processing unit 1004 calculates, for example, a second selection probability at which each option is selected using a model 2 specific selection model for the prediction data. The prediction processing unit 1004 calculates, for example, for each option, the sum of a value obtained by multiplying the calculated first selection probability by the class probability as a weight and a value obtained by multiplying the calculated second selection probability by (1-class probability) as a weight, as the selection probability. This allows the prediction processing unit 1004 to accurately calculate the selection probability at which each option is selected.

[0169] The prediction processing unit 1004 transmits, for example, the calculated selection probability to the recording unit 1002. The recording unit 1002 records the selection probability at which each of the received options is selected in association with the prediction data. This allows the recording unit 1002 to make the selection probability at which each option is selected available. Next, we move on to the description of FIG. 13.

[0170] 13 shows a specific example of a functional configuration in which the information processing device 100 in the first operation example realizes a first learning process for learning a selection model and realizes a first prediction process using the selection model. In FIG. 13, the information processing device 100 includes an acquisition unit 1301, a recording unit 1302, a model setting unit 1303, a recognition flag assignment unit 1304, a learning data generation unit 1305, a learning processing unit 1306, a prediction data generation unit 1307, and a prediction processing unit 1308.

[0171] The acquisition unit 1301 corresponds to the acquisition unit 601 and the acquisition unit 1001. The recording unit 1302 corresponds to the recording unit 602 and the recording unit 1002. The model setting unit 1303 corresponds to the model setting unit 603. The recognition flag assignment unit 1304 corresponds to the recognition flag assignment unit 604. The learning data generation unit 1305 corresponds to the learning data generation unit 605. The learning processing unit 1306 corresponds to the learning processing unit 606. The prediction data generation unit 1307 corresponds to the prediction data generation unit 1003. The prediction processing unit 1308 corresponds to the prediction processing unit 1004.

[0172] As a result, the information processing device 100 can learn a selection model that can accurately estimate the selection probability of each option being selected, taking into account the recognition of a specific option J by a selector. The information processing device 100 can perform a simulation using the selection model. The simulation relates to, for example, people flow or purchasing.

[0173] (Application example of information processing device 100) The information processing device 100 can be applied, for example, to the field of traffic simulation. For example, the information processing device 100 can perform a simulation that simulates what kind of transportation means a moving person who is the main selection agent will select. For example, the information processing device 100 can analyze, by simulation, how the transportation means selected by the moving person will change due to changes in the environment such as traffic congestion, road tolls, and transportation costs. Therefore, the information processing device 100 can obtain information that contributes to the consideration of urban policies, for example. In addition, the information processing device 100 can be applied, for example, to the field of economics. For example, the information processing device 100 can perform a simulation that simulates what kind of product a customer who is the main selection agent will select.

[0174] (First learning process procedure) Next, an example of a first learning process procedure executed by the information processing device 100 in the first operation example will be described with reference to Fig. 14. The first learning process is realized by, for example, the CPU 301, storage areas such as the memory 302 and the recording medium 305, and the network I / F 303 shown in Fig. 3.

[0175] Fig. 14 is a flowchart showing an example of a procedure of the first learning process. In Fig. 14, the information processing device 100 acquires a plurality of pieces of selection history data (step S1401). The information processing device 100 initializes parameters of a selection model (step S1402).

[0176] The information processing device 100 calculates the expected value of the class probability corresponding to each of the multiple selection history data acquired (step S1403). The information processing device 100 corrects the expected value of the class probability corresponding to one or more selection history data whose recognition flag indicates recognition among the multiple selection history data acquired (step S1404).

[0177] The information processing device 100 calculates an expected value of the log likelihood (step S1405). The information processing device 100 updates parameters of the selection model (step S1406). The information processing device 100 determines whether or not the convergence condition is satisfied (step S1407). If the convergence condition is not satisfied (step S1407: No), the information processing device 100 returns to the process of step S1403. On the other hand, if the convergence condition is satisfied (step S1407: Yes), the information processing device 100 ends the first learning process.

[0178] (First prediction processing procedure) Next, an example of a first prediction process procedure executed by the information processing device 100 in the first operation example will be described with reference to Fig. 15. The first prediction process is realized by, for example, the CPU 301, storage areas such as the memory 302 and the recording medium 305, and the network I / F 303 shown in Fig. 3.

[0179] 15 is a flowchart illustrating an example of a procedure of the first prediction process. In FIG 15, the information processing device 100 acquires prediction target data (step S1501).

[0180] The information processing device 100 determines whether the personal number included in the acquired prediction target data is an existing personal number for which the class probability has been calculated (step S1502). If it is an existing personal number (step S1502: Yes), the information processing device 100 proceeds to processing of step S1503. On the other hand, if it is not an existing personal number (step S1502: No), the information processing device 100 proceeds to processing of step S1504.

[0181] In step S1503, the information processing device 100 sets the calculated class probability to the class probability corresponding to the target (step S1503). Then, the information processing device 100 proceeds to the process of step S1505.

[0182] In step S1504, the information processing device 100 sets a prior probability to the class probability corresponding to the target (step S1504). Then, the information processing device 100 proceeds to the process of step S1505.

[0183] In step S1505, the information processing device 100 calculates the selection probability of each option based on the set class probability (step S1505). The information processing device 100 ends the first prediction process.

[0184] (Second operation example of information processing device 100) 16 to 21, a second operation example of the information processing device 100 will be described. The second operation example differs from the first operation example, for example, in that the information processing device 100 uses personal attributes.

[0185] 16 to 21 are explanatory diagrams showing a second operation example of the information processing device 100. A specific example of a functional configuration for realizing a second learning process in which the information processing device 100 learns a selection model in the second operation example is similar to the specific example of the functional configuration shown in Fig. 6. The information processing device 100 includes an acquisition unit 601, a recording unit 602, a model setting unit 603, a recognition flag assignment unit 604, a learning data generation unit 605, and a learning processing unit 606.

[0186] The acquiring unit 601 acquires selection history data including an observation number that identifies one selection action of any selecting subject, option attributes of each option in the situation of the selection action, and the option selected in the selection action. The option attributes are, for example, time or monetary costs. The selection history data includes personal information regarding the selecting subject. The personal information includes, for example, a personal number that identifies the selecting subject. The personal information includes, for example, personal attributes that represent the characteristics of the selecting subject. The personal attributes are, for example, age and gender. The acquiring unit 601 transmits the selection history data to the recording unit 602.

[0187] The recording unit 602 receives selection history data from the acquisition unit 601. The recording unit 602 records the received selection history data in a selection history data management table 1600, which will be described later with reference to Fig. 16. Now, moving on to the explanation of Fig. 16, an example of the contents stored in the selection history data management table 1600 will be described. The selection history data management table 1600 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in Fig. 3.

[0188] Fig. 16 shows an example of the contents stored in the selection history data management table 1600. As shown in Fig. 16, the selection history data management table 1600 has fields for observation number, individual number, individual attribute, option attribute, and selection. The selection history data management table 1600 stores selection history data as record 1600-e by setting information in each field for each selection action. e is an arbitrary integer.

[0189] The observation number field is set with an observation number that identifies one selection action of any one of the selection subjects. The personal number field is set with an personal number that identifies any one of the selection subjects that performed the selection action. The personal attribute field has, for example, fields for gender and age. The gender field is set with the gender of any one of the selection subjects that performed the selection action. The age field is set with the age of any one of the selection subjects that performed the selection action.

[0190] In the option attribute field, an option attribute is set. The option attribute is, for example, a time or monetary cost. The option attribute field has, for example, fields for walking, car, and e-scooter. The walking field has, for example, fields for cost and required time. In the cost field, specifically, an index value representing the monetary cost incurred when the selecting subject selects walking as a means of transportation is set. In the required time field, specifically, a required time representing the time cost incurred when the selecting subject selects walking as a means of transportation is set.

[0191] The automobile field has, for example, fields for cost and required time. Specifically, the cost field has an index value set therein that represents the monetary cost incurred when the selecting subject selects an automobile as a means of transportation. Specifically, the required time field has an index value set therein that represents the time cost incurred when the selecting subject selects an automobile as a means of transportation.

[0192] The e-scooter field has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation. The selection field is set with the option selected in the above selection action.

[0193] The model setting unit 603 sets one of the multiple options as a specific option J that is not necessarily recognized by the selecting subject. The specific option J is, for example, an e-scooter. The model setting unit 603 sets an initial value of a parameter of a utility function, a priori probability of a class probability, and a parameter that specifies the distribution of personal attributes for each class, which are referred to by the selection model learned by the learning processing unit 606. The initial value is, for example, 0. The initial value may be, for example, a random real number. The class probability is, for example, the probability that the selecting subject belongs to a class that indicates that the specific option J has been recognized. The class probability is, for example, a value between 0 and 1. The priori probability may be, for example, a random real number.

[0194] Specifically, the information processing device 100 defines the personal number n that identifies the selecting subject as shown in the following formula (25).

[0195]

number

[0196] Specifically, the information processing device 100 defines a latent class index k for identifying a latent class into which a selector n is classified, as shown in the following formula (26). When k has a value of 1, it indicates a latent class representing that a specific option J has been recognized. When k has a value of 2, it indicates a latent class representing that a specific option J has not been recognized.

[0197]

number

[0198] Specifically, the information processing device 100 defines an option index for identifying each option when k=1 as shown in the following formula (27).

[0199]

number

[0200] Specifically, the information processing device 100 defines an option index for identifying each option when k=2 as shown in the following formula (28).

[0201]

number

[0202] Specifically, the information processing device 100 defines the number of selections t of each selecting subject n as shown in the following formula (29).

[0203]

number

[0204] The information processing device 100 calculates an indicator variable q that indicates whether the selecting subject n belongs to the latent class k. nk is defined as shown in the following formula (30) and formula (31).

[0205]

number

[0206]

number

[0207] The information processing device 100 calculates an indicator variable y indicating the option j selected by the selection subject n with the selection count t. njt is defined as shown in the following formula (32) and formula (33).

[0208]

number

[0209]

number

[0210] The information processing device 100 converts the option attribute vector into X njt The information processing device 100 defines the continuous-value personal attribute vector as S cn The information processing device 100 defines the individual attribute vector of discrete values ​​as S dn The information processing device 100 defines the prior probability of the class probability of the latent class k as π k The information processing device 100 defines π k is defined as shown in the following formula (34) and formula (35).

[0211]

number

[0212]

number

[0213] The information processing device 100 calculates a continuous-value personal attribute vector S cn The distribution of is defined as shown in the following equation (36).

[0214]

number

[0215] The information processing device 100 calculates a discrete-value personal attribute vector S dn The distribution of is defined as shown in the following equation (37).

[0216]

number

[0217] The information processing device 100 calculates the random utility U of the option j for the number of selections t of the selection subject n in the latent class k. njt|k is defined as shown in the following formula (38). The information processing device 100 defines the determined utility Vnjt|k is defined as shown in the following equation (39). jk are the parameters of the utility function of choice j in latent class k. The utility function may be common to latent classes k=1,2, for example.

[0218]

number

[0219]

number

[0220] The information processing device 100 defines the class 1 specific selection model, which is a selection model for calculating the probability p(·) of a selector n in class k=1 selecting an option j with a selection count t, as shown in the following formula (40). Here, q nk , "q nk|k=1 " should be written.

[0221]

number

[0222] The information processing device 100 defines the class 2 specific selection model, which is a selection model for calculating the probability p(·) that a selector n in class k=2 selects an option j with a selection count t, as shown in the following formula (41). Here, q nk , "q nk|k=2 " should be written.

[0223]

number

[0224] Next, we move on to the description of Fig. 17 and describe an example of a graphical model 1700 corresponding to a class 1 eigenselection model and a class 2 eigenselection model. The graphical model 1700 is formed by, for example, a plurality of nodes. The plurality of nodes may be, for example, π k Determined by q nk The node that calculates X njt A node that calculates q nk and β jk and X njt y is determined by njt The nodes include a node that calculates, for example, q nk and μ c and Σ c S is determined by cn A node that calculates q nk and μ d S is determined by dn and a node for calculating

[0225] Specifically, the information processing device 100 calculates the determined utility V njt|k =V walk is defined as shown in the above equation (16). walk is common to latent classes k=1, 2, for example. njt|k =V auto is defined as shown in the above equation (17). auto is common to latent classes k=1, 2, for example. Specifically, the information processing device 100 calculates the determined utility V njt|k =V e-scooter is defined as shown in the above equation (18). e-scooter For example, is common to latent classes k=1,2.

[0226] The recognition flag assignment unit 604 refers to the selection history data management table 1600 and assigns a recognition flag of 1 to selection history data in which a specific option J has been selected. The recognition flag assignment unit 604 refers to the selection history data management table 1600 and assigns a recognition flag of 0 to selection history data in which a specific option J has not been selected. The recognition flag assignment unit 604 may refer to the selection history data management table 1600 and assign a recognition flag of 1 to selection history data including a personal number identifying a selection subject who has selected a specific option J. The recognition flag assignment unit 604 transmits the selection history data to which the recognition flag has been assigned to the learning data generation unit 605.

[0227] The learning data generation unit 605 generates learning data to be used by the learning processing unit 606. The learning data generation unit 605 receives selection history data to which a recognition flag has been assigned from the recognition flag assignment unit 604. The learning data generation unit 605 generates learning data by replacing, from the received selection history data, a character string representing a means of transportation with a number corresponding to the means of transportation, and replacing a character string representing the gender of the personal attributes with a number corresponding to the gender.

[0228] The learning data generating unit 605 records the generated learning data in a learning data management table 1800, which will be described later with reference to Fig. 18. Now, moving on to the explanation of Fig. 18, an example of the contents stored in the learning data management table 1800 will be described. The learning data management table 1800 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in Fig. 3.

[0229] Fig. 18 shows an example of the contents stored in the learning data management table 1800. As shown in Fig. 18, the learning data management table 1800 has fields for observation number, individual number, individual attribute, option attribute, selection, and recognition flag. The learning data management table 1800 stores learning data as record 1800-f by setting information in each field for each selection action. f is an arbitrary integer.

[0230] The observation number field is set with an observation number that identifies one selection action of any one of the selection subjects. The personal number field is set with an personal number that identifies any one of the selection subjects that performed the selection action. The personal attribute field has, for example, fields for gender and age. The gender field is set with a number indicating the gender of any one of the selection subjects that performed the selection action. The age field is set with the age of any one of the selection subjects that performed the selection action.

[0231] In the option attribute field, an option attribute is set. The option attribute is, for example, a time or monetary cost. The option attribute field has, for example, a field for number 1 indicating walking, a field for number 2 indicating a car, and a field for number 3 indicating an e-scooter.

[0232] The field 1 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects walking as a means of transportation. Specifically, the required time field is set with a required time representing the time cost incurred when the selecting subject selects walking as a means of transportation.

[0233] The field 2 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an automobile as the means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an automobile as the means of transportation.

[0234] The field 3 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation. The selection field is set with a number indicating the option selected in the above selection action. The recognition flag field is set with a recognition flag.

[0235] The learning processing unit 606 acquires a convergence condition. The convergence condition is, for example, that the difference value of the log likelihood is less than a threshold. The convergence condition is, for example, that the parameter β of the utility function is jk may be learned a predetermined number of times. The learning processing unit 606 acquires learning data by referring to the learning data management table 1800. The learning processing unit 606 learns parameters that define the class 1 eigenselection model and the class 2 eigenselection model based on the acquired learning data until the convergence condition is satisfied. For example, the learning processing unit 606 learns parameters β of the utility function that defines the class 1 eigenselection model and the class 2 eigenselection model until the convergence condition is satisfied. jk Learn.

[0236] Specifically, the learning processing unit 606 calculates the parameter β jk The initialization is, for example, setting 0. The initialization may be, for example, setting a random real number. Specifically, the learning processing unit 606 initializes any of the parameters β jk If any of the parameters β jk A constant may be set to x, y ...

[0237] (18-1) More specifically, the learning processing unit 606 sets an expected value of the class probability corresponding to each piece of learning data including a personal number identifying a selection subject who has never selected a specific option J, according to the following formula (42) and formula (43): More specifically, the learning processing unit 606 sets an expected value of the class probability corresponding to each piece of learning data including a personal number identifying a selection subject who has selected a specific option J, according to the following formula (44):

[0238]

number

[0239]

number

[0240]

number

[0241] (18-2) More specifically, the learning processing unit 606 calculates the expected value of the log likelihood according to the following formula (45). More specifically, the learning processing unit 606 calculates the parameter β of the utility function according to the following formulas (46) to (50) based on the expected value of the log likelihood. k More specifically, the learning processing unit 606 updates the class probabilities according to the following equation (51).

[0242]

number

[0243]

number

[0244]

number

[0245]

number

[0246]

number

[0247]

number

[0248]

number

[0249] (18-3) More specifically, the learning processing unit 606 repeats the processes shown in (18-1) and (18-2) until the convergence condition is satisfied. As a result, the learning processing unit 606 adjusts the parameter β k Therefore, the learning processing unit 606 can appropriately update the parameter β k The learning processing unit 606 can appropriately learn a selection model that refers to the parameter β k and the class probabilities to the recording unit 602.

[0250] The recording unit 602 records the parameter β of the utility function, which is a parameter of the selection model learned by the learning processing unit 606. k The recording unit 602 records the class probability learned by the learning processing unit 606 for any one of the selecting subjects as a posterior probability in association with the personal number that identifies the selecting subject. This allows the recording unit 602 to make the selection model and the posterior probability available. Next, we move on to the explanation of FIG. 19.

[0251] 19 illustrates a specific example of a functional configuration for implementing the second prediction process in which the information processing device 100 in the second operation example uses a selection model. In FIG. 19, the information processing device 100 includes an acquisition unit 1901, a recording unit 1902, a prediction data generation unit 1903, and a prediction processing unit 1904.

[0252] The acquiring unit 1901 acquires prediction target data including a prediction number that identifies one selection action of any selecting subject, and option attributes of each option in the situation of the selection action. The selection action may be, for example, a selection action that is planned to be performed and has not yet been performed. The prediction target data includes an individual number that identifies the selecting subject. The individual number in the prediction target data is common to the individual number in the prediction target data. The prediction target data includes individual attributes that represent the characteristics of the selecting subject. The acquiring unit 1901 transmits the prediction target data to the recording unit 1902.

[0253] The recording unit 1902 receives prediction target data from the acquisition unit 1901. The recording unit 1902 records the received prediction target data in a prediction target data management table 2000, which will be described later with reference to FIG. 20. Now, moving to the explanation of FIG. 20, an example of the contents stored in the prediction target data management table 2000 will be explained. The prediction target data management table 2000 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in FIG. 3.

[0254] Fig. 20 shows an example of the contents stored in the prediction target data management table 2000. As shown in Fig. 20, the prediction target data management table 2000 has fields for a prediction number, a personal number, a personal attribute, and an option attribute. The prediction target data management table 2000 stores prediction target data as a record 2000-g by setting information in each field for each selection action that is planned to be performed. g is an arbitrary integer.

[0255] The prediction number field is set with a prediction number that identifies one selection action of any one of the selection subjects. The personal number field is set with a personal number that identifies any one of the selection subjects that performed the selection action. The personal attribute field has, for example, fields for gender and age. The gender field is set with the gender of any one of the selection subjects that performed the selection action. The age field is set with the age of any one of the selection subjects that performed the selection action.

[0256] In the option attribute field, an option attribute is set. The option attribute is, for example, a time or monetary cost. The option attribute field has, for example, fields for walking, car, and e-scooter. The walking field has, for example, fields for cost and required time. In the cost field, specifically, an index value representing the monetary cost incurred when the selecting subject selects walking as a means of transportation is set. In the required time field, specifically, a required time representing the time cost incurred when the selecting subject selects walking as a means of transportation is set.

[0257] The automobile field has, for example, fields for cost and required time. Specifically, the cost field has an index value set therein that represents the monetary cost incurred when the selecting subject selects an automobile as a means of transportation. Specifically, the required time field has an index value set therein that represents the time cost incurred when the selecting subject selects an automobile as a means of transportation.

[0258] The e-scooter field has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation.

[0259] Returning to the explanation of Fig. 19, the calculation data generating unit 1903 refers to the prediction target data management table 2000 and generates calculation data to be referenced by the prediction processing unit 1904 based on the prediction target data. The calculation data generating unit 1903 generates calculation data by replacing, in the received selection history data, a character string indicating a means of transportation with a number corresponding to the means of transportation and replacing a character string indicating the gender of the selecting subject with a number corresponding to the gender.

[0260] The prediction data generating unit 1903 assigns a class probability to the prediction data. When the prediction data includes a personal number, the prediction data generating unit 1903 uses the recording unit 602 to assign a posterior probability corresponding to the personal number as a class probability to the prediction data.

[0261] When the prediction data generating unit 1903 determines, using the recording unit 602, that there is no posterior probability corresponding to the personal number, it assigns a class probability to the prediction data according to Bayes' theorem. For example, the prediction data generating unit 1903 calculates a posterior probability based on the prior probability and the distribution of personal attributes according to Bayes' theorem, and assigns it to the prediction data as a class probability. The prediction data generating unit 1903 may normalize the value of the prediction data.

[0262] The calculation data generating unit 1903 records the generated calculation data in a calculation data management table 2100, which will be described later with reference to Fig. 21. Now, moving to the explanation of Fig. 21, an example of the stored contents of the calculation data management table 2100 will be explained. The calculation data management table 2100 is realized by, for example, a storage area such as the memory 302 or the recording medium 305 of the information processing device 100 shown in Fig. 3.

[0263] Fig. 21 shows an example of the contents stored in the prediction data management table 2100. As shown in Fig. 21, the prediction data management table 2100 has fields for a prediction number, a personal number, a personal attribute, an option attribute, and a class probability. The prediction data management table 2100 stores prediction data as a record 1200-h by setting information in each field for each selection action. h is an arbitrary integer.

[0264] The prediction number field is set with a prediction number that identifies one selection action of any one of the selection subjects. The personal number field is set with a personal number that identifies any one of the selection subjects that performed the selection action. The personal attribute field has, for example, fields for gender and age. The gender field is set with a number indicating the gender of any one of the selection subjects that performed the selection action. The age field is set with the age of any one of the selection subjects that performed the selection action.

[0265] In the option attribute field, an option attribute is set. The option attribute is, for example, a time or monetary cost. The option attribute field has, for example, a field of number 1 indicating walking, a field of number 2 indicating a car, and a field of number 3 indicating an e-scooter. The field of number 1 has, for example, a field of cost and a required time. In the cost field, specifically, an index value indicating the monetary cost incurred when the selecting subject selects walking as a means of transportation is set. In the required time field, specifically, a required time indicating the time cost incurred when the selecting subject selects walking as a means of transportation is set.

[0266] The field 2 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an automobile as the means of transportation. Specifically, the required time field is set with the required time representing the time cost incurred when the selecting subject selects an automobile as the means of transportation.

[0267] The field 3 has, for example, fields for cost and required time. Specifically, the cost field is set with an index value representing the monetary cost incurred when the selecting subject selects an e-scooter as a means of transportation. Specifically, the required time field is set with a required time representing the time cost incurred when the selecting subject selects an e-scooter as a means of transportation. The class probability field is set with a class probability.

[0268] 19, the prediction processing unit 1904 acquires parameters of the selection model from the recording unit 602. The prediction processing unit 1904 refers to the calculation data management table 2100 and calculates the selection probability that each option is selected for each piece of calculation data.

[0269] The prediction processing unit 1904 calculates, for example, a first selection probability at which each option is selected using a model 1 specific selection model for the prediction data. The prediction processing unit 1904 calculates, for example, a second selection probability at which each option is selected using a model 2 specific selection model for the prediction data. The prediction processing unit 1904 calculates, for example, for each option, the sum of a value obtained by multiplying the calculated first selection probability by the class probability as a weight and a value obtained by multiplying the calculated second selection probability by (1-class probability) as a weight, as the selection probability. This allows the prediction processing unit 1904 to accurately calculate the selection probability at which each option is selected.

[0270] The prediction processing unit 1904 transmits, for example, the calculated selection probability to the recording unit 1902. The recording unit 1902 records the received selection probability at which each option is selected in association with the prediction data. In this way, the recording unit 1902 can make the selection probability at which each option is selected available.

[0271] A specific example of a functional configuration in which the information processing device 100 in the second operation example realizes a second learning process in which the selection model is learned and a second prediction process that uses the selection model is similar to the specific example of the functional configuration shown in Figure 13, so description thereof will be omitted.

[0272] As a result, the information processing device 100 can learn a selection model that can accurately estimate the selection probability of each option being selected, taking into account the recognition of a specific option J by a selector. The information processing device 100 can perform a simulation using the selection model. The simulation relates to, for example, people flow or purchasing.

[0273] (Application example of information processing device 100) The information processing device 100 can be applied to the field of traffic simulation. For example, the information processing device 100 can perform a simulation to simulate what kind of transportation means a moving person who is the main selection agent will select. For example, the information processing device 100 can analyze, by a simulation, how the transportation means selected by the moving person will change due to changes in the environment such as traffic congestion, road tolls, and the cost of the transportation means. Therefore, the information processing device 100 can obtain information that contributes to the consideration of urban policies, for example. In addition, the information processing device 100 can be applied to the field of economics, for example. For example, the information processing device 100 can perform a simulation to simulate what kind of product a customer who is the main selection agent will select.

[0274] (Second learning process procedure) An example of the second learning processing procedure performed by the information processing device 100 in the second operation example is specifically similar to an example of the first learning processing procedure performed by the information processing device 100 in the first operation example shown in FIG. 22, and therefore will not be described.

[0275] (Second prediction processing procedure) Next, an example of a second prediction process procedure executed by the information processing device 100 in the second operation example will be described with reference to Fig. 22. The second prediction process is realized by, for example, the CPU 301, storage areas such as the memory 302 and the recording medium 305, and the network I / F 303 shown in Fig. 3.

[0276] 22 is a flowchart illustrating an example of a procedure of the second prediction process. In FIG 22, the information processing device 100 acquires prediction target data (step S2201).

[0277] The information processing device 100 determines whether the personal number included in the acquired prediction target data is an existing personal number for which the class probability has been calculated (step S2202). If it is an existing personal number (step S2202: Yes), the information processing device 100 proceeds to processing of step S2203. On the other hand, if it is not an existing personal number (step S2202: No), the information processing device 100 proceeds to processing of step S2204.

[0278] In step S2203, the information processing device 100 sets the calculated class probability to the class probability corresponding to the target (step S2203). Then, the information processing device 100 proceeds to the process of step S2205.

[0279] In step S2204, the information processing device 100 calculates a posterior probability based on the prior probability and the attribute value, and sets the posterior probability to the class probability corresponding to the target (step S2204). Then, the information processing device 100 proceeds to the process of step S2205.

[0280] In step S2205, the information processing device 100 calculates the selection probability of each option based on the set class probability (step S2205). The information processing device 100 ends the second prediction process.

[0281] As described above, the information processing device 100 can obtain a plurality of records each representing an option previously selected from a plurality of options. The information processing device 100 can specify one or more records corresponding to a selecting subject who has recognized the first option from among the plurality of records. The information processing device 100 can set a first value representing definite belonging to a first class in which the selecting subject has recognized the first option, as an expected value of the probability that the selecting subject belongs to the first class in which the first option has been recognized, for each of the specified one or more records. The information processing device 100 can store a reference value of the probability that the selecting subject belongs to the first class and a formula representing the effect of each of the multiple options. The information processing device 100 can set a second value representing probabilistic belonging to a first class, as an expected value of the probability that the selecting subject belongs to the first class, for each of the remaining records other than the specified one or more records, based on the reference value and the formula. The information processing device 100 can learn a reference value and a parameter included in the formula, based on each of the obtained multiple records and the expected value. This allows the information processing device 100 to accurately learn the reference value and the parameters included in the formula. Therefore, the information processing device 100 can easily learn a model that uses the formula to accurately estimate the option to be selected by the selection subject.

[0282] According to the information processing device 100, by learning the reference value and the parameters of the formula, it is possible to learn a model that estimates the probability that each of a plurality of options will be selected by utilizing the formula in response to the input of the feature amount related to each option. This allows the information processing device 100 to learn a model that accurately estimates the option that the selection subject will select.

[0283] According to the information processing device 100, by learning the reference value and the parameters of the formula, a first model can be learned for the case where the selecting subject belongs to the first class. According to the information processing device 100, when the selecting subject does not belong to the first class, a second model can be learned. In this way, the information processing device 100 can learn different models depending on whether the selecting subject belongs to the first class or not. By using different models, the information processing device 100 can make it possible to accurately estimate the option to be selected by the selecting subject.

[0284] According to the information processing device 100, a first value indicating definite belonging can be set as an expected value for each record of one or more records representing the first option. This allows the information processing device 100 to set a first value indicating definite belonging to a record that is determined to correspond to a selecting subject who has recognized the first option. Therefore, the information processing device 100 can easily reflect in a model what option a selecting subject who has recognized the first option will select.

[0285] According to the information processing device 100, a first value representing definite belonging can be set as an expected value for each of one or more records having the same or similar attribute value as the record representing the first option. This allows the information processing device 100 to set a first value representing definite belonging to a record corresponding to a selection subject who is relatively likely to have recognized the first option. Therefore, the information processing device 100 can easily reflect in the model what option a selection subject who has recognized the first option will select.

[0286] According to the information processing device 100, it is possible to store a formula expressing the effect of each of a plurality of options when the selecting subject belongs to the first class. According to the information processing device 100, it is possible to store a formula expressing the effect of each of the remaining options, which are different from the first option, among the plurality of options when the selecting subject does not belong to the first class. In this way, the information processing device 100 can make available a formula that takes into account whether or not the selecting subject has recognized the first option.

[0287] According to the information processing device 100, it is possible to store the similarity between each of the remaining records and the record representing the first option. According to the information processing device 100, for each of the remaining records, a second value representing probabilistic belonging can be set as the expected value so that the greater the similarity with the record representing the first option, the greater the expected value becomes. This allows the information processing device 100 to adjust the expected value based on the similarity, and makes it easier to learn a model that accurately estimates the option to be selected by the selection subject.

[0288] According to the information processing device 100, a plurality of options each representing a means of transportation can be processed. This makes it possible for the information processing device 100 to easily and accurately estimate what means of transportation the selection subject will select.

[0289] The information processing method described in this embodiment can be realized by executing a prepared program on a computer such as a PC or a workstation. The information processing program described in this embodiment is recorded on a computer-readable recording medium, and is executed by being read from the recording medium by the computer. The recording medium may be a hard disk, a flexible disk, a CD (Compact Disc)-ROM, an MO (Magneto Optical disc), a DVD (Digital Versatile Disc), or the like. The information processing program described in this embodiment may also be distributed via a network such as the Internet.

[0290] The following supplementary notes are further disclosed regarding the above-described embodiment.

[0291] (Appendix 1) Obtain multiple records representing previously selected options from multiple options, For each of the one or more records corresponding to a selecting subject who has recognized a first option among the plurality of records obtained, a first value is set that represents a definite belongingness as an expected value of the probability that the selecting subject belongs to a first class that has recognized the first option; setting a second value representing a probability of the selecting subject belonging to the first class as the expected value for each remaining record different from the one or more records among the acquired multiple records based on a reference value of the probability that the selecting subject belongs to the first class and a formula representing an effect of each of the multiple options; learning the reference value and the parameters included in the formula based on each of the acquired records and the expected value set for each of the acquired records; An information processing method characterized in that the processing is executed by a computer.

[0292] (Additional Note 2) The learning process is and learning a model that estimates the probability that each of the multiple options will be selected using the formula in response to input of features related to each of the multiple options by learning the reference value and parameters of the formula.

[0293] (Additional Note 3) The learning process is The information processing method according to claim 2, characterized in that by learning the reference value and the parameters of the formula, a first model is learned that, when a selecting subject belongs to the first class, estimates the probability that each of the multiple options will be selected using the formula in response to input of features related to each of the multiple options, and a second model is learned that, when a selecting subject does not belong to the first class, estimates the probability that each of the remaining options will be selected using the formula in response to input of features related to each of the remaining options different from the first option.

[0294] (Additional Note 4) The process of setting the first value includes: 3. The information processing method according to claim 1 or 2, further comprising setting a first value representing a definite membership as the expected value for each record of one or more records representing the first option.

[0295] (Appendix 5) Each of the plurality of records includes an attribute value of a selection subject; The process of setting the first value includes: 3. The information processing method according to claim 1 or 2, further comprising setting, as the expected value, a first value representing definite belonging to one or more records having an attribute value identical or similar to a record representing the first option.

[0296] (Additional Note 6) The process of setting the second value includes: The information processing method described in Appendix 1 or 2, characterized in that a second value representing probabilistic belonging is set as the expected value for each record of the remaining records based on the reference value, a formula representing the effect of each of the multiple options when the selecting subject belongs to the first class, and a formula representing the effect of each of the remaining options different from the first option among the multiple options when the selecting subject does not belong to the first class.

[0297] (Supplementary Note 7) The process of setting the second value includes: The information processing method according to claim 1 or 2, further comprising: setting, as the expected value, a second value representing probabilistic belonging to each of the remaining records, based on the reference value, a formula representing the effect of each of the multiple options, and a similarity between each of the remaining records and a record representing the first option, such that the expected value becomes larger as the similarity between each of the remaining records and the record representing the first option increases.

[0298] (Supplementary Note 8) The information processing method according to any one of Supplementary Notes 1 to 7, wherein each of the plurality of options represents a means of transportation.

[0299] (Appendix 9) Obtain multiple records representing previously selected options from multiple options, For each of the one or more records corresponding to a selecting subject who has recognized a first option among the plurality of records obtained, a first value is set that represents a definite belongingness as an expected value of the probability that the selecting subject belongs to a first class that has recognized the first option; setting a second value representing a probability of the selecting subject belonging to the first class as the expected value for each remaining record different from the one or more records among the acquired multiple records based on a reference value of the probability that the selecting subject belongs to the first class and a formula representing an effect of each of the multiple options; learning the reference value and the parameters included in the formula based on each of the acquired records and the expected value set for each of the acquired records; An information processing program that causes a computer to execute a process. [Explanation of symbols]

[0300] 100 Information processing device 110 Choices 111 First Option 120 Records 130 First Class 501,502 class 200 Information Processing Systems 201 Client device 210 Network 300 Bus 301 CPU 302 Memory 303 Network Interface 304 Recording media I / F 305 Recording media 400 Storage section 401,601,1001,1301,1901 Acquisition part 402 First Setting Section 403 Second Setting Section 404 Learning Department 405 Prediction Department 406 Output section 511,512 Selected Models 602,1002,1302,1902 Recording Department 603,1303 Model setting section 604,1304 Recognition flag assignment section 605,1305 Learning data generation unit 606,1306 Learning processing unit 700,1600 Selection history data management table 800,1700 Graphical Models 900,1800 Learning data management table 1003,1003,1307,1903 Prediction data generation section 1004, 1308, 1904 Prediction processing section 1100,2000 Prediction target data management table 1200,2100 Prediction data management table

Claims

1. Obtaining a plurality of records each representing a previously selected option from among the plurality of options; For each of the one or more records corresponding to a selecting subject who has recognized a first option among the acquired multiple records, a first value is set that represents a definite belongingness as an expected value of the probability that the selecting subject belongs to a first class that has recognized the first option; setting a second value representing probabilistic belonging to the first class as the expected value for each remaining record different from the one or more records among the acquired multiple records based on a reference value of the probability that the selecting subject belongs to the first class and a formula representing an effect of each of the multiple options; learning the reference value and the parameters included in the formula based on each of the acquired records and the expected value set for each of the acquired records; An information processing method characterized in that the processing is executed by a computer.

2. The learning process includes:

2. The information processing method according to claim 1, further comprising: learning the reference value and parameters of the formula to learn a model that uses the formula to estimate the probability that each of the multiple options will be selected in response to input of features related to each of the multiple options.

3. The learning process includes:

3. The information processing method according to claim 2, characterized in that, by learning the reference value and parameters of the formula, a first model is learned that, when a selecting subject belongs to the first class, uses the formula to estimate the probability that each of the multiple options will be selected in response to input of features for each of the multiple options, and a second model is learned that, when a selecting subject does not belong to the first class, uses the formula to estimate the probability that each of the remaining options will be selected in response to input of features for each of the remaining options different from the first option.

4. The process of setting the first value includes:

3. The information processing method according to claim 1, further comprising setting, as the expected value, a first value representing a definite membership for each of the one or more records representing the first option.

5. each of the plurality of records includes an attribute value of a selection subject; The process of setting the first value includes: The information processing method according to claim 1 or 2, characterized in that for each of one or more records having an attribute value identical or similar to a record representing the first option, a first value representing definite belonging is set as the expected value.

6. The process of setting the second value includes: The information processing method according to claim 1 or 2, characterized in that a second value representing probabilistic belonging is set as the expected value for each record of the remaining records based on the reference value, a formula representing the effect of each of the multiple options when the selecting subject belongs to the first class, and a formula representing the effect of each of the remaining options different from the first option among the multiple options when the selecting subject does not belong to the first class.

7. The process of setting the second value includes: The information processing method according to claim 1 or 2, characterized in that a second value representing probabilistic belonging is set as the expected value for each record of the remaining records based on the reference value, a formula representing the effect of each of the multiple options, and a similarity between each record of the remaining records and a record representing the first option, so that the expected value becomes larger as the similarity between each record of the remaining records and the record representing the first option increases.

8. Obtaining a plurality of records each representing a previously selected option from among the plurality of options; For each of the one or more records corresponding to a selecting subject who has recognized a first option among the acquired multiple records, a first value is set that represents a definite belongingness as an expected value of the probability that the selecting subject belongs to a first class that has recognized the first option; setting a second value representing probabilistic belonging to the first class as the expected value for each remaining record different from the one or more records among the acquired multiple records based on a reference value of the probability that the selecting subject belongs to the first class and a formula representing an effect of each of the multiple options; learning the reference value and the parameters included in the formula based on each of the acquired records and the expected value set for each of the acquired records; An information processing program that causes a computer to execute a process.

Images